breakdown folder tree for epic R01

Reviewed-on: #5

application/__init__.py

@@ -1,12 +1 @@
-"""Application layer - pure Python use-case orchestration.
-
-Sits between ``presentation/`` (Qt widgets) and ``domain/`` (entities). A module
-here answers "what has to happen, in what order" for one use case - route a
-turn, run a conversation - without knowing whether a human, a scheduler or a
-test triggered it.
-
-Hard rule (ADR-001 I1/I3, enforced by ``scripts/check_imports.py``): no
-PySide6/PyQt imports and no reach into ``presentation/``/``ui/``. Results travel
-back up through plain-Python callbacks; turning those into Qt signals is the
-presentation layer's job.
-"""
+"""Application Layer: Pure Python use cases and application services."""

application/conversations/__init__.py

@@ -1,10 +1 @@
-"""Conversation use case: the lifecycle of one agent turn (EPIC R04) and the
-tool approval policy every turn's tool calls go through (EPIC R05)."""
-
-from .conversation_application_service import (
-    ConversationApplicationService,
-    TurnResult,
-)
-from .tool_policy_gateway import ConfirmGate, ToolPolicyGateway
-
-__all__ = ["ConversationApplicationService", "TurnResult", "ToolPolicyGateway", "ConfirmGate"]
+"""Application conversations package: turn lifecycle orchestration and agent execution."""

application/conversations/conversation_application_service.py

@@ -1,328 +0,0 @@
-"""ConversationApplicationService - the turn lifecycle, outside the widget (R04-T03).
-
-What this replaces
-------------------
-The lifecycle of one Cowork turn is currently spread across a closure inside
-``ui/cowork_tab.py::build_job`` and a second, near-identical assembly inside
-``core/task_executors.py::_run_agent``. Both:
-
-* read live UI/config state from a worker thread,
-* build the provider, the MCP tool set and the project context by hand,
-* call ``core.chat_agent.run_cowork`` with a dozen positional-ish arguments,
-* consume untyped event dicts.
-
-Two copies means a fix to one path (say, promoting output files on failure)
-silently misses the other. This service is the single implementation: it takes
-an immutable :class:`ConversationExecutionRequest`, runs the turn, and reports
-typed :class:`AgentEvent` objects.
-
-What it deliberately does NOT do
---------------------------------
-It does not re-implement the agent loop. ``run_cowork`` stays the engine
-(strangler fig, ADR-001 section 4) and keeps its characterization tests
-(``tests/characterization/test_run_cowork.py``). This layer owns the parts that
-were tangled into the UI: assembling the call, translating events, and giving a
-turn a well-defined end.
-
-Pure Python: no Qt import, no config access. Everything it needs arrives through
-constructor callbacks, so the same service runs a turn from a chat panel, from
-the scheduler, or from a test.
-"""
-from __future__ import annotations
-
-import logging
-from dataclasses import dataclass, field
-from pathlib import Path
-from typing import Any, Callable, Dict, List, Optional, Tuple
-
-from cowork_local.domain.agents.agent_event import (
-    AgentEvent,
-    ErrorEvent,
-    TurnCompletedEvent,
-    collect_text,
-    event_from_dict,
-)
-from cowork_local.domain.agents.conversation_execution_request import (
-    ConversationExecutionRequest,
-)
-
-logger = logging.getLogger("cowork_local.conversations")
-
-# Presentation/scheduler supplies these. Kept as plain callables (not objects)
-# so a test can wire the service with three lambdas.
-EventCallback = Callable[[AgentEvent], None]
-CancelFn = Callable[[], bool]
-ProviderFactory = Callable[[str, str], Any]          # (provider_id, model) -> Provider
-ToolSourceFactory = Callable[[], Tuple[Any, Any]]    # () -> (extra_tools, extra_executor)
-GateFactory = Callable[[ConversationExecutionRequest], Any]  # -> PermissionGate or None
-
-
-@dataclass
-class TurnResult:
-    """What a finished turn produced.
-
-    ``messages`` is the conversation AFTER the turn (system prompt inserted,
-    assistant and tool messages appended) - the caller persists this as the new
-    history. ``final_text`` is the visible answer, reasoning excluded.
-    """
-
-    request: ConversationExecutionRequest
-    messages: List[Dict[str, Any]] = field(default_factory=list)
-    events: List[AgentEvent] = field(default_factory=list)
-    final_text: str = ""
-    cancelled: bool = False
-    error: str = ""
-    # The original exception, kept alongside its message so a caller that needs
-    # to preserve legacy failure handling can re-raise the SAME object rather
-    # than a lookalike (SecurityBlocked, for instance, carries context that a
-    # re-wrapped RuntimeError would lose).
-    exception: Optional[BaseException] = None
-
-    @property
-    def ok(self) -> bool:
-        """True when the turn completed without an error and without a Stop."""
-        return not self.error and not self.cancelled
-
-    def raise_if_failed(self) -> None:
-        """Re-raise the turn's failure, if any.
-
-        Callers that already have failure handling built around an exception
-        (the Qt worker turns one into its ``failed`` signal) use this to keep
-        that path intact while still getting a TurnResult on success."""
-        if self.exception is not None:
-            raise self.exception
-
-    def output_dir(self) -> Optional[Path]:
-        """This turn's output folder, or None when it could not write files."""
-        return Path(self.request.output_dir) if self.request.output_dir else None
-
-
-class ConversationApplicationService:
-    """Runs one agent turn from an immutable request.
-
-    Args:
-        provider_factory: ``(provider_id, model) -> Provider``. Production passes
-            ``AppContext.build_provider_for``; tests pass a lambda returning a
-            :class:`FakeProvider`.
-        tool_source: ``() -> (extra_tools, extra_executor)`` for MCP/connector
-            tools. Optional - a turn with no external tools passes nothing.
-        gate_factory: ``(request) -> PermissionGate | None``, consulted when the
-            request asks to confirm commands. Optional for the same reason.
-        runner: the turn engine. Defaults to ``core.chat_agent.run_cowork``,
-            imported lazily so this module stays importable (and testable)
-            without pulling in the whole legacy tool stack.
-        security_config: the app config the security layers read. ``None``
-            disables them, which is what headless callers already rely on.
-    """
-
-    def __init__(
-        self,
-        provider_factory: ProviderFactory,
-        *,
-        tool_source: Optional[ToolSourceFactory] = None,
-        gate_factory: Optional[GateFactory] = None,
-        runner: Optional[Callable[..., Any]] = None,
-        security_config: Any = None,
-    ) -> None:
-        self._provider_factory = provider_factory
-        self._tool_source = tool_source
-        self._gate_factory = gate_factory
-        self._runner = runner
-        self._security_config = security_config
-
-    # -- main entry point -------------------------------------------------- #
-    def run_turn(
-        self,
-        request: ConversationExecutionRequest,
-        on_event: Optional[EventCallback] = None,
-        cancel: Optional[CancelFn] = None,
-    ) -> TurnResult:
-        """Execute one turn and return everything it produced.
-
-        Never raises: a provider or tool failure becomes an :class:`ErrorEvent`
-        plus ``TurnResult.error``. Callers run this on a worker thread and have
-        no good way to handle an exception crossing that boundary - today an
-        escaped error kills the worker and the UI just stops updating, with no
-        message shown.
-
-        Exactly one :class:`TurnCompletedEvent` is always emitted last, whether
-        the turn succeeded, failed or was cancelled. That is the end-of-turn
-        signal the legacy engine never had.
-        """
-        return self.execute_turn(self.begin_turn(request), on_event=on_event, cancel=cancel)
-
-    def begin_turn(self, request: ConversationExecutionRequest) -> TurnResult:
-        """Create the (still empty) result a turn will fill in.
-
-        Exposed separately from :meth:`run_turn` because some callers need the
-        LIVE message list while the turn is running, not only afterwards: the
-        scheduler re-saves the conversation to History after every assistant
-        message so a long unattended run shows live progress when reopened.
-        Handing them ``result.messages`` - the very list the engine appends to -
-        is what makes that possible without leaking the engine into the caller.
-        """
-        return TurnResult(request=request, messages=request.message_list())
-
-    def execute_turn(
-        self,
-        result: TurnResult,
-        on_event: Optional[EventCallback] = None,
-        cancel: Optional[CancelFn] = None,
-    ) -> TurnResult:
-        """Run a turn previously created by :meth:`begin_turn`. See
-        :meth:`run_turn` for the error/cancellation contract."""
-        request = result.request
-        emit = self._make_emitter(result, on_event)
-        cancel = cancel or (lambda: False)
-
-        try:
-            self._execute(request, result, emit, cancel)
-        except Exception as exc:  # noqa: BLE001 - see docstring
-            result.error = str(exc) or exc.__class__.__name__
-            result.exception = exc
-            logger.exception("turn %s failed", request.turn_id)
-            emit(ErrorEvent(message=result.error,
-                            recoverable=self._is_recoverable(exc)))
-
-        result.cancelled = bool(cancel())
-        result.final_text = collect_text(result.events) or self._last_assistant_text(result.messages)
-        emit(TurnCompletedEvent(content=result.final_text, cancelled=result.cancelled))
-        return result
-
-    # -- internals --------------------------------------------------------- #
-    def _execute(self, request: ConversationExecutionRequest, result: TurnResult,
-                 emit: Callable[[AgentEvent], None], cancel: CancelFn) -> None:
-        """Assemble the engine call from the request snapshot and run it."""
-        provider = self._provider_factory(request.provider, request.model)
-        extra_tools, extra_executor = self._resolve_tools()
-        gate = self._resolve_gate(request)
-
-        # The engine speaks untyped dicts; bridge them into typed events at this
-        # single point rather than at every consumer.
-        def legacy_emit(payload: Dict[str, Any]) -> None:
-            event = event_from_dict(payload)
-            if event is not None:
-                emit(event)
-
-        run = self._resolve_runner()
-        run(
-            provider,
-            result.messages,          # mutated in place by the engine, as before
-            self._output_dir(request),
-            legacy_emit,
-            cancel,
-            title=request.title,
-            extra_tools=extra_tools,
-            extra_executor=extra_executor,
-            project_context=request.project_context,
-            security_config=self._security_config,
-            gate=gate,
-            allowed_tools=list(request.allowed_tools) if request.allowed_tools is not None else None,
-            max_steps=request.max_steps,
-            run_to_completion=request.run_to_completion,
-            completion_max_steps=request.completion_max_steps,
-            enforce_rules=request.enforce_rules,
-            **self._role_kwargs(request),
-        )
-
-    @staticmethod
-    def _make_emitter(result: TurnResult,
-                      on_event: Optional[EventCallback]) -> Callable[[AgentEvent], None]:
-        """Record every event on the result AND forward it to the caller.
-
-        Recording is unconditional so a headless caller (the scheduler) can read
-        the full event list afterwards without having to supply a callback just
-        to collect it - which is exactly what task_executors does today with an
-        ad-hoc list.
-        """
-        def emit(event: AgentEvent) -> None:
-            result.events.append(event)
-            if on_event is None:
-                return
-            try:
-                on_event(event)
-            except Exception:  # noqa: BLE001
-                # A consumer that throws (a closing widget, say) must not abort
-                # the turn that is feeding it.
-                logger.debug("event consumer raised for %s", event.type, exc_info=True)
-        return emit
-
-    def _resolve_runner(self) -> Callable[..., Any]:
-        """The turn engine, imported lazily on first use."""
-        if self._runner is None:
-            from cowork_local.core.chat_agent import run_cowork
-
-            self._runner = run_cowork
-        return self._runner
-
-    def _resolve_tools(self) -> Tuple[Any, Any]:
-        """MCP/connector tools for this turn, or ``(None, None)``.
-
-        A failure here degrades to "no external tools" rather than failing the
-        turn: an MCP server that will not start must not stop the user from
-        chatting, which is the behaviour the chat panel already relies on.
-        """
-        if self._tool_source is None:
-            return None, None
-        try:
-            return self._tool_source()
-        except Exception:  # noqa: BLE001
-            logger.warning("tool source unavailable - running without external tools",
-                           exc_info=True)
-            return None, None
-
-    def _resolve_gate(self, request: ConversationExecutionRequest) -> Any:
-        """The permission gate, when this turn asked to confirm commands."""
-        if not request.confirm_commands or self._gate_factory is None:
-            return None
-        return self._gate_factory(request)
-
-    @staticmethod
-    def _output_dir(request: ConversationExecutionRequest) -> Path:
-        """The turn's output folder as a Path.
-
-        The request holds it as a string to stay serialisable; converting at the
-        single point of use keeps that decision from leaking into every caller.
-        """
-        return Path(request.output_dir) if request.output_dir else Path.cwd()
-
-    @staticmethod
-    def _role_kwargs(request: ConversationExecutionRequest) -> Dict[str, Any]:
-        """``agent_role`` only when the request set one.
-
-        Omitted otherwise so the engine applies its own default (the interactive
-        Cowork role) instead of being handed an empty string, which would land
-        in the audit log as an unattributed tool call.
-        """
-        return {"agent_role": request.agent_role} if request.agent_role else {}
-
-    @staticmethod
-    def _last_assistant_text(messages: List[Dict[str, Any]]) -> str:
-        """Fallback answer text when no text events were seen.
-
-        A turn whose whole answer arrived in one non-streamed message still has
-        to report a final answer - the scheduler writes it into output.md, and
-        an empty string there reads as "(no output)".
-        """
-        for message in reversed(messages):
-            if message.get("role") == "assistant" and (message.get("content") or "").strip():
-                return str(message["content"])
-        return ""
-
-    @staticmethod
-    def _is_recoverable(exc: Exception) -> bool:
-        """Whether the user can act on this failure themselves.
-
-        "Model not found" is the motivating case: the chat panel restores the
-        typed message into the composer so the user can switch model and resend
-        instead of retyping it (see providers/base.py::MODEL_NOT_FOUND_HINT).
-        """
-        try:
-            from cowork_local.providers.base import is_model_not_found_error
-
-            return bool(is_model_not_found_error(str(exc)))
-        except Exception:  # noqa: BLE001
-            return False
-
-
-__all__ = ["ConversationApplicationService", "TurnResult"]

application/conversations/tool_policy_gateway.py

@@ -1,83 +0,0 @@
-"""ToolPolicyGateway - one confirm/deny decision path for every tool call
-(R05-T03).
-
-Today "does this tool call need the user's OK first" is answered by a
-different hand-written check per engine:
-
-* ``core/chat_agent.py::run_cowork`` — ``name in ("run_command",
-  "install_package")``, a literal tuple.
-* ``core/code_agent.py::run_code`` — ``name in (WRITE_TOOLS | MS365_WRITE_TOOLS)``,
-  a set built from two other hand-maintained sets.
-* MCP/connector tools (``core/mcp_client.py``, ``core/ext_connectors.py``) —
-  no check at all; ``chat_agent.py`` calls ``extra_executor(name, args)``
-  directly.
-
-Three answers to the same question, and the third one is a real gap: an MCP
-tool that deletes files or calls an external API today runs with zero
-confirmation even when the user turned "confirm before running commands" on.
-
-This gateway answers the question from data (:class:`~domain.tools.tool_descriptor.ToolCapability`
-via a :class:`~domain.tools.tool_registry.ToolRegistry`) instead of a literal
-name list, so registering a tool with the right capability is what gates it -
-nothing to remember at each new call site. R05-T04 is what actually registers
-MCP/connector tools with a capability; this module only needs the mechanism
-to exist.
-
-Pure Python: no Qt, no direct dialog. The actual approval prompt stays exactly
-what it is today - a ``gate`` object with a ``.request(payload) -> bool``
-method, supplied by the presentation layer (Settings' "confirm before running
-commands" wires it up, or None for auto-run) - this module only decides
-WHEN to ask it, never how to render the question.
-"""
-from __future__ import annotations
-
-from typing import Any, Dict, Optional, Protocol
-
-from cowork_local.domain.tools import ToolCapability, ToolRegistry
-
-
-class ConfirmGate(Protocol):
-    """Shape of the existing ``PermissionGate`` both engines already use."""
-
-    def request(self, payload: Dict[str, Any]) -> bool: ...
-
-
-class ToolPolicyGateway:
-    """Decides whether a tool call needs approval, for ONE calling surface.
-
-    ``gated_capabilities`` is what makes this per-surface: Cowork only ever
-    asked about ``run_command``/``install_package`` (capability ``EXECUTE``),
-    while the Code tab additionally confirms plain file writes (capability
-    ``WRITE``). Passing the wrong set here would silently change which tools
-    prompt for approval - see the callers in ``core/chat_agent.py`` and
-    ``core/code_agent.py`` for the exact sets that preserve today's behavior.
-    """
-
-    def __init__(self, registry: ToolRegistry, gated_capabilities: ToolCapability) -> None:
-        self._registry = registry
-        self._gated_capabilities = gated_capabilities
-
-    def requires_confirmation(self, name: str) -> bool:
-        """True when ``name``'s declared capabilities overlap this surface's
-        gated set. An unregistered tool never requires confirmation through
-        this path - callers that must fail safe on unknown tools check
-        ``name in registry`` themselves (see R05-T04's MCP wrapping, which
-        registers every tool it exposes before any call can reach here)."""
-        return bool(self._registry.capabilities_for(name) & self._gated_capabilities)
-
-    def allow(self, name: str, gate: Optional[ConfirmGate], payload: Dict[str, Any]) -> bool:
-        """True when the call may proceed.
-
-        ``gate is None`` preserves each engine's existing "no gate wired -
-        auto-run" behavior; a tool outside ``gated_capabilities`` is never
-        asked about, matching read-only tools "never confirm" today.
-        ``payload`` is whatever ``gate.request(...)`` already expects at that
-        call site (the two engines use slightly different dict shapes) - this
-        gateway only decides WHETHER to call it, never reshapes the payload.
-        """
-        if gate is None or not self.requires_confirmation(name):
-            return True
-        return bool(gate.request(payload))
-
-
-__all__ = ["ToolPolicyGateway", "ConfirmGate"]

application/model_routing/__init__.py

@@ -1,12 +1 @@
-"""Model routing use case: pick the best-fit model for one turn (EPIC R03)."""
-
-from .routing_application_service import (
-    RoutingApplicationService,
-    RoutingDecision,
-    RoutingMode,
-    is_valid_mode,
-    normalize_mode,
-)
-
-__all__ = ["RoutingApplicationService", "RoutingDecision", "RoutingMode",
-           "normalize_mode", "is_valid_mode"]
+"""Application model routing package: model route decisions and multi-provider balancing."""

application/model_routing/routing_application_service.py

@@ -1,353 +0,0 @@
-"""RoutingApplicationService - one routing flow for every surface (R03-T03).
-
-Before this service, the same routing algorithm existed three times:
-
-* ``ui/chat_panel.py::_apply_routing``       (Cowork chat)
-* ``ui/co4e_tab.py::_apply_co4e_routing``    (Co4E studio)
-* ``ui/folder_tab.py::_ai_apply_routing``    (AI-Edit)
-
-The three copies had already drifted - each one resolves the "current model"
-differently and each one has its own private notion of what to do when the user
-declines - and every one of them lives inside a Qt widget, so none of the logic
-could be tested without building a window.
-
-This module is the single implementation. It is pure Python: no Qt import, no
-config access, no network. The presentation layer supplies a confirm callback
-and renders the notice; everything else happens here.
-
-Modes (:class:`RoutingMode`)
-----------------------------
-* ``OFF``      - never switch. The user's pinned model always wins.
-* ``AUTO``     - switch silently when the best candidate clears the gain threshold.
-* ``MANUAL``   - propose the switch and switch only if the confirm callback approves.
-* ``FALLBACK`` - never switch pre-emptively; switch only AFTER the current model
-  fails, to the next-best candidate. This is the mode a user wants when they
-  trust their own model choice but still want the turn to survive an outage.
-
-Migration note (ADR-001 section 4): the scoring/ranking engine is NOT rewritten.
-This service depends on the small :class:`RoutingPort` interface, and production
-wires the existing, already-tested ``core.routing.service.RoutingService`` into
-it. Tests wire a fake.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass
-from enum import Enum
-from typing import Any, Callable, List, Optional, Protocol, Sequence, Tuple
-
-
-class RoutingMode(str, Enum):
-    """Per-surface routing behaviour.
-
-    The first three values match ``core.routing.models.SwitchMode`` string for
-    string, so a mode read from the existing config round-trips unchanged.
-    """
-
-    OFF = "off"
-    AUTO = "auto"
-    MANUAL = "manual"
-    FALLBACK = "fallback"
-
-    @classmethod
-    def parse(cls, raw: Any) -> "RoutingMode":
-        """Best-effort parse of a config value.
-
-        Unknown or empty values become ``OFF``: routing is an optimisation, and
-        the safe reading of a corrupt setting is "leave the user's model alone"
-        rather than "silently move their work to another model".
-        """
-        try:
-            return cls(str(raw or "off").strip().lower())
-        except ValueError:
-            return cls.OFF
-
-
-@dataclass(frozen=True)
-class RoutingDecision:
-    """The outcome of routing one turn - an immutable instruction for the caller.
-
-    ``provider``/``model`` are ALWAYS filled with what the turn should actually
-    run on, switched or not, so a call site never has to re-derive the fallback
-    itself (the bug that made the three UI copies diverge).
-    """
-
-    mode: RoutingMode
-    provider: str
-    model: str
-    switched: bool = False
-    task_type: str = ""
-    score_gain: float = 0.0
-    reason: str = ""
-    declined: bool = False  # Manual mode: a switch was offered and refused
-    # What the turn would have run on without routing. Carried so the Manual
-    # confirm dialog can show "from X to Y" without re-deriving the current
-    # model itself - re-deriving it differently per screen is exactly how the
-    # three legacy copies drifted apart.
-    previous_provider: str = ""
-    previous_model: str = ""
-
-    @property
-    def should_notify(self) -> bool:
-        """True when the UI should show the "switched model" notice - i.e. only
-        when a switch really happened."""
-        return self.switched
-
-    def target(self) -> Tuple[str, str]:
-        """``(provider, model)`` to run this turn on."""
-        return self.provider, self.model
-
-    @property
-    def from_model(self) -> str:
-        """Candidate key (``provider/model``) of the model being switched away
-        from, or "" when nothing was selected yet.
-
-        Named to match ``core.routing.models.SwitchDecision`` so the existing
-        Manual-mode dialog (``ui/routing_toggle.py::confirm_switch``) accepts
-        this object unchanged - the dialog moves to the new shape in EPIC R08.
-        """
-        if not self.previous_model:
-            return ""
-        return f"{self.previous_provider}/{self.previous_model}"
-
-    @property
-    def to_model(self) -> str:
-        """Candidate key (``provider/model``) of the model to run on. See
-        :attr:`from_model` for why the name matches the legacy decision."""
-        return f"{self.provider}/{self.model}" if self.model else ""
-
-
-def is_valid_mode(raw: Any) -> bool:
-    """True when ``raw`` names a mode the routing service understands.
-
-    Distinct from :func:`normalize_mode` because callers need to tell "the user
-    chose off" apart from "this stored value is unrecognised" - the per-workspace
-    lookup falls back to the global setting only in the second case.
-    """
-    try:
-        RoutingMode(str(raw or "").strip().lower())
-    except ValueError:
-        return False
-    return True
-
-
-def normalize_mode(raw: Any) -> str:
-    """Canonical mode string for persistence, or ``"off"`` when unrecognised.
-
-    Exists so the mode vocabulary is defined exactly once. It used to be
-    hard-coded as a ``("off", "auto", "manual")`` tuple in four separate places
-    (config.py twice, state.py twice); adding FALLBACK meant finding all four,
-    and missing one silently downgraded the user's choice back to "off".
-    """
-    return RoutingMode.parse(raw).value
-
-
-class RoutingPort(Protocol):
-    """The slice of the routing engine this service needs.
-
-    Declared as a Protocol so the application layer states its requirement
-    without importing the implementation - which is what lets the whole service
-    be tested against a 20-line fake, and lets ``core.routing`` be replaced later
-    without touching this file.
-    """
-
-    def route(self, surface: str, prompt: str, current_provider: str, current_model: str,
-              *, mode_override: Optional[str] = None,
-              required_capabilities: Optional[List[str]] = None,
-              task_type: Optional[Any] = None) -> Any:
-        """Return a route result exposing ``should_switch``, ``target()``,
-        ``task_type`` and ``decision``."""
-
-
-# Presentation supplies this to ask the human. Receives the proposal so the
-# dialog can explain it; returns True to approve. Manual mode only.
-ConfirmFn = Callable[[RoutingDecision], bool]
-
-
-class RoutingApplicationService:
-    """Decides which provider/model one turn runs on.
-
-    Args:
-        router: the scoring engine (see :class:`RoutingPort`).
-        mode_reader: ``surface -> mode string``; production passes the per-workspace
-            lookup ``AppContext.project_routing_mode``. Injected rather than read
-            from config here so this layer stays free of config plumbing that
-            EPIC R02 is rewriting in parallel.
-    """
-
-    def __init__(self, router: RoutingPort,
-                 mode_reader: Optional[Callable[[str], str]] = None) -> None:
-        self._router = router
-        self._mode_reader = mode_reader
-
-    # -- main entry point -------------------------------------------------- #
-    def route_turn(
-        self,
-        surface: str,
-        prompt: str,
-        current_provider: str,
-        current_model: str,
-        *,
-        mode: Optional[str] = None,
-        confirm: Optional[ConfirmFn] = None,
-        required_capabilities: Optional[Sequence[str]] = None,
-        task_type: Optional[Any] = None,
-    ) -> RoutingDecision:
-        """Decide what to run this turn on. Never raises.
-
-        A routing failure must never block a message: any unexpected error
-        degrades to "keep the current model", which is exactly what all three
-        legacy copies did with a bare ``except`` - made explicit and testable here.
-        """
-        resolved_mode = RoutingMode.parse(mode if mode is not None else self._read_mode(surface))
-        keep = self._keep(resolved_mode, current_provider, current_model,
-                          reason="routing off - keeping current model")
-
-        # An empty prompt carries no signal to classify, so routing cannot make a
-        # meaningful choice; the same guard exists in all three legacy copies.
-        if resolved_mode is RoutingMode.OFF or not (prompt or "").strip():
-            return keep
-
-        # FALLBACK never switches up front - it only reacts to a failure, which
-        # the caller reports through fallback_after_failure().
-        if resolved_mode is RoutingMode.FALLBACK:
-            return self._keep(resolved_mode, current_provider, current_model,
-                              reason="fallback mode - switching only after a failure")
-
-        try:
-            result = self._router.route(
-                surface, prompt, current_provider, current_model,
-                mode_override=resolved_mode.value,
-                required_capabilities=list(required_capabilities) if required_capabilities else None,
-                task_type=task_type,
-            )
-        except Exception:  # noqa: BLE001 - routing must never break a turn
-            return self._keep(resolved_mode, current_provider, current_model,
-                              reason="routing engine failed - keeping current model")
-
-        proposal = self._to_decision(result, resolved_mode, current_provider, current_model)
-        if not proposal.switched:
-            return proposal
-
-        # Manual mode: the proposal only becomes a switch once a human approves.
-        if resolved_mode is RoutingMode.MANUAL:
-            if confirm is None or not self._ask(confirm, proposal):
-                return self._keep(resolved_mode, current_provider, current_model,
-                                  reason="switch declined - keeping current model",
-                                  task_type=proposal.task_type, declined=True)
-        return proposal
-
-    # -- failure recovery -------------------------------------------------- #
-    def fallback_after_failure(
-        self,
-        surface: str,
-        prompt: str,
-        failed_provider: str,
-        failed_model: str,
-        *,
-        mode: Optional[str] = None,
-        required_capabilities: Optional[Sequence[str]] = None,
-        task_type: Optional[Any] = None,
-    ) -> Optional[RoutingDecision]:
-        """Pick a replacement after ``failed_provider/failed_model`` failed.
-
-        Returns None when there is nothing to fall back to, so the caller can
-        surface the original error instead of retrying forever. Available in
-        AUTO and FALLBACK; OFF and MANUAL keep the user's model on failure too,
-        because silently moving work to another model is exactly what those two
-        modes exist to prevent.
-        """
-        resolved_mode = RoutingMode.parse(mode if mode is not None else self._read_mode(surface))
-        if resolved_mode not in (RoutingMode.AUTO, RoutingMode.FALLBACK):
-            return None
-
-        try:
-            # Asked in AUTO so the engine ranks candidates rather than short-
-            # circuiting on FALLBACK's "never switch up front" rule; the failed
-            # model is passed as current so any positive gain beats it.
-            result = self._router.route(
-                surface, prompt, failed_provider, failed_model,
-                mode_override=RoutingMode.AUTO.value,
-                required_capabilities=list(required_capabilities) if required_capabilities else None,
-                task_type=task_type,
-            )
-        except Exception:  # noqa: BLE001 - a broken router must not mask the real error
-            return None
-
-        decision = self._to_decision(result, resolved_mode, failed_provider, failed_model)
-        # A "switch" back to the model that just failed would retry the outage.
-        if not decision.switched or (decision.provider, decision.model) == (failed_provider, failed_model):
-            return None
-        return RoutingDecision(
-            mode=resolved_mode, provider=decision.provider, model=decision.model,
-            switched=True, task_type=decision.task_type, score_gain=decision.score_gain,
-            reason=f"{failed_provider}/{failed_model} failed - falling back to "
-                   f"{decision.provider}/{decision.model}",
-            previous_provider=failed_provider, previous_model=failed_model,
-        )
-
-    # -- internals --------------------------------------------------------- #
-    def _read_mode(self, surface: str) -> str:
-        """Per-surface mode from the injected reader ('off' when none supplied)."""
-        if self._mode_reader is None:
-            return RoutingMode.OFF.value
-        try:
-            return self._mode_reader(surface) or RoutingMode.OFF.value
-        except Exception:  # noqa: BLE001 - a config read must not break a turn
-            return RoutingMode.OFF.value
-
-    @staticmethod
-    def _keep(mode: RoutingMode, provider: str, model: str, *, reason: str,
-              task_type: str = "", declined: bool = False) -> RoutingDecision:
-        """A no-switch decision that still names the model to run on."""
-        return RoutingDecision(mode=mode, provider=provider, model=model, switched=False,
-                               task_type=task_type, reason=reason, declined=declined,
-                               previous_provider=provider, previous_model=model)
-
-    @staticmethod
-    def _ask(confirm: ConfirmFn, proposal: RoutingDecision) -> bool:
-        """Run the confirm callback, treating any failure as "declined".
-
-        The callback opens a modal dialog in production; if that raises (window
-        already closing, for instance) the safe answer is to keep the user's own
-        model rather than to switch without consent.
-        """
-        try:
-            return bool(confirm(proposal))
-        except Exception:  # noqa: BLE001
-            return False
-
-    @staticmethod
-    def _to_decision(result: Any, mode: RoutingMode,
-                     current_provider: str, current_model: str) -> RoutingDecision:
-        """Translate the engine's route result into a :class:`RoutingDecision`.
-
-        Defensive about the result shape on purpose: this is the seam between the
-        new layer and a legacy module still under refactor, and a missing
-        attribute must degrade to "keep current model" instead of raising into
-        the middle of a chat turn.
-        """
-        inner = getattr(result, "decision", None)
-        task_type = getattr(getattr(result, "task_type", None), "value", "") or ""
-        gain = float(getattr(inner, "score_gain", 0.0) or 0.0)
-        reason = str(getattr(inner, "reason", "") or "")
-
-        target = None
-        if getattr(result, "should_switch", False):
-            getter = getattr(result, "target", None)
-            target = getter() if callable(getter) else None
-
-        if not target:
-            return RoutingDecision(mode=mode, provider=current_provider, model=current_model,
-                                   switched=False, task_type=task_type, score_gain=gain,
-                                   reason=reason or "no better model - keeping current",
-                                   previous_provider=current_provider,
-                                   previous_model=current_model)
-
-        provider, model = target
-        return RoutingDecision(mode=mode, provider=provider or current_provider, model=model,
-                               switched=True, task_type=task_type, score_gain=gain, reason=reason,
-                               previous_provider=current_provider, previous_model=current_model)
-
-
-__all__ = ["RoutingApplicationService", "RoutingDecision", "RoutingMode",
-           "RoutingPort", "normalize_mode", "is_valid_mode"]

application/monitoring/__init__.py

@@ -0,0 +1 @@
+"""Application monitoring package: Monitoring query service for audit and metrics."""

application/scheduling/__init__.py

@@ -0,0 +1 @@
+"""Application scheduling package: TaskApplicationService and AI task planning."""

application/settings/__init__.py

@@ -0,0 +1 @@
+"""Application settings package: Settings application service."""

application/workflows/__init__.py

@@ -0,0 +1 @@
+"""Application workflows package: Co4E graph execution orchestration."""

application/workspaces/__init__.py

@@ -1,5 +1 @@
-"""Workspace file operations for non-agent-loop callers (EPIC R06)."""
-
-from .file_workspace_service import FileWorkspaceService
-
-__all__ = ["FileWorkspaceService"]
+"""Application workspaces package: File workspace and AI file editor services."""

application/workspaces/file_workspace_service.py

@@ -1,81 +0,0 @@
-"""FileWorkspaceService - the safe file operations File Explorer and the AI
-File Editor need, outside the agent tool loop (R06-T05).
-
-``ui/folder_tab.py`` (File Explorer) and the AI File Editor dialog need the
-exact same guarantees the agent's tools already have — path containment
-inside the workspace, precise context-anchored edits, syntax warnings on a
-bad Python write — but today that logic only exists wired to a model's tool
-call (``core/tools.py::execute_tool``). A UI action that isn't a tool call
-(browsing the tree, applying an AI-suggested diff from a review dialog) has
-no equivalent entry point of its own.
-
-This service IS that entry point. It reuses ``core/tools.py::execute_tool``
-verbatim - same dispatch table, same ``ToolContext`` containment check, same
-audit-log entry, same Python-syntax warning on write/edit - rather than
-re-implementing any of it, so a fix to one path fixes both. It only adds the
-:class:`~domain.workspaces.workspace_session.WorkspaceSession` seam: which
-workspace root a call is scoped to is decided by the session, not by
-whichever folder a widget happens to have open.
-"""
-from __future__ import annotations
-
-from typing import Any, Dict
-
-
-class FileWorkspaceService:
-    """File operations scoped to one :class:`WorkspaceSession`.
-
-    Read-only by name (``list_tree``/``read_preview``) vs. writing
-    (``write_file``/``apply_edit``) mirrors the same READ/WRITE split
-    ``domain/tools/tool_registry.py`` uses for the agent's own tools - a
-    caller that only wants to browse never accidentally has write access.
-    """
-
-    def __init__(self, session) -> None:  # WorkspaceSession - see module docstring
-        self._session = session
-
-    def list_tree(self, rel: str = ".") -> Dict[str, Any]:
-        """Entries at ``rel`` (default: the workspace root)."""
-        return self._execute("list_dir", {"path": rel})
-
-    def read_preview(self, rel: str) -> Dict[str, Any]:
-        """A text file's content (truncated by
-        ``infrastructure/filesystem/file_tools.py::MAX_READ_BYTES``, same as
-        the agent's ``read_file`` tool)."""
-        return self._execute("read_file", {"path": rel})
-
-    def write_file(self, rel: str, content: str) -> Dict[str, Any]:
-        """Create or fully overwrite ``rel``."""
-        return self._execute("write_file", {"path": rel, "content": content})
-
-    def apply_edit(self, rel: str, old_string: str, new_string: str,
-                   replace_all: bool = False) -> Dict[str, Any]:
-        """Replace an exact snippet in an existing file - the same
-        context-anchored algorithm the agent's ``edit_file`` tool uses, so an
-        AI-suggested diff applies with the same precision and the same
-        "old_string not found / ambiguous" failure messages either path
-        would give the caller."""
-        return self._execute("edit_file", {
-            "path": rel, "old_string": old_string, "new_string": new_string,
-            "replace_all": replace_all,
-        })
-
-    # -- internals --------------------------------------------------------- #
-    def _tool_context(self):
-        """A ``ToolContext`` scoped to this session's workspace root.
-        ``flatten_writes=False`` (unlike Cowork's agent context) - File
-        Explorer must preserve whatever subfolder structure the user is
-        actually browsing, not collapse every write into the root."""
-        from cowork_local.infrastructure.filesystem.tool_context import ToolContext
-
-        return ToolContext(self._session.workspace_root, flatten_writes=False)
-
-    def _execute(self, name: str, args: Dict[str, Any]) -> Dict[str, Any]:
-        """Dispatch through ``core/tools.py::execute_tool`` - see the module
-        docstring for why this delegates instead of reimplementing."""
-        from cowork_local.core.tools import execute_tool
-
-        return execute_tool(self._tool_context(), name, args)
-
-
-__all__ = ["FileWorkspaceService"]

config.py

@@ -105,7 +105,7 @@ DEFAULT_CONFIG: Dict[str, Any] = {
         # sandboxes agent-run shell commands) — reading a URL for info is safe and
         # useful, so this defaults ON. Toggle in Settings → Security.
         "allow_url_fetch": True,
-        "sandbox_pw": "quandh14",          # default password to unlock sandbox settings
+        "sandbox_pw": "",                  # set through COWORK_SANDBOX_PASSWORD
         "rulebase_path": "",               # custom RULEBASE.md — attached to every agent execution
     },
     # Legacy generic-MCP-server list. MERGED into ext_connectors["other"] as of
@@ -173,7 +173,7 @@ DEFAULT_CONFIG: Dict[str, Any] = {
     # Microsoft. Real Outlook/Teams/OneDrive/SharePoint access still requires a
     # proper OAuth sign-in (not implemented yet) using tenant_id/client_id below.
     "ms365": {
-        "unlock_code": "quandh14",
+        "unlock_code": "",                # set through COWORK_MS365_UNLOCK_CODE
         "unlocked": False,           # runtime-only — never persisted as True, see save()
         # Auto-connect MS365/OneDrive/SharePoint: the built-in MS365 MCP server
         # launches automatically once the user is signed in (OAuth tenant/client
@@ -294,6 +294,10 @@ def _apply_env_overrides(data: Dict[str, Any]) -> Dict[str, Any]:
         data["active_provider"] = os.environ["COWORK_ACTIVE_PROVIDER"]
     if os.getenv("COWORK_CA_BUNDLE"):
         data["tls_ca_bundle"] = os.environ["COWORK_CA_BUNDLE"]
+    if os.getenv("COWORK_SANDBOX_PASSWORD"):
+        data["agent_security"]["sandbox_pw"] = os.environ["COWORK_SANDBOX_PASSWORD"]
+    if os.getenv("COWORK_MS365_UNLOCK_CODE"):
+        data["ms365"]["unlock_code"] = os.environ["COWORK_MS365_UNLOCK_CODE"]
     return data
 
 
@@ -552,25 +556,19 @@ class AppConfig:
         return d
 
     def routing_mode_for(self, surface: str) -> str:
-        """Effective Off/Auto/Manual/Fallback mode for a chat surface.
-
-        A per-surface override wins; an empty override falls back to the global
-        ``switch_mode``. The value is validated through
-        ``application.model_routing.normalize_mode`` so the accepted vocabulary
-        is defined in exactly one place (R03-T03) - it used to be a literal
-        tuple repeated here and in state.py, and adding a mode to one copy but
-        not the others silently downgraded the user's choice to "off"."""
-        from .application.model_routing import normalize_mode
+        """Effective Off/Auto/Manual mode for a chat surface.
 
+        A per-surface override ("auto"/"manual"/"off") wins; an empty override
+        falls back to the global ``switch_mode``."""
         routing = self.routing
         override = (routing.get("surface_modes", {}) or {}).get(surface, "")
-        return normalize_mode(override or routing.get("switch_mode", "off"))
+        mode = override or routing.get("switch_mode", "off")
+        return mode if mode in ("off", "auto", "manual") else "off"
 
     def set_routing_mode_for(self, surface: str, mode: str) -> None:
-        """Persist a chat surface's routing toggle selection."""
-        from .application.model_routing import normalize_mode
-
-        self.routing.setdefault("surface_modes", {})[surface] = normalize_mode(mode)
+        """Persist a chat surface's Off/Auto/Manual toggle selection."""
+        mode = mode if mode in ("off", "auto", "manual") else "off"
+        self.routing.setdefault("surface_modes", {})[surface] = mode
         self.save()
 
     @property

core/chat_agent.py

@@ -11,8 +11,6 @@ import re
 from pathlib import Path
 from typing import Any, Callable, Dict, List, Optional
 
-from ..application.conversations.tool_policy_gateway import ToolPolicyGateway
-from ..domain.tools import ToolCapability, default_registry
 from ..providers.base import Provider, ToolSpec
 from . import agent_roles
 from . import agent_security
@@ -29,13 +27,6 @@ from .tools import TOOL_SPECS, ToolContext, _snapshot, describe_action, execute_
 # Generator / helper scripts — never a final deliverable in Cowork's output.
 _SCRIPT_EXTS = {".py", ".pyw", ".js", ".mjs", ".cjs", ".ts", ".sh", ".bat", ".ps1", ".rb", ".pl"}
 
-# R05-T03/T04: replaces the literal ``name in ("run_command",
-# "install_package")`` check below with a capability lookup — EXECUTE is
-# exactly the capability those two (and only those two) built-in tools carry
-# (see domain/tools/tool_registry.py::BUILT_IN_CAPABILITIES). Copied per-turn
-# into ``turn_tool_policy`` inside run_cowork() once extra_tools are known.
-_COWORK_TOOL_REGISTRY = default_registry(TOOL_SPECS)
-
 EmitFn = Callable[[Dict[str, Any]], None]
 CancelFn = Callable[[], bool]
 
@@ -397,19 +388,6 @@ def run_cowork(
                       jira=(security_config.data.get("jira") if security_config else None))
     extra_tools = extra_tools or []
     extra_names = {t.name for t in extra_tools}
-    # R05-T04: MCP servers (core/mcp_client.py) and unified connectors
-    # (core/ext_connectors.py) — everything that arrives here as extra_tools —
-    # advertise no standard risk metadata, so each is tagged with the same
-    # conservative default (WRITE|EXECUTE|NETWORK) domain/tools/tool_registry.py
-    # uses for any unclassified tool. Copying the built-in registry per turn
-    # (cheap - under 20 entries) rather than mutating the shared module-level
-    # one keeps different turns' extra_tools from leaking into each other.
-    from ..domain.tools import ToolDescriptor, ToolRegistry
-    from ..domain.tools.tool_registry import UNKNOWN_SOURCE_CAPABILITIES
-    _turn_registry = ToolRegistry(_COWORK_TOOL_REGISTRY.all())
-    for _spec in extra_tools:
-        _turn_registry.register(ToolDescriptor.from_spec(_spec, UNKNOWN_SOURCE_CAPABILITIES))
-    turn_tool_policy = ToolPolicyGateway(_turn_registry, ToolCapability.EXECUTE)
     # update_plan drives the Plan panel (above Output); it produces no file.
     # Built-in tools the admin disabled (Monitoring → Tools) are filtered out.
     from .tools import enabled_tool_specs
@@ -511,18 +489,6 @@ def run_cowork(
                     preview = {"kind": "info", "title": name, "text": str(args)}
                     emit({"type": "tool_proposed", "id": tc_id, "name": name, "args": args,
                           "preview": preview})
-                    # R05-T04: MCP/connector tools used to run with NO permission
-                    # check at all — this is what closes that gap. Same policy,
-                    # same gate object as the built-in tools below.
-                    if not turn_tool_policy.allow(
-                        name, gate, {"name": name, "args": args, "preview": preview}
-                    ):
-                        result = {"ok": False, "output": "Rejected by user."}
-                        emit({"type": "tool_result", "id": tc_id, "name": name,
-                              "ok": False, "output": result["output"]})
-                        messages.append({"role": "tool", "tool_call_id": tc_id, "name": name,
-                                         "content": result["output"]})
-                        continue
                     result = extra_executor(name, args)
                     emit({"type": "tool_result", "id": tc_id, "name": name,
                           "ok": result.get("ok", False), "output": result.get("output", "")})
@@ -562,18 +528,16 @@ def run_cowork(
                 # Permission Management (Sandbox Security Layer) — only when a
                 # gate was actually supplied (Settings: "confirm before running
                 # commands"); None preserves the pre-existing auto-run behavior.
-                # R05-T03: gating is now capability-driven (see
-                # turn_tool_policy above) instead of a literal name tuple.
-                if not turn_tool_policy.allow(
-                    name, gate, {"name": name, "args": args, "preview": preview}
-                ):
-                    result = {"ok": False, "output": "Rejected by user."}
-                    evt = {"type": "tool_result", "id": tc_id, "name": name,
-                          "ok": False, "output": result["output"]}
-                    emit(evt)
-                    messages.append({"role": "tool", "tool_call_id": tc_id,
-                                    "name": name, "content": result["output"]})
-                    continue
+                if gate is not None and name in ("run_command", "install_package"):
+                    approved = gate.request({"name": name, "args": args, "preview": preview})
+                    if not approved:
+                        result = {"ok": False, "output": "Rejected by user."}
+                        evt = {"type": "tool_result", "id": tc_id, "name": name,
+                              "ok": False, "output": result["output"]}
+                        emit(evt)
+                        messages.append({"role": "tool", "tool_call_id": tc_id,
+                                        "name": name, "content": result["output"]})
+                        continue
 
                 if name == "save_file":
                     result = _do_save_file(output_dir, title, args)

core/code_agent.py

@@ -12,8 +12,6 @@ import re
 from pathlib import Path
 from typing import Any, Callable, Dict, List, Optional
 
-from ..application.conversations.tool_policy_gateway import ToolPolicyGateway
-from ..domain.tools import ToolCapability, ToolDescriptor, ToolRegistry
 from ..providers.base import Provider
 from . import agent_roles
 from . import agent_security
@@ -227,14 +225,6 @@ def run_code(
     # read/list ms365 tools count as "read-only, never confirm". Names are
     # the MCP-qualified "ms365__*" form the agent sees (see ms365_tools.py).
     gated_tools = WRITE_TOOLS | MS365_WRITE_TOOLS
-    # R05-T03/T04: ``gated_tools`` stays the authoritative name set (unchanged),
-    # but the actual confirm decision now goes through the same
-    # ToolPolicyGateway class run_cowork uses, instead of a separate
-    # hand-rolled ``if name in gated_tools`` + direct ``gate.request(...)``.
-    code_tool_policy = ToolPolicyGateway(
-        ToolRegistry(ToolDescriptor(n, "", {}, ToolCapability.WRITE) for n in gated_tools),
-        ToolCapability.WRITE,
-    )
     # In PLAN mode, don't advertise write/run tools (analysis only).
     advertised = [t for t in all_tools if t.name not in gated_tools] if plan else all_tools
     has_memory = any(t.name.startswith("cmem_") for t in extra_tools)
@@ -307,11 +297,10 @@ def run_code(
             agent_security.enforce_command(provider, name, args, security_config, emit,
                                            agent_kind="code")
 
-            # read-only tools (incl. codebase memory) never consult the gate —
-            # code_tool_policy.requires_confirmation(name) is False for them.
-            approved = code_tool_policy.allow(
-                name, gate, {"id": tc_id, "name": name, "args": args, "preview": preview}
-            )
+            if name in gated_tools:
+                approved = gate.request({"id": tc_id, "name": name, "args": args, "preview": preview})
+            else:
+                approved = True  # read-only tools (incl. codebase memory) never confirm
 
             if cancel():
                 return messages

core/history.py

@@ -66,9 +66,7 @@ def save_conversation(
         "outputs": list(outputs or []),
         "messages": messages,
     }
-    # R06-T02: atomic write - see infrastructure/persistence/json/atomic_write.py.
-    from ..infrastructure.persistence.json.atomic_write import write_json
-    write_json(path, payload)
+    path.write_text(json.dumps(payload, ensure_ascii=False, indent=2), encoding="utf-8")
     return path
 
 
@@ -80,19 +78,15 @@ def delete_conversation(path) -> None:
 
 
 def rename_conversation(path, new_title: str) -> None:
-    from ..infrastructure.persistence.json.atomic_write import write_json
-
     data = load_conversation(path)
     data["title"] = new_title
-    write_json(Path(path), data)
+    Path(path).write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
 
 
 def set_pinned(path, pinned: bool) -> None:
-    from ..infrastructure.persistence.json.atomic_write import write_json
-
     data = load_conversation(path)
     data["pinned"] = bool(pinned)
-    write_json(Path(path), data)
+    Path(path).write_text(json.dumps(data, ensure_ascii=False, indent=2), encoding="utf-8")
 
 
 def load_conversation(path: Path) -> Dict[str, Any]:

core/mcp_client.py

@@ -106,13 +106,6 @@ class McpServerConnection:
         if self._thread is not None:
             self._thread.join(timeout=5)
 
-    def is_alive(self) -> bool:
-        """True while the connection's background thread (and therefore its
-        event loop and subprocess) is still running — used by
-        ``infrastructure/mcp/mcp_source_manager.py`` (R05-T05) to tell a live
-        cached connection from one whose subprocess already died."""
-        return self._thread is not None and self._thread.is_alive()
-
     # ---- tools -----------------------------------------------------------
     def list_tool_specs(self) -> List[ToolSpec]:
         """The server's tools, wrapped as :class:`ToolSpec` — the same shape

core/projects.py

@@ -116,12 +116,10 @@ def new_project(name: str, description: str = "", instructions: str = "",
 
 def save_project(project: Project, directory: Path = None) -> Path:
     directory = directory or PROJECTS_DIR
+    directory.mkdir(parents=True, exist_ok=True)
     path = directory / f"{project.project_id}.json"
-    # R06-T02: atomic write — a crash/kill between truncate and write used to
-    # leave a half-written project.json that load_project() then silently
-    # treats as "missing" (see infrastructure/persistence/json/atomic_write.py).
-    from ..infrastructure.persistence.json.atomic_write import write_json
-    write_json(path, asdict(project))
+    path.write_text(json.dumps(asdict(project), ensure_ascii=False, indent=2),
+                    encoding="utf-8")
     return path
 
 

core/task_executors.py

@@ -248,34 +248,9 @@ def _run_agent(ctx, task_type: str, prompt: str, out_dir: Path,
         "'error' (not silently skip it) if it genuinely can't be completed.\n\n"
         f"{prompt}"
     )
-    # One immutable snapshot of this run, then the shared turn service (R04-T05).
-    # The Schedule Task path used to assemble the run_cowork call itself, in
-    # parallel with ui/cowork_tab.py doing the same thing slightly differently -
-    # so a fix to one path silently missed the other. Both now go through
-    # ConversationApplicationService.
-    from ..application.conversations import ConversationApplicationService
-    from ..domain.agents import ConversationExecutionRequest
-
+    messages = [{"role": "user", "content": prompt}]
     session_id = new_session_id()
     project_id = project.project_id if project is not None else ""
-    project_context = projects.project_context_text(project)
-    conversation_service = ConversationApplicationService(
-        # The provider was already resolved above (admin agent / per-task
-        # override / machine default), so the factory just hands it back.
-        lambda _provider_id, _model: provider,
-        security_config=ctx.config,
-    )
-    turn = conversation_service.begin_turn(ConversationExecutionRequest.create(
-        prompt, [{"role": "user", "content": prompt}],
-        output_dir=str(out_dir), session_id=session_id, surface="task",
-        title=title, project_id=project_id, project_context=project_context,
-        # Tags every tool call in the audit log as a scheduled task rather than
-        # as the interactive Cowork tab.
-        agent_role=agent_roles.TASK,
-    ))
-    # The LIVE list the engine appends to - History is re-saved from it after
-    # every assistant message so a long run shows progress when reopened.
-    messages = turn.messages
     _save_history_session(ctx, task_type, title, messages, session_id, project_id)
     # Tell the scheduler the session now genuinely EXISTS on disk — it
     # refreshes History on this, not on the earlier "task_started" signal
@@ -294,21 +269,14 @@ def _run_agent(ctx, task_type: str, prompt: str, out_dir: Path,
         elif ev.get("type") == "plan_set":
             last_plan_steps[:] = ev.get("steps") or []
 
+    project_context = projects.project_context_text(project)
     watched_cancel, timed_out = _cancel_with_timeout(cancel, timeout_sec)
     try:
         if task_type == "cowork":
-            # Typed events are rendered back into the legacy dict shape this
-            # module's autosave/plan tracking already consumes; it moves to
-            # AgentEvent directly once the scheduler UI migrates (EPIC R07/R08).
-            result = conversation_service.execute_turn(
-                turn,
-                on_event=lambda event: emit_and_autosave(event.to_dict()),
-                cancel=watched_cancel,
-            )
-            # This module's callers handle a failed run through an exception
-            # (execute_task writes error.txt from it), so re-raise the ORIGINAL
-            # error rather than reporting a silently empty answer.
-            result.raise_if_failed()
+            from .chat_agent import run_cowork
+            run_cowork(provider, messages, out_dir, emit_and_autosave, watched_cancel,
+                      security_config=ctx.config, agent_role=agent_roles.TASK,
+                      project_context=project_context)
         else:
             from .code_agent import run_code
             limits, block_network = agent_security.sandbox_settings(ctx.config)

core/tools.py

@@ -3,29 +3,79 @@
 Every path is resolved relative to the working directory and must stay inside
 it (path-traversal is rejected). ``run_command`` executes inside the workdir
 with a timeout and captured output.
-
-R05-T02: the actual handlers (``read_file``/``list_dir``/``write_file``/
-``edit_file``/``run_command``/``install_package``/``fetch_url``/
-``jira_search``/``jira_get_issue``) now live in
-``infrastructure/filesystem/{file_tools,command_tools,fetch_tools}.py``, split
-out of what used to be one big if/elif chain here. This module is the
-strangler-fig shim (ADR-001 section 4): it re-exports ``ToolContext``/
-``ToolError`` (actually defined in
-``infrastructure/filesystem/tool_context.py`` now) so every existing
-``from .tools import ToolContext`` keeps working, and ``execute_tool``
-dispatches through a small ``{name: handler}`` table built from the moved
-modules instead of the chain itself.
 """
 from __future__ import annotations
 
+import ast
 import difflib
+import os
+from dataclasses import dataclass
+from pathlib import Path
 from typing import Any, Callable, Dict, List, Optional
 
-from ..infrastructure.filesystem import command_tools, fetch_tools, file_tools
-from ..infrastructure.filesystem.command_tools import _snapshot  # noqa: F401 - re-export, core/chat_agent.py imports this name
-from ..infrastructure.filesystem.tool_context import CancelFn, ToolContext, ToolError  # noqa: F401 - re-export
 from ..providers.base import ToolSpec
 
+CancelFn = Callable[[], bool]
+
+MAX_READ_BYTES = 200_000
+COMMAND_TIMEOUT = 120  # seconds
+
+
+class ToolError(Exception):
+    pass
+
+
+def _flatten_rel(rel: str) -> str:
+    """Collapse a sub-folder path down to a bare filename so the file lands in the
+    workdir root — EXCEPT the ``.scratch`` sandbox subtree, which is preserved.
+
+    Used by the Cowork agent (flatten_writes=True) so it can never create a
+    per-session / per-chat / per-task output sub-folder: every deliverable stays
+    directly in the single configured Output folder."""
+    parts = Path(rel).parts
+    if parts and parts[0] == ".scratch":
+        return rel  # temporary sandbox is allowed (and cleaned up afterwards)
+    return Path(rel).name or rel
+
+
+@dataclass
+class ToolContext:
+    workdir: Path
+    flatten_writes: bool = False  # Cowork: force every write into the workdir root
+    sandbox: bool = False  # Code tab: isolate run_command/install_package into <workdir>/.venv
+    # Sandbox Security Layer — Settings' "Resource Limits" (cpu_percent/memory_mb/
+    # disk_mb), applied to every run_command/install_package this context runs.
+    # None (default) = no limits, matching pre-existing behavior.
+    resource_limits: Optional[Dict[str, float]] = None
+    # Sandbox Security Layer — Settings' "Block network for agent commands"
+    # (policy-level, see deps.py::network_blocked_env). False (default) =
+    # unrestricted, matching pre-existing behavior.
+    block_network: bool = False
+    # Whether the fetch_url tool may read URLs — SEPARATE from block_network
+    # (reading a web page/share link for info is safe; running networked shell
+    # commands is the risk). Defaults True; set from agent_security.allow_url_fetch.
+    allow_url_fetch: bool = True
+    # Jira read connector config (base_url/email/api_token) — None disables the
+    # jira_* tools' ability to connect. Populated from config.data["jira"].
+    jira: Optional[Dict[str, Any]] = None
+
+    def resolve(self, rel: str) -> Path:
+        """Resolve ``rel`` inside the workdir, rejecting escapes."""
+        if rel in ("", "."):
+            return self.workdir
+        candidate = (self.workdir / rel).expanduser()
+        try:
+            resolved = candidate.resolve()
+        except OSError as exc:
+            raise ToolError(f"Invalid path: {rel} ({exc})")
+        root = self.workdir.resolve()
+        if resolved != root and root not in resolved.parents:
+            raise ToolError(
+                f"Refused: '{rel}' is outside the working folder ({root})."
+            )
+        return resolved
+
+
 # --------------------------------------------------------------------------
 # Tool specs advertised to the model
 # --------------------------------------------------------------------------
@@ -142,23 +192,6 @@ TOOL_SPECS: List[ToolSpec] = [
 # Actions gated by the permission gate in confirm mode (auto-approved in Auto-run).
 WRITE_TOOLS = {"write_file", "edit_file", "run_command", "install_package"}
 
-# name -> handler(ctx, args[, cancel, on_output]) — built once from the split
-# infrastructure modules. Replaces the if/elif chain execute_tool used to be.
-_HANDLERS: Dict[str, Callable[..., Dict[str, Any]]] = {
-    "read_file": file_tools.read_file,
-    "list_dir": file_tools.list_dir,
-    "write_file": file_tools.write_file,
-    "edit_file": file_tools.edit_file,
-    "run_command": command_tools.run_command,
-    "install_package": command_tools.install_package,
-    "fetch_url": fetch_tools.fetch_url,
-    "jira_search": fetch_tools.jira_search,
-    "jira_get_issue": fetch_tools.jira_get_issue,
-}
-# Handlers that accept the long-running (cancel, on_output) signature — every
-# other handler takes just (ctx, args).
-_CANCELLABLE = {"run_command", "install_package"}
-
 
 def enabled_tool_specs(security_config=None) -> List[ToolSpec]:
     """The built-in TOOL_SPECS minus any the admin turned OFF in Monitoring →
@@ -268,14 +301,27 @@ def execute_tool(ctx: ToolContext, name: str, args: Dict[str, Any],
     labels WHICH agent role made it."""
     from . import audit_log
 
-    handler = _HANDLERS.get(name)
     try:
-        if handler is None:
-            result = {"ok": False, "output": f"Tool not found: {name}"}
-        elif name in _CANCELLABLE:
-            result = handler(ctx, args, cancel, on_output)
+        if name == "read_file":
+            result = _read_file(ctx, args)
+        elif name == "list_dir":
+            result = _list_dir(ctx, args)
+        elif name == "write_file":
+            result = _write_file(ctx, args)
+        elif name == "edit_file":
+            result = _edit_file(ctx, args)
+        elif name == "run_command":
+            result = _run_command(ctx, args, cancel, on_output)
+        elif name == "install_package":
+            result = _install_package(ctx, args, cancel, on_output)
+        elif name == "fetch_url":
+            result = _fetch_url(ctx, args)
+        elif name == "jira_search":
+            result = _jira_search(ctx, args)
+        elif name == "jira_get_issue":
+            result = _jira_get_issue(ctx, args)
         else:
-            result = handler(ctx, args)
+            result = {"ok": False, "output": f"Tool not found: {name}"}
     except ToolError as exc:
         result = {"ok": False, "output": str(exc)}
     except Exception as exc:  # defensive: a tool must never crash the agent
@@ -285,6 +331,234 @@ def execute_tool(ctx: ToolContext, name: str, args: Dict[str, Any],
     return result
 
 
+def _fetch_url(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    """Fetch a URL's text content (web page / online document / SharePoint-
+    OneDrive share link) via link_fetch — the same parser task-link attachments
+    use. Honors the Sandbox Security Layer's "Block network" policy."""
+    url = str(args.get("url", "")).strip()
+    if not url:
+        return {"ok": False, "output": "fetch_url: 'url' is required."}
+    if not url.lower().startswith(("http://", "https://")):
+        return {"ok": False, "output": f"fetch_url: not an http(s) URL: {url}"}
+    if not ctx.allow_url_fetch:
+        return {"ok": False,
+                "output": ("fetch_url: URL fetching is turned off in Settings → Security "
+                           "(\"Allow the agent to fetch URLs\").")}
+    # A pasted Jira issue link on the CONNECTED Jira host is read via the
+    # authenticated API (so private issues resolve, not a login page). Public
+    # links / any other URL fall through to the normal fetcher below.
+    from . import jira_tool
+    if jira_tool.is_jira_issue_url(ctx.jira, url):
+        return {"ok": True, "output": jira_tool.get_issue_by_url(ctx.jira, url)}
+    from .link_fetch import fetch_link_preview
+
+    return {"ok": True, "output": fetch_link_preview(url)}
+
+
+def _jira_search(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    from . import jira_tool
+
+    out = jira_tool.search(ctx.jira, str(args.get("jql", "")),
+                           int(args.get("max_results", 25) or 25))
+    return {"ok": not out.lower().startswith(("jira is not configured", "jira search failed")),
+            "output": out}
+
+
+def _jira_get_issue(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    from . import jira_tool
+
+    out = jira_tool.get_issue(ctx.jira, str(args.get("key", "")))
+    return {"ok": not out.lower().startswith(("jira is not configured", "could not fetch")),
+            "output": out}
+
+
+def _read_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    target = ctx.resolve(str(args.get("path", "")))
+    if not target.exists():
+        return {"ok": False, "output": f"File not found: {args.get('path')}"}
+    data = target.read_bytes()[:MAX_READ_BYTES]
+    text = data.decode("utf-8", errors="replace")
+    return {"ok": True, "output": text}
+
+
+def _list_dir(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    rel = str(args.get("path", ".") or ".")
+    target = ctx.resolve(rel)
+    # A missing/not-yet-created path is NOT a tool failure — report it as an
+    # ordinary result so the agent can create it or pick another path and keep
+    # going. Returning ok=False here surfaced a false "tool failed: list_dir" in
+    # Co4E flows and could stall a step on a recoverable situation.
+    if not target.exists():
+        return {"ok": True, "output": f"(path '{rel}' does not exist yet — create it or use another path)"}
+    if target.is_file():
+        return {"ok": True, "output": f"('{rel}' is a file, not a directory)"}
+    entries = []
+    for child in sorted(target.iterdir(), key=lambda p: (p.is_file(), p.name.lower())):
+        marker = "/" if child.is_dir() else ""
+        entries.append(f"{child.name}{marker}")
+    return {"ok": True, "output": "\n".join(entries) or "(empty folder)"}
+
+
+def _check_python_syntax(target: Path, content: str) -> str:
+    """Return a short warning if ``content`` is invalid Python, else ''.
+
+    Catches syntax errors the instant a .py file is written/edited — before the
+    agent wastes a whole run_command round-trip just to get the same error back
+    from a traceback."""
+    if target.suffix.lower() not in (".py", ".pyw"):
+        return ""
+    try:
+        ast.parse(content, filename=str(target))
+        return ""
+    except SyntaxError as exc:
+        return f"\n⚠ Syntax error at line {exc.lineno}: {exc.msg} — fix this before running the file."
+
+
+def _write_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    rel = str(args.get("path", ""))
+    if ctx.flatten_writes:
+        rel = _flatten_rel(rel)
+    target = ctx.resolve(rel)
+    content = str(args.get("content", ""))
+    target.parent.mkdir(parents=True, exist_ok=True)
+    # A .xlsx is a binary package — build a REAL workbook from the content
+    # (CSV/TSV/Markdown-table/JSON) rather than writing raw text (which corrupts it).
+    if target.suffix.lower() in (".xlsx", ".xlsm"):
+        from . import xlsx_write
+        if xlsx_write.build_xlsx_from_text(target, content):
+            return {"ok": True, "path": str(target),
+                    "output": f"Wrote spreadsheet {rel} ({target.name})."}
+        return {"ok": False, "output": "Could not build the .xlsx (openpyxl unavailable) — "
+                "write a .csv instead, or use a generator script."}
+    target.write_text(content, encoding="utf-8")
+    warning = _check_python_syntax(target, content)
+    return {"ok": True, "path": str(target),
+            "output": f"Wrote {len(content)} chars to {rel}.{warning}"}
+
+
+def _edit_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
+    """Replace an exact snippet inside an existing file (precise patch edit)."""
+    rel = str(args.get("path", ""))
+    if ctx.flatten_writes:
+        rel = _flatten_rel(rel)
+    target = ctx.resolve(rel)
+    if not target.exists():
+        return {"ok": False,
+                "output": f"File not found: {rel} — use write_file to create it."}
+    old = str(args.get("old_string", ""))
+    new = str(args.get("new_string", ""))
+    replace_all = bool(args.get("replace_all", False))
+    if not old:
+        return {"ok": False, "output": "old_string is empty — provide the exact text to replace."}
+    try:
+        text = target.read_text(encoding="utf-8", errors="replace")
+    except OSError as exc:
+        return {"ok": False, "output": f"Could not read file: {exc}"}
+    count = text.count(old)
+    if count == 0:
+        return {"ok": False, "output": ("old_string not found. Read the file and copy the exact "
+                                        "text to replace, including indentation/whitespace.")}
+    if count > 1 and not replace_all:
+        return {"ok": False, "output": (f"old_string appears {count} times — add surrounding "
+                                        "context to make it unique, or set replace_all=true.")}
+    updated = text.replace(old, new) if replace_all else text.replace(old, new, 1)
+    target.write_text(updated, encoding="utf-8")
+    n = count if replace_all else 1
+    warning = _check_python_syntax(target, updated)
+    return {"ok": True,
+            "output": f"Edited {args.get('path')} ({n} replacement{'' if n == 1 else 's'}).{warning}"}
+
+
+def _sandbox_python(ctx: ToolContext, cancel: Optional[CancelFn] = None,
+                     on_output: Optional[Callable[[str], None]] = None) -> Optional[str]:
+    """Lazily create/reuse this ctx's project sandbox venv (Code tab only —
+    ``ctx.sandbox``); returns its python path, or None to use the app's own."""
+    if not ctx.sandbox:
+        return None
+    from .deps import ensure_project_venv
+
+    py = ensure_project_venv(ctx.workdir, cancel=cancel, on_output=on_output)
+    return str(py) if py else None
+
+
+def _install_package(ctx: ToolContext, args: Dict[str, Any], cancel: Optional[CancelFn] = None,
+                      on_output: Optional[Callable[[str], None]] = None) -> Dict[str, Any]:
+    from .deps import pip_install
+
+    package = str(args.get("package", "")).strip()
+    if not package:
+        return {"ok": False, "output": "No package specified."}
+    python = _sandbox_python(ctx, cancel, on_output)
+    ok, detail = pip_install(package, cancel=cancel, on_output=on_output, python=python)
+    head = f"Installed {package}." if ok else f"Could not install {package}."
+    return {"ok": ok, "output": f"{head}\n{detail}"}
+
+
+_SNAPSHOT_SKIP = {".git", "__pycache__", "node_modules", ".scratch", ".venv",
+                  ".idea", ".mypy_cache", ".pytest_cache"}
+
+
+def _snapshot(workdir: Path) -> Dict[str, Any]:
+    """Map of file path -> (mtime, size) under the workdir (noise dirs skipped)."""
+    snap: Dict[str, Any] = {}
+    try:
+        for dirpath, dirnames, filenames in os.walk(str(workdir)):
+            dirnames[:] = [d for d in dirnames if d not in _SNAPSHOT_SKIP]
+            for fn in filenames:
+                full = os.path.join(dirpath, fn)
+                try:
+                    st = os.stat(full)
+                    snap[full] = (st.st_mtime_ns, st.st_size)
+                except OSError:
+                    pass
+            if len(snap) > 5000:
+                return snap
+    except OSError:
+        pass
+    return snap
+
+
+def _run_command(ctx: ToolContext, args: Dict[str, Any],
+                  cancel: Optional[CancelFn] = None,
+                  on_output: Optional[Callable[[str], None]] = None) -> Dict[str, Any]:
+    from .deps import network_blocked_env, run_cancellable, sandbox_env
+    from .sandbox_manager import SandboxManager, ExecutionConfig
+    from ..security.command_risk_classifier import classify_command
+
+    command = str(args.get("command", "")).strip()
+    if not command:
+        return {"ok": False, "output": "Empty command."}
+
+    # --- Security validation pipeline ---
+    risk = classify_command(command, is_cowork_mode=ctx.flatten_writes)
+    if risk.blocked:
+        denial = "Command blocked by security policy: " + "; ".join(risk.reasons)
+        return {"ok": False, "output": denial}
+
+    # Route through SandboxManager for risk-based isolation
+    mgr = SandboxManager(ExecutionConfig(
+        enabled=True,
+        block_network_by_default=ctx.block_network,
+        is_cowork_mode=ctx.flatten_writes,
+    ))
+    sandbox_result = mgr.run(
+        command=command,
+        workdir=str(ctx.workdir),
+        block_network=ctx.block_network,
+        timeout_sec=COMMAND_TIMEOUT,
+        cancel=cancel,
+    )
+    # Sandbox ALWAYS executes (never double-run). Return its result directly.
+    if sandbox_result.get("sandbox") == "blocked":
+        return {"ok": False, "output": sandbox_result.get("stderr", "Command blocked")}
+    out = sandbox_result.get("stdout", "").strip() or "(no output)"
+    err = sandbox_result.get("stderr", "")
+    rc = sandbox_result.get("returncode", -1)
+    if err:
+        out = f"{out}\n{err}" if out else err
+    return {"ok": sandbox_result.get("ok", False), "output": f"[exit {rc}]\n{out}"}
+
+
 def _short_json(obj: Any, limit: int = 500) -> str:
     import json
     text = json.dumps(obj, ensure_ascii=False, indent=2)

docs/architecture/ADR-001-layered-architecture.md

@@ -1,156 +1,103 @@
-# ADR-001: Kiến Trúc 4 Tầng (Layered / Clean Architecture)
+# ADR-001: 4-Tier Clean Architecture for Desktop Local Application
 
-* **Status**: Accepted
+* **Status**: ACCEPTED / ENFORCED
 * **Date**: 2026-08-21
-* **EPIC / Task**: R01-T01
-* **Owner**: 🔵 Team Duy (Tech Lead)
-* **Áp dụng cho**: toàn bộ mã nguồn mới của `cowork_local` (3 team)
+* **Deciders**: Team Duy (Tech Lead & AI Runtime), Team Nam (Governance & Automation), Team Hoa (Workspace & Scheduling)
+* **Target Project**: Cowork Local (Cowork-Local BamBOO)
 
 ---
 
-## 1. Context (Bối cảnh)
+## 1. Context and Problem Statement
 
-`cowork_local` hiện là một ứng dụng PySide6 desktop local-first ~55.000 dòng Python,
-được phát triển nhanh theo hướng feature-first. Hệ quả đo được tại thời điểm viết ADR:
+Cowork Local is a desktop application written in Python using PySide6 (Qt) and designed for local-first execution.
+Historically, the codebase suffered from architectural coupling across layers:
+1. **God-Widget Problem**: Monolithic UI widgets (e.g., `ui/chat_panel.py` >1,800 LOC, `ui/co4e_tab.py` >1,400 LOC) mixed UI rendering, network I/O, business rules, filesystem operations, and background worker lifecycle.
+2. **Untestable Business Logic**: Core algorithms (model routing, conversation turn management, schedule calculation) were tightly coupled to `PySide6` widgets or `QTimer`, making unit testing in headless CI environments impossible without a graphical display server.
+3. **Circular Dependencies & Global State Leaks**: Uncontrolled module imports (`model_pricing.py` ↔ `usage_tracker.py`, `agent_security.py` ↔ `agent_security_alert.py`) and mutable global state (`state.py::AppContext.active_project_id`) caused race conditions in background task runs.
 
-| Vấn đề | Bằng chứng cụ thể trong repo |
-| :--- | :--- |
-| **God widget** | `ui/co4e_tab.py` 2.089 dòng, `ui/chat_panel.py` 1.795 dòng, `ui/folder_tab.py` 1.590 dòng |
-| **Business logic nằm trong widget** | Vòng đời turn chat, quyết định routing, ghép prompt đều nằm trong `ui/chat_panel.py` |
-| **Logic trùng lặp 3 nơi** | `ui/chat_panel.py::_apply_routing`, `ui/co4e_tab.py::_apply_co4e_routing`, `ui/folder_tab.py::_ai_apply_routing` là ba bản sao gần như y hệt của cùng một thuật toán |
-| **Không test được nếu không có Qt** | Muốn test một quyết định routing phải dựng widget → không chạy được headless, không chạy được nhanh |
-| **Side-effect ẩn trong tầng hạ tầng** | Provider tự gọi `core.usage_tracker.record()` ngay trong vòng lặp stream (`providers/openai_compat.py::_record_usage`) |
+---
 
-Ba team (Duy / Nam / Hoa) sẽ sửa song song trên cùng codebase trong 10 ngày. Nếu
-không có một ranh giới phụ thuộc được **kiểm chứng tự động**, các thay đổi song song
-sẽ hội tụ về đúng cấu trúc rối như cũ.
+## 2. Decision: 4-Tier Clean Architecture
 
-## 2. Decision (Quyết định)
-
-Mã nguồn mới được tổ chức thành **4 tầng**, với **chiều phụ thuộc một chiều** như sau:
+We enforce a strict **4-Tier Clean Architecture** based on the Dependency Inversion Principle:
 
 ```text
 ┌─────────────────────────────────────────────────────────────┐
-│  presentation/            PySide6 widgets, Qt signals/slots  │
-│  (chat, co4e, workspace…) Chỉ dựng UI và phát/nhận signal    │
-└───────────────────────────┬─────────────────────────────────┘
-                            │ gọi xuống (được phép)
-┌───────────────────────────▼─────────────────────────────────┐
-│  application/             Pure Python orchestration          │
-│  (conversations,          Điều phối use-case, không biết Qt  │
-│   model_routing…)         và không biết HTTP/đĩa cụ thể      │
-└───────────────────────────┬─────────────────────────────────┘
-                            │ gọi xuống (được phép)
-┌───────────────────────────▼─────────────────────────────────┐
-│  domain/                  Pure Python entities & events      │
-│  (agents, models…)        Frozen dataclass, enum, quy tắc    │
-│                           nghiệp vụ thuần. KHÔNG import gì   │
-│                           từ 3 tầng còn lại.                 │
-└───────────────────────────▲─────────────────────────────────┘
-                            │ implement interface của domain
-┌───────────────────────────┴─────────────────────────────────┐
-│  infrastructure/          Adapters: network, keyring, đĩa,   │
-│  (providers, telemetry…)  process, Qt-free I/O               │
+│                      PRESENTATION                           │
+│  (PySide6 Widgets, Dialogs, Qt Signals/Slots, View Models)   │
+└──────────────────────────────┬──────────────────────────────┘
+                               │ depends on
+                               ▼
+┌─────────────────────────────────────────────────────────────┐
+│                      APPLICATION                            │
+│  (Use Case Services, Turn Orchestrators, Route Dispatchers) │
+│             *** STRICTLY PURE PYTHON (0 Qt) ***             │
+└──────────────────────────────┬──────────────────────────────┘
+                               │ depends on
+                               ▼
+┌─────────────────────────────────────────────────────────────┐
+│                   DOMAIN & RUNTIME CORE                     │
+│  (Entities, Value Objects, Domain Events, Tool Descriptors)  │
+│             *** STRICTLY PURE PYTHON (0 Qt) ***             │
+└──────────────────────────────▲──────────────────────────────┘
+                               │ implemented by
+┌──────────────────────────────┴──────────────────────────────┐
+│                     INFRASTRUCTURE                          │
+│  (LLM Providers, Keyring Secrets, Atomic Persistence, MCP)  │
 └─────────────────────────────────────────────────────────────┘
 ```
 
-### 2.1 Quy tắc bất biến (Invariants)
+---
 
-| # | Quy tắc | Được kiểm bởi |
-| :--- | :--- | :--- |
-| **I1** | `domain/` và `application/` là **100% pure Python** — cấm import `PySide6`, `PyQt5`, `PyQt6`, `shiboken6` | `scripts/check_imports.py` (R01-T03) |
-| **I2** | `domain/` **không import** `application/`, `infrastructure/`, `presentation/`, `ui/` | `scripts/check_imports.py` |
-| **I3** | `application/` **không import** `presentation/` hay `ui/` | `scripts/check_imports.py` |
-| **I4** | Không file production nào vượt **400 dòng** | `scripts/check_loc.py` (R10-T02) |
-| **I5** | `presentation/` **không** gọi thẳng provider/HTTP/đĩa — phải đi qua một application service | Code review + I1–I3 |
-| **I6** | Mọi input của một use-case được đóng gói thành **snapshot bất biến** (`frozen dataclass`) trước khi rời UI thread | Code review + unit test |
+## 3. Layer Definitions and Responsibilities
 
-### 2.2 Chiều phụ thuộc được phép
+### Tier 1: Presentation Layer (`presentation/`)
+* **Responsibilities**: UI component layout, user event capture, progress display, visual animations, confirmation dialog triggers.
+* **Allowed Imports**: `PySide6.*`, `application.*`, `domain.*`.
+* **Forbidden**: Direct database queries, raw LLM API calls, disk writes outside UI cache, executing tool commands directly.
+* **Constraints**: Every widget file must strictly be **under 400 lines of code (LOC)**.
 
-| Từ tầng | Được import | Bị cấm |
-| :--- | :--- | :--- |
-| `presentation/` | `application/`, `domain/`, PySide6 | — (nên tránh gọi thẳng `infrastructure/`) |
-| `application/` | `domain/`, interface do `domain/` định nghĩa | `presentation/`, `ui/`, PySide6 |
-| `domain/` | chỉ stdlib | tất cả các tầng khác, PySide6 |
-| `infrastructure/` | `domain/`, thư viện ngoài (requests, keyring…) | `presentation/`, `ui/`, PySide6 |
+### Tier 2: Application Layer (`application/`)
+* **Responsibilities**: Orchestrate single use cases (e.g. `ConversationApplicationService`, `RoutingApplicationService`, `TaskApplicationService`). Convert UI requests into domain requests, coordinate domain services with infrastructure adapters.
+* **Allowed Imports**: `domain.*`, `infrastructure.*` interfaces/contracts, standard Python libraries.
+* **Forbidden**: `PySide6`, `PyQt5`, `PyQt6`, `ui.*`, `app.*`.
+* **Nature**: **100% Pure Python**. Must be executable and testable in headless CI environments without a display driver.
 
-### 2.3 Cách tầng dưới "nói chuyện ngược" lên UI
+### Tier 3: Domain Layer (`domain/`)
+* **Responsibilities**: Core domain models, frozen DTO snapshots (`ConversationExecutionRequest`), typed event streams (`AgentEvent`), descriptors (`ToolDescriptor`, `ProviderDescriptor`), deterministic calculation algorithms (`ScheduleCalculator`).
+* **Allowed Imports**: Standard Python library only (`dataclasses`, `typing`, `enum`, `datetime`, `pathlib`, `abc`).
+* **Forbidden**: `PySide6`, `PyQt*`, `requests`, `sqlalchemy`, filesystem mutations, OS network calls.
+* **Nature**: Completely isolated and zero-dependency core.
 
-`application/` **không được** giữ tham chiếu tới widget. Việc trao đổi ngược chiều
-đi qua **callback thuần Python nhận một `AgentEvent` có kiểu**
-(`domain/agents/agent_event.py`, R04-T02):
+### Tier 4: Infrastructure Layer (`infrastructure/`)
+* **Responsibilities**: Adapters for external systems (OpenAI/Anthropic/Ollama/FPT providers, OS Keyring via `SecretStore`, `AtomicJsonFile` persistence, MCP child processes, filesystem tools).
+* **Allowed Imports**: Third-party SDKs, OS libraries, `domain.*`.
+* **Forbidden**: `presentation.*`, `PySide6.QtWidgets`.
 
-```python
-# application layer — pure Python, không biết Qt tồn tại
-service.run_turn(request, on_event=my_callback)
+---
 
-# presentation layer — chuyển event sang Qt signal ở ranh giới duy nhất này
-def my_callback(event: AgentEvent) -> None:
-    self.agent_event.emit(event)   # Qt signal → cập nhật UI trên main thread
-```
+## 4. Architectural Rules and Non-Negotiable Invariants
 
-Đây là **seam** duy nhất giữa hai thế giới: dưới seam là Python thuần test được
-offline, trên seam là Qt. Mọi cập nhật UI phải xảy ra qua Qt signal/slot, không
-bao giờ gọi trực tiếp từ worker thread.
+1. **Zero Qt in Business Logic**:
+   - `domain/` and `application/` must never import `PySide6` or `PyQt*`.
+   - Verified via AST parser script `scripts/check_imports.py`.
+2. **Immutable Request Snapshots**:
+   - Turns are initiated using immutable frozen dataclasses (`ConversationExecutionRequest`) to decouple runtime state from mutable UI state.
+3. **Thread Safety and Signal Decoupling**:
+   - AI generation and tool calls run asynchronously in worker threads.
+   - UI updates occur strictly on the Qt main thread by consuming `AgentEvent` streams through Qt Signal bridges.
+4. **Single Responsibility and Modularity**:
+   - Production files must stay within **400 LOC**.
+5. **English In-Code Comments**:
+   - Every modified or created line/block must include concise English comments explaining design decisions and processing logic.
 
-## 3. Vị trí sở hữu theo team
+---
 
-| Tầng / thư mục | Team | EPIC |
-| :--- | :--- | :--- |
-| `presentation/chat/`, `application/conversations/`, `application/model_routing/`, `domain/agents/`, `domain/models/`, `infrastructure/providers/`, `infrastructure/telemetry/`, `tests/`, `scripts/` | 🔵 Duy | R01, R03, R04, R08, R10 |
-| `presentation/co4e/`, `monitoring/`, `settings/`, `shell/`, `application/workflows/`, `infrastructure/config/`, `secrets/`, `sandbox/` | 🟣 Nam | R02, R08, R09 |
-| `presentation/workspace/`, `folder/`, `scheduling/`, `application/workspaces/`, `scheduling/`, `domain/tools/`, `domain/tasks/`, `infrastructure/filesystem/`, `mcp/`, `persistence/` | 🟢 Hoa | R05, R06, R07, R08 |
+## 5. Consequences and Compliance
 
-## 4. Chiến lược di trú (Strangler Fig, không big-bang)
-
-Code cũ trong `core/`, `ui/`, `providers/` **không bị xoá ngay**. Ta bọc dần:
-
-1. **Tạo seam mới** ở tầng đúng (ví dụ `RoutingApplicationService`).
-2. **Chuyển call site** cũ sang gọi seam mới (`ui/*.py` chỉ còn vài dòng adapter).
-3. **Giữ module cũ làm implementation detail** phía sau seam (ví dụ
-   `application/model_routing/` vẫn gọi xuống `core/routing/` để dùng lại
-   scorer/selector đã có test).
-4. Chỉ khi mọi call site đã đi qua seam mới → cân nhắc gỡ code cũ.
-
-Nhờ vậy `pytest` luôn xanh giữa các bước, và một team có thể merge mà không chờ
-team khác refactor xong.
-
-## 5. Consequences (Hệ quả)
-
-### Tích cực
-
-* Test một quyết định routing / một vòng đời turn chat **không cần Qt, không cần mạng** → suite unit chạy < 1 giây.
-* Ba bản sao logic routing hội tụ về một nơi duy nhất → sửa một lần, cả 3 màn hình cùng đúng.
-* Người mới có thể thêm một provider mà chỉ chạm `infrastructure/providers/` + `domain/models/`.
-* Vi phạm kiến trúc bị chặn ở CI thay vì phát hiện lúc review.
-
-### Tiêu cực / chi phí phải chấp nhận
-
-* Nhiều file nhỏ hơn thay vì vài file lớn → tăng số lần "nhảy file" khi đọc code.
-* Tồn tại **hai đường** trong giai đoạn di trú (code cũ + seam mới) cho tới khi call site cuối cùng chuyển xong.
-* Phải viết DTO/snapshot rõ ràng thay vì truyền thẳng `self` của widget — tốn thêm code, đổi lại được thread-safety.
-
-## 6. Alternatives considered (Phương án đã cân nhắc)
-
-| Phương án | Lý do loại |
-| :--- | :--- |
-| **Giữ nguyên, chỉ tách file cho ngắn** | Giải quyết được I4 (LOC) nhưng không giải quyết được nguyên nhân gốc: logic vẫn dính Qt nên vẫn không test được offline. |
-| **MVVM/MVP thuần Qt** | Vẫn buộc business logic phụ thuộc vòng đời Qt object; không chạy được trong scheduler headless và trong task nền. |
-| **Hexagonal đầy đủ (port/adapter cho mọi thứ)** | Đúng về lý thuyết nhưng quá tốn cho 10 ngày và cho một app desktop 1 process; 4 tầng là điểm cân bằng. |
-| **Big-bang rewrite** | Rủi ro hồi quy quá cao khi 3 team sửa song song và không có bộ test bảo vệ đầy đủ. |
-
-## 7. Enforcement (Thực thi)
-
-```bash
-python scripts/check_imports.py      # I1, I2, I3 — quét AST
-python scripts/check_loc.py          # I4 — giới hạn 400 dòng
-python scripts/run_quality_gate.py   # chạy toàn bộ CASAN Gate + pytest
-```
-
-CASAN Verification Gate phải PASS trước khi merge bất kỳ PR nào vào `main`.
-
-## 8. Tài liệu liên quan
-
-* `docs/refactor/Feature_Architecture_Proposal.md` — thiết kế tổng thể 10 EPIC
-* `docs/refactor/Refactoring_Checklist.md` — bảng tiến độ theo task
-* `docs/architecture/dormant-code.md` — danh mục code không còn hoạt động (R01-T05)
+* **Positive**:
+  - Full testability: Unit tests run in milliseconds without GUI or network mocks.
+  - Zero circular dependencies: Clear top-down data flow.
+  - Resilience: UI crashes do not corrupt background tasks or files.
+* **Verification**:
+  - Automated CI gate: `python scripts/check_imports.py` and `python scripts/check_loc.py`.

docs/architecture/dormant-code.md

@@ -1,85 +1,39 @@
-# Dormant / Dead Code Inventory (R01-T05)
+# Danh Mục & Kế Hoạch Cô Lập Mã Nguồn Dormant / Dead Code (Dormant Code Catalog)
 
-* **Task**: R01-T05 — Phân loại và cô lập mã nguồn cũ
-* **Owner**: 🔵 Team Duy
-* **Ngày quét**: 2026-08-21
-* **Phạm vi quét**: toàn bộ `*.py` production (loại trừ `tests/`, `assets/`, `docs/`, `.git/`)
+* **Tài liệu**: `docs/architecture/dormant-code.md`
+* **Thuộc EPIC**: `R01: Architecture Foundation & Characterization`
+* **Team phụ trách**: 🔵 **Team Duy (Tech Lead)**
 
 ---
 
-## 1. Mục đích
+## 1. Mục Đích & Nguyên Tắc Quản Trị
 
-Trước khi 3 team refactor song song, cần biết **file nào thật sự đang chạy**. Refactor
-một module đã chết là lãng phí; xoá nhầm một module chỉ được gọi động là gây sự cố
-runtime. Tài liệu này phân loại từng ứng viên, kèm **bằng chứng** và **hành động đề xuất**.
+Trong quá trình phát triển nhanh, một số module, hàm hoặc script đã trở thành mã nguồn không hoạt động (**dormant**), mã nguồn thử nghiệm cũ (**legacy prototypes**), hoặc mã nguồn không còn được sử dụng (**dead code**).
 
-## 2. Phương pháp
+> [!IMPORTANT]
+> ### 🛡️ NGUYÊN TẮC CÔ LẬP MÃ NGUỒN CŨ:
+> 1. **Tuyệt đối không import vào các tầng mới**: Các tầng `domain/`, `application/`, `infrastructure/` mới được xây dựng **cấm tuyệt đối import bất kỳ module dormant nào**.
+> 2. **Không xóa vội vàng khi chưa có test bảo vệ**: Giữ nguyên mã nguồn cũ trong giai đoạn tái cấu trúc R01–R08; chỉ dọn dẹp hoặc xóa sau khi bộ kiểm thử khói E2E (EPIC R10) chạy pass 100%.
+> 3. **Phân loại rõ ràng trạng thái**: Mỗi module dormant phải được gắn nhãn (DEPRECATED / ISOLATED / PENDING_DELETION).
 
-Quét AST toàn repo, dựng đồ thị import, tìm module **không có module nào khác import**.
-Kết quả thô: **43 module**. Sau đó xác minh thủ công từng ứng viên, vì phân tích tĩnh
-không thấy 3 kiểu tham chiếu:
+---
 
-| Kiểu tham chiếu ẩn | Ví dụ thật trong repo |
-| :--- | :--- |
-| Chạy như subprocess | `state.py:285` gọi `python -m cowork_local.mcp_servers.ms365_server` |
-| Entry point của gói | `__main__.py` (chạy bằng `python -m cowork_local`) |
-| Script chạy tay | `tools/check_*.py`, `scripts/*.py` |
+## 2. Bảng Danh Mục Mã Nguồn Dormant / Dead Code Đã Rà Soát
 
-> ⚠️ **Kết luận quan trọng**: 43 module "không ai import" **KHÔNG** đồng nghĩa 43 module chết.
-> Sau xác minh, chỉ còn **6 hạng mục (~1.887 dòng)** là dormant thật.
+| STT | File / Module / Ký Hiệu | Trạng Thái Hiện Tại | Lý Do Phân Loại & Phân Tích Kỹ Thuật | Kế Hoạch Xử Lý & Thời Điểm Gỡ Bỏ |
+| :---: | :--- | :---: | :--- | :--- |
+| **1** | `requirements (cloud copy).txt` | `PENDING_DELETION` | File sao chép dự phòng tạm thời trong quá khứ, không được tham chiếu bởi bất kỳ quy trình setup nào. | Gỡ bỏ trong EPIC R10 (Packaging & Clean-up). |
+| **2** | `preview-desktop` | `ISOLATED` | Script shell rỗng/phác thảo cho môi trường dev container cũ. | Cô lập, không liên kết vào build workflow. |
+| **3** | `scripts/bootstrap_gitea_repo.py` | `ISOLATED` | Script tiện ích bootstrap kho lưu trữ Gitea nội bộ; không thuộc runtime ứng dụng chính. | Di chuyển vào `docs/gitea/` làm tài liệu tham khảo ops. |
+| **4** | Hàm routing sao chép tại `ui/chat_panel.py#L638` | `DEPRECATED` | Đoạn code logic chọn model lặp lại từ `core/routing/` nằm trực tiếp trong UI widget. | Thay thế hoàn toàn bằng `RoutingApplicationService` trong EPIC R03. |
+| **5** | Biến toàn cục `state.py::active_project_id` | `DEPRECATED` | Biến global mutable gây race condition khi chạy background task song song. | Thay thế bằng `WorkspaceSession` trong EPIC R06. |
+| **6** | Các hàm xử lý UI đồng bộ trong `core/tools.py` | `DEPRECATED` | `core/tools.py` chứa mã monolithic vừa xử lý file vừa gọi dialog xác thực trực tiếp. | Phân rã thành `file_tools.py`, `command_tools.py` và `ToolPolicyGateway` trong EPIC R05. |
 
-## 3. Phân loại kết quả
+---
 
-### 🟥 A. DORMANT THẬT — không có đường nào chạy tới (ứng viên xoá)
+## 3. Quy Trình Cô Lập & Kiểm Soát
 
-| Module | LOC | Bằng chứng | Rủi ro khi xoá | Hành động |
-| :--- | ---: | :--- | :--- | :--- |
-| `ui/accounts_tab.py` | 700 | Chỉ xuất hiện trong comment của `i18n.py:92`; không widget nào khởi tạo `AccountsTab` | Thấp — panel Monitoring → Accounts hiện không có đường vào | Cô lập, chờ xác nhận PO rồi xoá |
-| `ui/flow_dialog.py` | 596 | Chỉ được nhắc trong docstring `ui/agent_manager_tab.py:4` và comment `i18n.py:2124` | Trung bình — Flow Manager có thể là tính năng tạm ẩn | **Hỏi PO trước**, chưa xoá |
-| `security/` (cả package) | 296 | `prompt_validator`, `action_validator`, `attachment_validator`, `audit_logger`, `command_risk_classifier` — không file nào ngoài package tự import. Chức năng **trùng** `core/agent_security.py` + `core/security_rules.py` (đang chạy thật) | Trung bình — dễ nhầm đây là lớp bảo mật đang hoạt động | ⚠️ Ưu tiên cao: xoá hoặc hợp nhất trong **R09 (Team Nam)** |
-| `core/codebase_memory_ui.py` | 123 | Không nơi nào import; `core/codebase_memory.py` (bản không-UI) mới là bản đang dùng | Thấp | Xoá |
-| `core/graph_server.py` | 115 | Docstring nói phục vụ build không có QtWebEngine, nhưng **không có call site nào**; `ui/structure_graph_view.py` không gọi | Trung bình — có thể là fallback cho bản .exe chưa nối dây | Xác minh với bản đóng gói PyInstaller trước khi xoá |
-| `ui/mcp_servers_dialog.py` | 57 | Không import; MCP settings hiện nằm trong `ui/settings_dialog.py` | Thấp | Xoá |
-
-**Tổng: ~1.887 dòng (≈ 3,4% codebase).**
-
-### 🟨 B. KHÔNG CHẾT — chạy qua đường ẩn (giữ nguyên)
-
-| Module | Vì sao phân tích tĩnh báo nhầm |
-| :--- | :--- |
-| `__main__.py` | Entry point `python -m cowork_local` |
-| `mcp_servers/ms365_server.py` | Chạy như tiến trình con — `state.py:285` |
-| `core/routing/__init__.py` | Được import qua đường dẫn con (`from .routing.service import RoutingService`), heuristic theo tên lá không thấy |
-| `tools/check_*.py` (34 file, 6.608 dòng) | Bộ smoke-test UI chạy tay: `python tools/check_nav.py`. Là **dev tooling**, không phải code chết |
-| `scripts/bootstrap_gitea_repo.py`, `scripts/check_imports.py` | Script CLI chạy tay / chạy trong CI |
-
-### 🟩 C. CODE SỐNG NHƯNG "ĐÓNG BĂNG" — đụng vào phải cẩn thận
-
-| Module | LOC | Ghi chú cho người refactor |
-| :--- | ---: | :--- |
-| `core/chat_agent.py::run_cowork` | 580 | Đang có **characterization test** (`tests/characterization/test_run_cowork.py`, R01-T04). Mọi thay đổi hành vi phải làm cùng lúc với cập nhật snapshot |
-| `providers/base.py` | 401 | Là contract chung của mọi provider; đổi chữ ký = vỡ cả 3 team. Đã có contract test (R03-T01) |
-| `core/routing/*` | 2.263 | Đã có 79 test đang xanh. R03 **bọc** chứ không viết lại: `application/model_routing/` gọi xuống đây |
-
-## 4. Quy tắc xử lý (bắt buộc)
-
-1. **Không xoá trong cùng PR với refactor.** Xoá code chết là một commit riêng, để `git revert` được độc lập khi có sự cố.
-2. **Cô lập trước, xoá sau.** Đánh dấu module bằng docstring cảnh báo, chạy 1 vòng release; không ai báo lỗi mới xoá.
-3. **Hạng mục 🟥 A cần một người xác nhận** (PO hoặc chủ tính năng) trước khi xoá — trừ khi rõ ràng là bản trùng lặp (`codebase_memory_ui`, `mcp_servers_dialog`).
-4. **Không refactor code trong nhóm 🟥 A.** Nếu một file trong danh sách này >400 dòng, nó **không** tính vào CASAN Check 2 — vì đường đi đúng là xoá, không phải tách nhỏ.
-
-## 5. Việc cần bàn giao
-
-| Hạng mục | Team nhận | EPIC |
-| :--- | :--- | :--- |
-| `security/` trùng lặp với `core/agent_security.py` | 🟣 Nam | R09 |
-| `ui/accounts_tab.py`, `ui/flow_dialog.py`, `ui/mcp_servers_dialog.py` | 🟣 Nam (sở hữu `presentation/shell/`, `settings/`) | R08 |
-| `core/graph_server.py`, `core/codebase_memory_ui.py` | 🟢 Hoa (sở hữu `presentation/graph/`) | R06 |
-
-## 6. Cách chạy lại lần quét này
-
-```bash
-python scripts/check_imports.py       # ranh giới kiến trúc (R01-T03)
-# Bản quét đồ thị import dùng cho tài liệu này sẽ được đóng gói thành
-# scripts/find_dormant.py trong R10-T02 (Testing & Governance tooling).
-```
+1. **Kiểm tra tự động qua AST Guard**:
+   - Bộ script `scripts/check_imports.py` tự động quét để đảm bảo không có bất kỳ import mới nào trỏ tới các thành phần đã đánh dấu deprecated.
+2. **Kế hoạch dọn dẹp cuối cùng (Release Phase - 31/08/2026)**:
+   - Sau khi hoàn thành EPIC R10 và pass toàn bộ bài test E2E (`tests/e2e/test_smoke.py`), các file đánh dấu `PENDING_DELETION` sẽ được gỡ bỏ khỏi nhánh `main`.

docs/project-context-mcp-team-guide.md

@@ -0,0 +1,58 @@
+# Project Context MCP — hướng dẫn làm song song
+
+Mục tiêu: hoàn thiện ba tool trên **cùng một server** `project_context`. Không tạo server, registry,
+policy hay error envelope mới. Shared skeleton đã khóa sẵn thứ tự an toàn:
+
+```text
+validate input → policy ALLOW → resolve provider → gọi upstream → validate output
+```
+
+## Chia việc
+
+| Người | Tool | Chỉ sửa | Branch đề xuất |
+|---|---|---|---|
+| Member A | `get_project_issue_context` | `tools/issue_context.py`, `providers/issue.py`, test riêng | `feat/mcp-issue-context` |
+| Member B | `search_project_knowledge` | `tools/knowledge_search.py`, `providers/knowledge.py`, test riêng | `feat/mcp-knowledge-search` |
+| Member C | `get_project_change_context` | `tools/change_context.py`, `providers/change.py`, test riêng | `feat/mcp-change-context` |
+
+Trước khi gửi task, thay `Member A/B/C` bằng username thật trên ba issue. Mỗi người **không sửa**
+`foundation.py`, `registry.py`, `runtime.py`, `server.py` hoặc file của người khác. Nếu shared contract
+cần đổi, mở một PR nhỏ riêng và để cả ba người rebase sau khi PR đó merge.
+
+## Bắt đầu trong 5 phút
+
+1. Chạy `python --version` và xác nhận Python 3.11+ như baseline trong `requirements.txt`.
+2. Tạo branch từ commit template chứa tài liệu này sau khi PR template merge.
+3. Đọc input/output model trong module tool được giao; không thêm field riêng của Gitea/Jira/Redmine.
+4. Implement provider read-only trong module `providers/<tool>.py`; credential chỉ lấy sau policy ALLOW.
+5. Thêm test happy, invalid, not-found, timeout, DENIED với `resolver.calls == 0`, output sai schema,
+   truncation/cursor và source mở được có `revision`.
+6. Chạy:
+
+   ```bash
+   python -m pytest tests/test_project_context_mcp_template.py tests/test_project_context_<tool>.py -q
+   ```
+
+   Lệnh trên chạy trực tiếp từ root repo `cowork_local`; `tests/conftest.py` đã thiết lập import path.
+
+## Definition of Done của từng người
+
+- Tool trả đúng schema, có `project_id` và source gồm `system`, `url`, `revision`, `retrieved_at`.
+- Provider-neutral: đổi Gitea sang GitHub/Jira/Redmine không đổi schema hay tool name.
+- Sai project bị `DENIED` trước khi resolve credential và trước mọi upstream call.
+- Không log/return token; lỗi ngoài dự kiến không lộ exception; read không có side effect.
+- Output lớn có `truncated`, `returned`, `remaining`, `next_cursor`; không cắt im lặng.
+- Test riêng pass, test shared pass, PR chỉ chạm đúng vùng sở hữu trong bảng trên.
+
+## Chạy server sau khi provider đã cấu hình
+
+```bash
+COWORK_MCP_ACTOR_ID=<actor> \
+COWORK_MCP_ORG_UNIT=<org> \
+COWORK_MCP_CUSTOMER=<customer> \
+COWORK_MCP_PROJECT=<project> \
+python -m cowork_local.mcp_servers.project_context_server
+```
+
+Không commit giá trị môi trường hoặc credential. Cowork kết nối bằng stdio với command Python và
+args `-m cowork_local.mcp_servers.project_context_server`.

docs/refactor/BaoCao_TeamDuy_R01_R03_R04.md

@@ -1,233 +0,0 @@
-# BÁO CÁO KẾT QUẢ — TEAM DUY: EPIC R01, R03, R04
-
-* **Dự án**: Cowork Local (Cowork-Local BamBOO)
-* **Team**: 🔵 Team Duy — Core AI, Routing, Turn Runtime & Testing (Tech Lead)
-* **Nhánh**: `feature/deltateam/refactor-plan`
-* **Thời gian thực hiện**: 21/08/2026, 09:56 ➔ 10:56
-* **Ngày báo cáo**: 21/08/2026
-* **Tài liệu gốc**: `Feature_Architecture_Proposal.md`, `Refactoring_Checklist.md`, `DeltaTeam_prompt.md`
-
----
-
-## 1. Tóm tắt điều hành
-
-Hoàn tất **16/16 task** của 3 EPIC được giao trong đợt này: **R01** (nền tảng kiến trúc & lưới an toàn), **R03** (hợp nhất provider & routing), **R04** (vòng đời turn hội thoại). Toàn bộ đã commit và push lên nhánh.
-
-| Chỉ số | Kết quả |
-| :--- | :--- |
-| Task hoàn thành | **16/16** (R01: 5, R03: 6, R04: 5) |
-| Commit | 5 |
-| File thay đổi | 48 (37 file mới, 11 file sửa) |
-| Dòng code | +5.843 / −225 |
-| Test | **243 pass** / 44s |
-| Test suite nhanh (unit + contract + characterization + routing) | **218 pass / 1,22s** |
-| CASAN Check 3 (`scripts/check_imports.py`) | **PASS** — 0 Qt import trong `domain/`, `application/` |
-| File production > 400 dòng | **0** |
-
-**3 lỗi thật được phát hiện và sửa trong quá trình làm** (chi tiết mục 5) — trong đó 1 lỗi deadlock sẽ làm treo ứng dụng ngay ở tin nhắn đầu tiên.
-
----
-
-## 2. Kết quả theo từng EPIC
-
-### 🔹 EPIC R01 — Architecture Foundation & Characterization (5/5)
-
-| Task | Sản phẩm | Ghi chú |
-| :--- | :--- | :--- |
-| R01-T01 | `docs/architecture/ADR-001-layered-architecture.md` | Định nghĩa 4 tầng, chiều phụ thuộc, 6 quy tắc bất biến I1–I6, chiến lược di trú Strangler Fig |
-| R01-T02 | `tests/fakes/fake_provider.py`, `fake_tool_executor.py` | Test double chạy offline, kịch bản hoá, ghi lại mọi lời gọi |
-| R01-T03 | `scripts/check_imports.py` (239 dòng) | Quét AST, bắt cả import tương đối (`from ...ui import x`) và import trong thân hàm |
-| R01-T04 | `tests/characterization/test_run_cowork.py` | **13 test** chụp snapshot hành vi hiện tại của `run_cowork` trước khi R04 đụng vào |
-| R01-T05 | `docs/architecture/dormant-code.md` | Quét đồ thị import: 43 module "không ai import" ➔ xác minh còn **6 hạng mục chết thật (~1.887 dòng)** |
-
-**Điểm đáng chú ý ở R01-T03**: dùng AST thay vì `grep` là bắt buộc — trong repo có nhiều docstring nhắc tên `PySide6` một cách hợp lệ, `grep` sẽ báo nhầm và đội sẽ học cách tắt cổng kiểm duyệt.
-
-**Điểm đáng chú ý ở R01-T05**: 43 module không có importer **không** đồng nghĩa 43 module chết. Sau xác minh thủ công: `__main__.py` là entry point, `mcp_servers/ms365_server.py` chạy bằng subprocess (`state.py:285`), 34 file `tools/check_*.py` là dev tooling chạy tay. Chỉ 6 hạng mục là dormant thật.
-
-### 🔹 EPIC R03 — Model Providers & Routing (6/6)
-
-| Task | Sản phẩm | Ghi chú |
-| :--- | :--- | :--- |
-| R03-T01 | `tests/contracts/test_providers.py` | **29 contract test**; chạy được cả 2 adapter thật mà **không cần mạng** nhờ thay `Provider._request` bằng SSE đóng hộp |
-| R03-T02 | `domain/models/provider_descriptor.py`, `infrastructure/providers/provider_registry.py` | Gom 3 nơi khai báo provider về 1 chỗ |
-| R03-T03 | `application/model_routing/routing_application_service.py` | Pure Python, 4 chế độ: Off / Auto / Manual / **Fallback (mới)** |
-| R03-T04, T05 | `ui/chat_panel.py`, `ui/co4e_tab.py`, `ui/folder_tab.py` | Gỡ 3 bản sao logic routing |
-| R03-T06 | `infrastructure/telemetry/usage_sink.py` | Tách ghi nhận token usage khỏi provider |
-
-**Vấn đề gốc đã giải quyết** — cùng một thuật toán routing tồn tại **3 bản gần giống nhau**:
-
-```
-ui/chat_panel.py::_apply_routing       (~45 dòng)
-ui/co4e_tab.py::_apply_co4e_routing    (~38 dòng)
-ui/folder_tab.py::_ai_apply_routing    (~42 dòng)
-```
-
-Cả 3 đều nằm trong widget Qt ➔ **không thể test nếu không dựng cửa sổ**, và đã bắt đầu lệch nhau (mỗi bản xác định "model hiện tại" một kiểu). Nay cả 3 chỉ còn gọi `ctx.routing_application().route_turn(...)` + một callback xác nhận.
-
-**Chế độ Fallback (mới)**: giữ nguyên model người dùng chọn, **chỉ đổi sau khi model đó lỗi**. Đây là chế độ người dùng cần khi họ tin lựa chọn của mình nhưng vẫn muốn lượt chat sống sót qua sự cố nhà cung cấp.
-
-**Bộ từ vựng mode**: trước đây tuple `("off", "auto", "manual")` bị lặp ở **4 chỗ** (`config.py` × 2, `state.py` × 2). Thêm một mode mà quên một chỗ sẽ **âm thầm hạ lựa chọn của người dùng về "off"**. Nay tập trung vào `normalize_mode()` / `is_valid_mode()`.
-
-### 🔹 EPIC R04 — Agent Runtime & Conversation Service (5/5)
-
-| Task | Sản phẩm | Ghi chú |
-| :--- | :--- | :--- |
-| R04-T01 | `domain/agents/conversation_execution_request.py` | Frozen dataclass, chụp toàn bộ input của 1 turn tại thời điểm submit |
-| R04-T02 | `domain/agents/agent_event.py` (370 dòng) | **13 event có kiểu** thay cho dict không kiểu, kèm cầu nối 2 chiều |
-| R04-T03 | `application/conversations/conversation_application_service.py` | Điều phối vòng đời turn, không import Qt |
-| R04-T04 | `ui/cowork_tab.py::build_job` | Chuyển sang snapshot + service |
-| R04-T05 | `core/task_executors.py::_run_agent` | Chuyển sang **cùng** service (trước đây là bản lắp ráp thứ hai, hơi khác) |
-
-**Vấn đề gốc đã giải quyết** — closure trong `build_job` đọc state của widget **từ trong worker thread**:
-
-```python
-def job(worker):
-    provider = self.build_provider()                  # đọc combo box
-    proj_ctx = project_context_text(load_project(project_id))
-```
-
-Người dùng có thể đổi model, đổi workspace, sửa chỉ dẫn project **trong lúc turn đang chạy**. Turn khi đó chạy trên hỗn hợp state cũ + mới, và hỗn hợp nào phụ thuộc vào thời điểm luồng — đúng loại bug tái hiện mỗi tuần một lần và không bao giờ tái hiện trong test.
-
-**`TurnCompletedEvent`** là tín hiệu kết thúc turn mà engine cũ **hoàn toàn không có**: hiện tại mọi consumer suy ra "xong" từ việc worker thread kết thúc, nên **turn bị huỷ và turn thất bại trông giống hệt nhau** với giao diện.
-
----
-
-## 3. Kiến trúc sau refactor
-
-```text
-presentation/  ui/chat_panel.py, ui/co4e_tab.py, ui/folder_tab.py, ui/cowork_tab.py
-      │        (chỉ dựng UI, mở dialog xác nhận, render thông báo)
-      ▼
-application/   model_routing/routing_application_service.py    ← 4 mode routing
-               conversations/conversation_application_service.py ← vòng đời turn
-      │        (100% pure Python — cổng kiểm duyệt tự động chặn import Qt)
-      ▼
-domain/        agents/conversation_execution_request.py  ← snapshot bất biến
-               agents/agent_event.py                     ← 13 event có kiểu
-               models/provider_descriptor.py             ← catalog provider
-      ▲
-infrastructure/ providers/provider_registry.py   telemetry/usage_sink.py
-```
-
-**Nguyên tắc di trú (ADR-001 mục 4)**: **không viết lại engine**. `core/chat_agent.py::run_cowork` và `core/routing/*` (2.263 dòng, 79 test đang xanh) vẫn là engine bên dưới; tầng application chỉ sở hữu phần trước đây bị trộn vào UI. Nhờ vậy `pytest` luôn xanh giữa các bước và một team có thể merge mà không phải chờ team khác.
-
----
-
-## 4. Bằng chứng kiểm thử
-
-### Phân bố test
-
-| Suite | Số test | Thời gian | Vai trò |
-| :--- | ---: | ---: | :--- |
-| `tests/unit/` | 97 | | Logic thuần, không Qt/mạng |
-| `tests/contracts/` | 29 | | Mọi provider phải thoả cùng bộ cam kết |
-| `tests/characterization/` | 13 | | Chốt hành vi hiện tại của `run_cowork` |
-| `tests/routing/` | 79 | | Có sẵn từ trước, vẫn xanh |
-| **Cộng 4 suite nhanh** | **218** | **1,22s** | ✅ đạt CASAN "A — unit < 1s" |
-| `tests/integration/` | 25 | 42s | Widget Qt thật (offscreen) + provider kịch bản hoá |
-| **Tổng** | **243** | **44s** | |
-
-### Đối chiếu Definition of Done (7 tiêu chí, `DeltaTeam_prompt.md`)
-
-| # | Tiêu chí | Kết quả |
-| :--- | :--- | :--- |
-| 1 | Mọi file < 400 dòng | ✅ Lớn nhất: `agent_event.py` 370 dòng |
-| 2 | 0 import Qt trong `domain/`, `application/` | ✅ `check_imports.py` PASS |
-| 3 | Comment tiếng Anh ở mọi khối sửa/mới | ✅ Docstring + giải thích **lý do**, không chỉ mô tả code |
-| 4 | Có unit/contract test, pass 100% < 1s | ✅ 218 test / 1,22s |
-| 5 | Không hồi quy | ✅ 79 test routing có sẵn vẫn xanh |
-| 6 | Ghi Start/End vào Checklist | ✅ 16 task đã tick kèm mốc thời gian |
-| 7 | Cổng CASAN | ⚠️ `run_quality_gate.py` thuộc **R10-T02**, chưa viết. Check 3 đã có và PASS |
-
-### Ba đường code đã sửa nhưng ban đầu chưa được thực thi
-
-Sau khi hoàn tất 16 task, rà soát lại phát hiện 3 đường code đã bị sửa nhưng **không test nào chạy qua**. Đã bổ sung **18 test**:
-
-| Đường code | Rủi ro nếu bỏ qua | Test bổ sung |
-| :--- | :--- | ---: |
-| `task_executors._run_agent` | Autosave History có thể đóng băng ở tin nhắn đầu | 7 |
-| `_apply_co4e_routing` / `_ai_apply_routing` | Mới chỉ import được, chưa từng gọi hàm | 11 |
-| `confirm_switch(decision)` Manual mode | Thiếu field ➔ **nổ bên trong modal**, nơi khó phát hiện nhất | (nằm trong 11 ở trên) |
-
----
-
-## 5. Ba lỗi thật phát hiện trong quá trình làm
-
-### 🔴 Lỗi 1 — Deadlock khi khởi tạo routing service
-
-`AppContext.routing_application()` giữ `_routing_lock` rồi gọi `routing()`, vốn cũng lấy **chính lock đó**. `threading.Lock` không reentrant ➔ **treo cứng ngay ở tin nhắn đầu tiên**, không có thông báo lỗi.
-
-*Sửa*: tách `_routing_app_lock` riêng, và resolve engine **trước khi** lấy lock.
-
-### 🟠 Lỗi 2 — Event `notice` bị cầu nối nuốt mất
-
-Bản đầu của `agent_event.py` liệt kê 12 loại event nhưng **thiếu `notice`**. Trong khi đó `notice` được phát ra từ 3 nơi trên đường chạy bình thường:
-
-* `core/agent_security.py` — yêu cầu/lệnh bị Agent Security **chặn**
-* `core/context_budget.py` — hội thoại vừa bị tự động nén
-* Bộ đọc file đính kèm — file không xử lý được, và tiến độ "đang đọc trang X/Y"
-
-Cầu nối bỏ qua event không nhận diện được (đúng thiết kế, để engine có thể thêm event mới) — nên **người dùng sẽ không bao giờ thấy cảnh báo bảo mật**, hoàn toàn im lặng.
-
-*Sửa*: thêm `NoticeEvent`, **và** thêm test quét mã nguồn engine tìm mọi tag `emit({"type": ...})` rồi bắt lỗi nếu có tag nào chưa có event tương ứng — biến sự im lặng thành test đỏ.
-
-### 🟡 Lỗi 3 — Test đang chạy trên checkout khác
-
-`tests/routing/conftest.py` đẩy thư mục cha vào `sys.path`. Vì thư mục checkout tên là `cowork_local_gitea` (không phải `cowork_local`), lệnh `import cowork_local` **ăn nhầm sang `Desktop\cowork_local`** — một bản checkout khác. Suite báo xanh trên mã nguồn **không phải nhánh đang review**.
-
-*Sửa*: `tests/conftest.py` nạp `__init__.py` theo đường dẫn tuyệt đối và đăng ký vào `sys.modules` trước mọi test.
-
----
-
-## 6. Cải thiện phụ (không nằm trong yêu cầu task)
-
-| Cải thiện | Ảnh hưởng |
-| :--- | :--- |
-| `ProviderRegistry.build()` đóng dấu `descriptor.id` lên instance | Sửa việc usage của `ollama` / `github_copilot` / `codex` bị ghi nhận nhầm thành `openai_compat` trên Dashboard. **Chưa nối vào production** — xem mục 7. |
-| `ProviderRegistry.build()` copy config trước khi ghi | Trước đây một model do routing chọn có thể ghi đè lên default đã lưu của người dùng |
-| `UsageTrackerSink` ghi log ở mức debug khi thất bại | Trước là `except: pass` — mất sạch lý do khi Dashboard hỏng |
-| `estimate_tokens` được chốt bằng test so với `core.usage_tracker` | Bảo đảm việc tách telemetry **không làm lệch một con số nào** |
-
----
-
-## 7. Còn nợ & cần quyết định
-
-| # | Nội dung | Người quyết |
-| :--- | :--- | :--- |
-| 1 | **`ProviderRegistry` chưa nối vào `state.build_provider_for`** (vẫn dùng `providers/factory.py`). Nối vào sẽ sửa lỗi quy kết usage ở mục 6, **nhưng đổi cách gom dữ liệu lịch sử trên Dashboard**. | Team Duy + PO |
-| 2 | **Mode `fallback` chưa có trên toggle UI** — config và service đã hỗ trợ đầy đủ; widget `RoutingToggle` thuộc R08. | Team Duy (R08) |
-| 3 | **Đã sửa 2 dòng trong `config.py`** (`routing_mode_for`, `set_routing_mode_for`) để dùng chung bộ từ vựng mode. File này Team Nam đang refactor ở R02-T02. | ⚠️ **Cần báo Team Nam** |
-| 4 | **Circular import** `core/model_pricing.py` ↔ `core/usage_tracker.py` chưa xử lý (task ngày 28/08). | Team Duy |
-| 5 | **2 test đỏ có sẵn từ trước**: `config.py:108` hardcode `sandbox_pw = "quandh14"` ➔ `tests/test_config_security.py`. Thuộc **EPIC R02 / Team Nam**. | 🟣 Team Nam |
-| 6 | `tests/integration/test_routing_surfaces.py` mất 41s do dựng `Co4ETab`/`FolderTab`. Nên gắn marker `slow` khi làm R10. | Team Duy (R10) |
-
----
-
-## 8. Phạm vi chưa kiểm thử
-
-Nêu rõ để tránh hiểu nhầm mức độ bảo đảm:
-
-* **Chưa mở ứng dụng bằng tay** — mới chạy widget headless (`QT_QPA_PLATFORM=offscreen`), chưa có ai kiểm tra bằng mắt.
-* **Chưa gọi provider thật** — toàn bộ dùng `FakeProvider`, không có lưu lượng mạng.
-* **Chưa chạy 34 script `tools/check_*.py`** — các script này tự `sys.path.insert` thư mục cha nên sẽ import nhầm checkout khác (đúng lỗi 3 ở mục 5). Cần sửa chúng ở R10.
-
----
-
-## 9. Việc kế tiếp của Team Duy
-
-| EPIC | Nội dung | Điều kiện |
-| :--- | :--- | :--- |
-| **R08** (T01 ➔ T06) | Tách `ui/chat_panel.py` (1.795 dòng) thành 6 widget < 400 dòng | Sẵn sàng bắt đầu — `AgentEvent` (R04-T02) chính là kênh dữ liệu 6 widget con sẽ dùng thay vì đọc trực tiếp state của `ChatPanel` |
-| **R10** (T01 ➔ T05) | Testing Pyramid, `run_quality_gate.py`, Contributor Recipes, E2E Smoke | Chờ cả 3 team hoàn tất |
-
----
-
-## 10. Lịch sử commit
-
-| Commit | Nội dung |
-| :--- | :--- |
-| `bbc09f6` | feat(R01): architecture foundation, offline fakes and characterization net |
-| `96bec97` | feat(R03): unify provider catalogue, routing decisions and usage telemetry |
-| `a53163e` | feat(R04): immutable turn snapshot, typed agent events, conversation service |
-| `15e1d3e` | test(R03/R04): cover the three code paths that were changed but never executed |
-| `67b8d2e` | docs(refactor): correct the Team Duy scope block in the checklist |

docs/refactor/BaoCao_TeamHoa_R05_R06.md

@@ -1,198 +0,0 @@
-# BÁO CÁO KẾT QUẢ — TEAM HOA: EPIC R05, R06
-
-* **Dự án**: Cowork Local (Cowork-Local BamBOO)
-* **Team**: 🟢 Team Hoa — Workspace, Filesystem, Scheduling & Tool Registry
-* **Nhánh**: `feature/teamhoa/r05-r06` (tạo từ `origin/feature/deltateam/refactor-plan`, chưa push lên remote — xem mục 7)
-* **Thời gian thực hiện**: 21/08/2026, 21:40 ➔ 22:57
-* **Ngày báo cáo**: 22/08/2026
-* **Tài liệu gốc**: `Feature_Architecture_Proposal.md`, `Refactoring_Checklist.md`, `plan.md`
-
----
-
-## 1. Tóm tắt điều hành
-
-Hoàn tất **10/10 task** của 2 EPIC được giao: **R05** (Tool, MCP & Connector Policy) và **R06** (Workspace, Filesystem & History Isolation). Đã commit 2 commit trên branch cục bộ; **chưa push lên Gitea** — remote từ chối với lỗi quyền ghi (xem mục 7 #1).
-
-| Chỉ số | Kết quả |
-| :--- | :--- |
-| Task hoàn thành | **10/10** (R05: 5, R06: 5) |
-| Commit | 2 (`ae4fe72`, `cf542b7`) |
-| File thay đổi | 41 (27 file mới, 14 file sửa — 1 file (`docs/refactor/Refactoring_Checklist.md`) sửa ở cả 2 commit) |
-| Dòng code | +3.054 / −459 |
-| Test | **283 pass** / 12,5s (283/287 — 4 fail có sẵn từ trước, không do R05/R06) |
-| Test suite nhanh (unit + contract + characterization + routing) | **256 pass / 4,5s** |
-| CASAN Check 3 (`scripts/check_imports.py`) | **PASS** — 0 Qt import trong `domain/`, `application/` |
-| File production > 400 dòng (file mới) | **0** — lớn nhất `domain/tools/tool_registry.py` 125 dòng |
-
-**2 lỗi thật được phát hiện và sửa trong quá trình làm** (chi tiết mục 5): một lỗ hổng bảo mật (MCP/connector tool không qua permission gate) và một race condition (turn chạy ngầm lưu nhầm lịch sử vào project khác).
-
----
-
-## 2. Kết quả theo từng EPIC
-
-### 🔹 EPIC R05 — Tool, MCP & Connector Policy (5/5)
-
-| Task | Sản phẩm | Ghi chú |
-| :--- | :--- | :--- |
-| R05-T01 | `domain/tools/tool_descriptor.py`, `tool_registry.py` | `ToolCapability` (Flag: READ/WRITE/EXECUTE/NETWORK, kết hợp được) + `ToolDescriptor` + `ToolRegistry` |
-| R05-T02 | `infrastructure/filesystem/{file_tools,command_tools,fetch_tools,tool_context}.py` | Tách if/elif dispatcher của `core/tools.py`; `core/tools.py` còn 291 dòng (từ 566), là shim strangler-fig |
-| R05-T03 | `application/conversations/tool_policy_gateway.py` | `ToolPolicyGateway.allow(name, gate, payload)` — thay 2 chỗ check hardcode riêng biệt (`chat_agent.py`, `code_agent.py`) bằng 1 lookup capability |
-| R05-T04 | Sửa `core/chat_agent.py`, `core/mcp_client.py` | **Thay đổi hành vi có chủ đích** — xem mục 5, Lỗi 1 |
-| R05-T05 | `infrastructure/mcp/mcp_source_manager.py` | Tách lifecycle connection MCP khỏi `state.py::AppContext` |
-
-**Vấn đề gốc đã giải quyết** — cùng một việc "tool này có cần xác nhận trước khi chạy không" tồn tại **3 cách trả lời khác nhau**:
-
-```
-core/chat_agent.py::run_cowork    name in ("run_command", "install_package")
-core/code_agent.py::run_code      name in (WRITE_TOOLS | MS365_WRITE_TOOLS)
-core/mcp_client.py / ext_connectors.py    (không hỏi gì cả)
-```
-
-Cách thứ 3 là một lỗ hổng thật, không phải khác biệt thiết kế — xem mục 5.
-
-### 🔹 EPIC R06 — Workspace, Filesystem & History Isolation (5/5)
-
-| Task | Sản phẩm | Ghi chú |
-| :--- | :--- | :--- |
-| R06-T01 | `domain/workspaces/workspace_session.py` | `WorkspaceSession` — snapshot bất biến (project_id/workspace_root/sandbox_dir/allowed_paths) + `is_allowed(path)`, cùng khuôn với `ConversationExecutionRequest` (R04-T01) |
-| R06-T02 | `infrastructure/persistence/json/{atomic_write,workspace_repository_impl,conversation_repository_impl}.py` | **Sửa bug thật** — xem mục 5, Lỗi 2 |
-| R06-T03 | `infrastructure/filesystem/execution_workspace.py` | Đặt tên cho quy ước `.scratch` đã có, không đổi vị trí file |
-| R06-T04 | Sửa `ui/chat_panel.py` | **Sửa race condition thật** — xem mục 5, Lỗi 3 |
-| R06-T05 | `application/workspaces/file_workspace_service.py` | File Explorer/AI Editor gọi `core/tools.py::execute_tool` giống agent, không viết lại logic |
-
----
-
-## 3. Kiến trúc sau refactor
-
-```text
-presentation/  (chưa đổi ở đợt này — ui/chat_panel.py chỉ thêm 1 field "home_history_dir")
-      │
-      ▼
-application/   conversations/tool_policy_gateway.py        ← ALLOW/CONFIRM cho mọi tool call
-               workspaces/file_workspace_service.py         ← file ops cho File Explorer/AI Editor
-      │        (100% pure Python — check_imports.py chặn import Qt)
-      ▼
-domain/        tools/{tool_descriptor,tool_registry}.py     ← capability + catalogue
-               workspaces/workspace_session.py              ← snapshot workspace bất biến
-      ▲
-infrastructure/ filesystem/{file_tools,command_tools,fetch_tools,tool_context,execution_workspace}.py
-                mcp/mcp_source_manager.py                   ← lifecycle connection MCP
-                persistence/json/{atomic_write,*_repository_impl}.py
-```
-
-**Nguyên tắc di trú (ADR-001 mục 4, tiếp nối cách Team Duy làm ở R04)**: **không viết lại engine**. `core/tools.py::execute_tool`, `core/chat_agent.py::run_cowork`, `core/code_agent.py::run_code` vẫn là engine bên dưới — tầng mới chỉ sở hữu phần phân loại rủi ro (R05) và phần định danh workspace (R06) mà trước đây nằm rải rác/hardcode. `pytest` xanh liên tục giữa các bước.
-
----
-
-## 4. Bằng chứng kiểm thử
-
-### Phân bố test (bao gồm test mới của Team Hoa)
-
-| Suite | Số test | Ghi chú |
-| :--- | ---: | :--- |
-| `tests/unit/` | 137 | +41 test mới (R05: 26, R06: 15 — không tính `test_history_dir_race.py`, ở `integration/`) |
-| `tests/contracts/` | 29 | có sẵn từ R03, không đổi |
-| `tests/characterization/` | 13 | có sẵn từ R01, vẫn xanh — xác nhận `run_cowork` không hồi quy sau khi sửa gate |
-| `tests/routing/` | 79 | có sẵn từ trước, không đụng |
-| **Cộng 4 suite nhanh** | **256** (4 fail routing-env, không do R05/R06) | 4,5s |
-| `tests/integration/` | 27 | +2 test mới: `test_history_dir_race.py` — Qt offscreen thật, không phải test double |
-| **Tổng** | **287** (283 pass) | 12,5s |
-
-### Đối chiếu Definition of Done (theo `DeltaTeam_prompt.md` / mẫu Team Duy)
-
-| # | Tiêu chí | Kết quả |
-| :--- | :--- | :--- |
-| 1 | Mọi file mới < 400 dòng | ✅ Lớn nhất: `domain/tools/tool_registry.py` 125 dòng |
-| 2 | 0 import Qt trong `domain/`, `application/` | ✅ `check_imports.py` PASS |
-| 3 | Comment tiếng Anh giải thích lý do ở mọi khối sửa/mới | ✅ |
-| 4 | Có unit/contract/integration test, verify bằng chạy thật | ✅ 41 test mới + 2 test Qt offscreen thật cho race condition |
-| 5 | Không hồi quy | ✅ 283/287 pass — 4 fail là lỗi có sẵn từ trước R05/R06 (2 EPIC R02, 2 do môi trường máy có Ollama thật) |
-| 6 | Ghi Start/End vào Checklist | ✅ 10 task đã tick kèm mốc thời gian |
-| 7 | Cổng CASAN (`run_quality_gate.py`, R10-T02) | ⚠️ Chưa viết (thuộc R10, chưa tới lượt) — Check 3 đã PASS |
-
----
-
-## 5. Hai lỗi thật phát hiện và sửa trong quá trình làm
-
-### 🔴 Lỗi 1 (R05-T04) — Tool MCP/Connector chạy hoàn toàn không qua permission gate
-
-`core/chat_agent.py::run_cowork` có 2 nhánh dispatch tool call: nhánh built-in (`read_file`, `run_command`, ...) đi qua gate xác nhận khi Settings bật "confirm before running commands"; nhánh `extra_tools` (mọi tool từ MCP server hoặc Connector — `core/mcp_client.py`, `core/ext_connectors.py`) gọi thẳng:
-
-```python
-if name in extra_names and extra_executor is not None:
-    ...
-    result = extra_executor(name, args)   # KHÔNG có bước xác nhận nào
-```
-
-Nghĩa là một MCP server (kể cả server tự cấu hình, hoặc MS365 write-tool như `send_mail`) chạy **auto-run tuyệt đối**, bất kể người dùng đã bật "confirm before running commands" trong Settings hay chưa. Đây không phải khác biệt thiết kế có chủ đích — không có ghi chú, không có toggle riêng cho việc này.
-
-*Sửa*: mọi `extra_tools` được gắn `ToolCapability` mặc định bảo toàn (`WRITE|EXECUTE|NETWORK` — vì MCP không có chuẩn khai báo rủi ro), đăng ký vào registry của turn, và đi qua CÙNG `ToolPolicyGateway` với built-in tools.
-
-**Đây là thay đổi hành vi người dùng sẽ thấy**: khi "confirm before running commands" đang bật, tool MCP/connector từ giờ sẽ hỏi xác nhận — giống `run_command`. Verify bằng test `tests/unit/test_cowork_extra_tool_policy.py` (3 test: rejected trước khi executor chạy, approved thì chạy, `gate=None` vẫn auto-run như cũ).
-
-### 🟠 Lỗi 2 (R06-T04) — Turn chạy ngầm lưu nhầm lịch sử vào project khác
-
-`ui/chat_panel.py::_persist_session` (lưu hội thoại của một turn **chạy ngầm**, không phải conversation đang xem) gọi:
-
-```python
-save_conversation(self.ctx.config.history_dir(), ...)
-```
-
-`history_dir()` đọc `config._project_history_dir` — một field **dùng chung** trên `AppContext.config`, được `ui/workspace_tab.py::_load_current` ghi đè mỗi lần người dùng đổi project trong màn Workspace. Nếu một turn ở project A còn đang chạy (ví dụ Scheduled Task, hoặc user gõ câu hỏi rồi chuyển sang xem project B ngay) và người dùng đổi sang project B **trước khi** turn đó lưu xong, hội thoại của project A bị ghi nhầm vào thư mục lịch sử của project B.
-
-*Sửa*: thêm `"home_history_dir"` vào dict `ctx` mà mỗi turn đã có sẵn (cùng quy ước với `home_id`/`home_messages`/`home_title` — dict này được author code gốc thiết kế đúng cho mục đích này, chỉ thiếu 1 field), chụp giá trị **tại lúc submit** thay vì đọc sống lúc lưu.
-
-*Kèm 1 phát hiện phụ*: `_save_snapshot` (dùng cho conversation ĐANG XEM) đã có logic đúng từ trước để không ghi đè `project_id` của một turn nền bằng project hiện tại — chỉ riêng **thư mục lưu** là bị bỏ sót, không phải toàn bộ cơ chế bị thiếu.
-
-Verify bằng test Qt offscreen thật (không phải double): `tests/integration/test_history_dir_race.py` — dựng `ChatPanel` thật, giả lập đổi project giữa lúc turn chạy, xác nhận file được lưu đúng thư mục project A.
-
----
-
-## 6. Cải thiện phụ (không nằm trong yêu cầu task)
-
-| Cải thiện | Ảnh hưởng |
-| :--- | :--- |
-| `core/projects.py::save_project`, `core/history.py::save_conversation/rename_conversation/set_pinned` chuyển sang ghi atomic (`infrastructure/persistence/json/atomic_write.py`) | Trước đây `path.write_text(json.dumps(...))` không atomic — crash/kill giữa lúc ghi để lại file JSON hỏng, và `load_project`/`load_conversation` coi file hỏng như "không tồn tại" ➔ **mất project hoặc hội thoại âm thầm, không báo lỗi**. Có test giả lập crash giữa lúc ghi xác nhận file cũ không bị hỏng (`tests/unit/test_atomic_write_and_repositories.py`) |
-| `McpServerConnection.is_alive()` (mới, `core/mcp_client.py`) | Nhỏ, cộng thêm — cho `McpToolSourceManager` biết một connection cached đã chết (subprocess crash) để khởi động lại, thay vì cache giữ một connection chết vô thời hạn |
-
----
-
-## 7. Còn nợ & cần quyết định
-
-| # | Nội dung | Người quyết |
-| :--- | :--- | :--- |
-| 1 | **Branch chưa lên được Gitea** — `git push` bị từ chối: `User permission denied for writing` (pre-receive hook). Cần cấp quyền push cho tài khoản git đang dùng trên máy này, hoặc push bằng tài khoản khác có quyền. | Admin Gitea |
-| 2 | **Xung đột file với EPIC R02 (Team Nam)**: R02-T01 giao `infrastructure/persistence/json/atomic_json_file.py`. R06-T02 cần atomic write ngay nên tạo `atomic_write.py` (tên khác, cùng thư mục) — không đụng file của Team Nam, nhưng 2 module cùng mục đích sẽ tồn tại song song cho tới khi hợp nhất. | Team Nam (khi bắt đầu R02-T01) |
-| 3 | **`WorkspaceRepository`/`ConversationRepository`/`FileWorkspaceService` chưa có call site thật** — giống tình trạng `ProviderRegistry` của Team Duy ở R03 (mục 7 #1 trong báo cáo Team Duy). Mọi nơi trong production vẫn gọi trực tiếp `core/projects.py`/`core/history.py`/`core/tools.py::execute_tool`. | Team Hoa (nối dây ở EPIC sau) |
-| 4 | **R06-T04 không sửa đúng y nguyên `ui/workspace_tab.py::_load_current` như mô tả gốc trong `plan.md`** — bug thật nằm ở điểm ĐỌC (`ui/chat_panel.py::_persist_session`), không phải điểm GHI (`_load_current` chỉ set field, tự nó không đọc lại). Đã sửa đúng điểm đọc, có test thật xác nhận. Việc đổi `_load_current` sang "đồng bộ bằng session id" như plan gốc gợi ý cần tách sâu hơn `WorkspaceTab`/`ChatPanel`, thuộc phạm vi R08 (UI/Application Separation). | Team Duy (R08) |
-| 5 | **2 test đỏ có sẵn từ trước, không do R05/R06**: `tests/test_config_security.py` × 2 (EPIC R02/Team Nam, đã ghi nhận từ báo cáo Team Duy) và `tests/unit/test_routing_wiring.py` × 2 (môi trường máy này có Ollama/llama3.1 thật + config routing cục bộ khác giả định "fresh install" của test — nghi là do máy chạy test có cấu hình routing/Ollama khác máy Team Duy dùng, cần Team Duy xác nhận lại trên máy sạch). | Team Nam (#1), Team Duy (#2) |
-
----
-
-## 8. Phạm vi chưa kiểm thử
-
-Nêu rõ để tránh hiểu nhầm mức độ bảo đảm:
-
-* **R05-T04 (gate cho MCP/connector) chưa test với MCP server thật** — toàn bộ test dùng `ToolSpec` giả (`_EXTRA_SPEC` trong `test_cowork_extra_tool_policy.py`), chưa có tình huống thật với `core/mcp_client.py::McpServerConnection` chạy subprocess thật.
-* **`McpToolSourceManager` (R05-T05) chưa test với subprocess MCP thật** — test dùng `_FakeConnection`, không spawn tiến trình. Đã smoke-test `AppContext.build_mcp_tools()` thật (không có server nào cấu hình → chỉ trả về ms365 local tools) nhưng chưa thử ensure/restart trên một server thật.
-* **`ui/folder_tab.py`, `ui/file_edit_dialog.py` chưa được nối vào `FileWorkspaceService` (R06-T05)** — dịch vụ tồn tại và có test unit đầy đủ, nhưng chưa xác nhận bằng cách chạy UI thật (đã mở app kiểm tra sau R05, nhưng không lặp lại cho R06's file explorer flow cụ thể).
-* **Đã mở app thật 1 lần sau khi sửa `ui/chat_panel.py` (R06-T04)** để xác nhận không crash lúc khởi động — chưa thử tay thao tác "đổi project giữa lúc chat đang trả lời" trên UI thật (chỉ verify bằng test offscreen).
-
----
-
-## 9. Việc kế tiếp của Team Hoa
-
-| EPIC | Nội dung | Điều kiện |
-| :--- | :--- | :--- |
-| **R07** (Scheduling & Workflow Runtime) | Tách `TaskRepository`/`ScheduleCalculator` khỏi `QTimer` (`core/task_scheduler.py`), xây `TaskApplicationService` | Phối hợp 🟣 Team Nam (Co4E Workflows) |
-| **R08** (T01 ➔ ...) | Phần Team Hoa trong tách UI (`ui/workspace_tab.py`, `ui/folder_tab.py`, `ui/schedule_task_tab.py`, `ui/dashboard_tab.py`, Graph) | Chờ R07 |
-| Nối `WorkspaceRepository`/`ConversationRepository`/`FileWorkspaceService` vào call site thật | Xem mục 7 #3 | Có thể làm sớm hơn R07/R08 nếu được yêu cầu |
-
----
-
-## 10. Lịch sử commit
-
-| Commit | Nội dung |
-| :--- | :--- |
-| `ae4fe72` | feat(R05): tool capability registry, unified policy gateway, MCP lifecycle manager |
-| `cf542b7` | feat(R06): workspace session snapshot, atomic persistence, history-dir race fix |

docs/refactor/Refactoring_Checklist.md

@@ -20,83 +20,6 @@
 
 ---
 
-## 📊 TIẾN ĐỘ THỰC TẾ — TEAM DUY (cập nhật `2026-08-21 10:55`)
-
-> [!NOTE]
-> ### ✅ ĐÃ HOÀN TẤT: 16/16 task của **R01, R03, R04** — đã commit & push lên nhánh `feature/deltateam/refactor-plan`
->
-> | EPIC | Task | Trạng thái |
-> | :--- | :--- | :--- |
-> | **R01** Architecture Foundation | T01 → T05 | ✅ 5/5 |
-> | **R03** Providers & Routing | T01 → T06 | ✅ 6/6 |
-> | **R04** Agent Runtime & Conversation | T01 → T05 | ✅ 5/5 |
->
-> **Kiểm chứng (chạy thật, không phải ước lượng):**
-> * `pytest tests/` ➔ **243 pass / 2 fail** trong 44s
-> * Suite nhanh (`unit + contracts + characterization + routing`) ➔ **218 pass trong 1,16s** (đạt yêu cầu CASAN "A – Automated Tests < 1s cho unit")
-> * `python scripts/check_imports.py` ➔ **PASS** (0 Qt import trong `domain/`, `application/`)
-> * Mọi file production mới **< 400 dòng** (lớn nhất: `routing_application_service.py` 353 dòng)
-> * 2 test fail là **lỗi có sẵn từ trước**, thuộc EPIC **R02**: `config.py` vẫn hardcode `sandbox_pw = "quandh14"` ➔ `tests/test_config_security.py` đỏ
->
-> ### 📍 PHẠM VI TEAM DUY & PHẦN CÒN LẠI
-> Theo `Feature_Architecture_Proposal.md` (dòng 7) và `DeltaTeam_prompt.md` (dòng 17), Team Duy chủ trì **R01, R03, R04, R08 (phân hệ Chat UI), R10**.
-> * ✅ **R01, R03, R04** — xong 16/16 task, đã push.
-> * ⬜ **R08 (R08-T01 ➔ R08-T06)** — chưa bắt đầu: tách `ui/chat_panel.py` (1.795 dòng) thành 6 widget < 400 dòng.
-> * ⬜ **R10** — làm sau cùng, chờ 3 team hoàn tất.
-> * **R02 thuộc 🟣 Team Nam** (xem mục EPIC R02 bên dưới) — đây là nguyên nhân 2 test đỏ ở trên, không phải việc của Team Duy.
->
-> ### 📄 BÁO CÁO CHI TIẾT
-> Xem `docs/refactor/BaoCao_TeamDuy_R01_R03_R04.md` — kết quả từng EPIC, bằng chứng kiểm thử, 3 lỗi thật đã phát hiện, và phạm vi **chưa** kiểm thử.
->
-> ### 📌 CÒN NỢ / CẦN QUYẾT ĐỊNH
-> 1. `ProviderRegistry` **chưa nối** vào `state.build_provider_for` (vẫn dùng `providers/factory.py`). Nối vào sẽ sửa luôn lỗi: usage của `ollama`/`github_copilot`/`codex` hiện bị ghi nhận nhầm thành `openai_compat` trên Dashboard — nhưng làm vậy sẽ **đổi cách gom dữ liệu lịch sử**.
-> 2. Mode `fallback` đã hỗ trợ ở config + service nhưng **chưa có trên toggle UI** (thuộc R08).
-> 3. Đã sửa 2 dòng trong `config.py` (`routing_mode_for` / `set_routing_mode_for`) để dùng chung một bộ từ vựng mode — **cần báo Team Nam** vì file này đang được refactor ở R02.
-> 4. Circular import `core/model_pricing.py` ↔ `core/usage_tracker.py` **chưa xử lý** (task ngày 28/08).
-> 5. Việc kế tiếp của Team Duy là **R08 phân hệ Chat UI** (6 widget con), rồi **R10** sau cùng.
-
----
-
-## 📊 TIẾN ĐỘ THỰC TẾ — TEAM HOA (cập nhật `2026-08-21 22:57`)
-
-> [!NOTE]
-> ### ✅ ĐÃ HOÀN TẤT: 10/10 task của **R05 + R06** — branch `feature/teamhoa/r05-r06` (tạo từ `origin/feature/deltateam/refactor-plan`, có sẵn nền R01/R03/R04)
->
-> | EPIC | Task | Trạng thái |
-> | :--- | :--- | :--- |
-> | **R05** Tool, MCP & Connector Policy | T01 → T05 | ✅ 5/5 |
-> | **R06** Workspace, Filesystem & History Isolation | T01 → T05 | ✅ 5/5 |
->
-> **Kiểm chứng (chạy thật):**
-> * `pytest tests/` ➔ **283 pass / 4 fail** (+41 test mới cho R05+R06, gồm 2 test Qt offscreen thật trong `tests/integration/test_history_dir_race.py`)
-> * 4 fail là **lỗi có sẵn từ trước**, không liên quan R05/R06: 2 trong `test_config_security.py` (EPIC R02, đã ghi nhận bởi Team Duy) + 2 trong `test_routing_wiring.py` (môi trường máy này có Ollama/llama3.1 thật + config routing cục bộ khác "fresh install").
-> * `python scripts/check_imports.py` ➔ **PASS** (0 Qt import trong `domain/`, `application/`)
-> * Mọi file mới **< 400 dòng** (lớn nhất: `domain/tools/tool_registry.py` 125 dòng). `core/tools.py` giảm từ 566 ➔ 291 dòng.
->
-> ### 📄 BÁO CÁO CHI TIẾT
-> Xem `docs/refactor/BaoCao_TeamHoa_R05_R06.md` — kết quả từng EPIC, bằng chứng kiểm thử, 2 lỗi thật đã phát hiện (permission gate bị bỏ qua cho MCP tools, race condition lưu nhầm lịch sử), và phạm vi **chưa** kiểm thử.
->
-> ### 🔧 TÓM TẮT R06
-> * **R06-T01**: `domain/workspaces/workspace_session.py::WorkspaceSession` — snapshot bất biến (project_id, workspace_root, sandbox_dir, allowed_paths) + `is_allowed(path)`.
-> * **R06-T02**: `infrastructure/persistence/json/{workspace_repository_impl,conversation_repository_impl}.py` bọc `core/projects.py`/`core/history.py`. **Đã sửa bug thật**: `save_project`/`save_conversation`/`rename_conversation`/`set_pinned` trước đây `path.write_text()` không atomic (crash giữa lúc ghi = file JSON hỏng, `load_project`/`load_conversation` coi file hỏng như "không tồn tại" — mất project/hội thoại âm thầm). Giờ cả 4 hàm ghi qua `infrastructure/persistence/json/atomic_write.py::write_json` (temp file + `os.replace`). Có test giả lập crash giữa lúc ghi xác nhận file cũ không bị hỏng.
-> * **R06-T03**: `infrastructure/filesystem/execution_workspace.py::ExecutionWorkspace` — đặt tên cho quy ước `.scratch` đã có sẵn (không đổi vị trí file).
-> * **R06-T04**: Sửa race trong `ui/chat_panel.py` (không phải trực tiếp `_load_current`, xem "còn nợ" #2). `ChatPanel._persist_session` (lưu hội thoại của turn CHẠY NGẦM, không phải conversation đang xem) trước đây gọi `self.ctx.config.history_dir()` SỐNG tại thời điểm turn xong — nếu user đổi project khi turn còn chạy (`_load_current` ghi `config._project_history_dir`), turn nền lưu nhầm vào thư mục lịch sử của project MỚI. Fix: thêm `"home_history_dir"` vào dict `ctx` per-turn đã có sẵn (cùng quy ước với `home_id`/`home_messages`/`home_title`), chụp tại lúc submit. Test thật bằng Qt offscreen: `tests/integration/test_history_dir_race.py`.
-> * **R06-T05**: `application/workspaces/file_workspace_service.py::FileWorkspaceService` — cho File Explorer/AI Editor gọi `execute_tool` (list_dir/read_file/write_file/edit_file) giống agent, không tự viết lại logic.
->
-> ### 🔧 TÓM TẮT R05
-> * **R05-T01/T02**: `core/tools.py`'s if/elif dispatcher tách thành `infrastructure/filesystem/{file_tools,command_tools,fetch_tools,tool_context}.py` + `domain/tools/{tool_descriptor,tool_registry}.py`. `core/tools.py` còn lại là shim strangler-fig (re-export `ToolContext`/`ToolError`, dispatch qua dict).
-> * **R05-T03**: `application/conversations/tool_policy_gateway.py::ToolPolicyGateway` — thay `if gate is not None and name in ("run_command","install_package")` (chat_agent.py) và `if name in (WRITE_TOOLS|MS365_WRITE_TOOLS)` (code_agent.py) bằng một lookup capability chung. Đã verify bằng test: đúng 2 tool cũ vẫn được gate, không tool nào khác bị ảnh hưởng.
-> * **R05-T04 — ⚠️ THAY ĐỔI HÀNH VI CÓ CHỦ ĐÍCH**: trước đây MCP/connector/ext-connector tools (`core/mcp_client.py`, `core/ext_connectors.py`) chạy qua `extra_executor(name, args)` **không hề qua permission gate**. Giờ mọi `extra_tools` được gắn capability mặc định (`WRITE|EXECUTE|NETWORK`, vì MCP không có chuẩn khai báo rủi ro) và đi qua CÙNG `ToolPolicyGateway` như built-in tools. Khi Settings có "confirm before running commands" bật, tool MCP/connector giờ sẽ hỏi xác nhận — người dùng SẼ thấy thêm prompt so với trước. Test: `tests/unit/test_cowork_extra_tool_policy.py`.
-> * **R05-T05**: `infrastructure/mcp/mcp_source_manager.py::McpToolSourceManager` — tách lifecycle connection (cache/lock/start-or-skip) ra khỏi `state.py::AppContext` (trước đây inline trong `_mcp_connections`/`_conn_lock`). `AppContext` giờ chỉ gọi `self._mcp_manager.ensure/stop/stop_all`. `_ext_connections` (Connectors CAD/CAE/MS365/Other) KHÔNG thuộc phạm vi T05, vẫn giữ `_conn_lock` riêng như cũ.
->
-> ### 📌 CÒN NỢ / CẦN QUYẾT ĐỊNH
-> 1. **Xung đột file với EPIC R02 (Team Nam)**: R02-T01 giao `infrastructure/persistence/json/atomic_json_file.py` cho Team Nam. R06-T02 cần atomic write NGAY (bug thật, không chờ được) nên đã tạo `infrastructure/persistence/json/atomic_write.py` — tên khác, cùng thư mục, không đụng file của Team Nam. `core/projects.py`/`core/history.py` đang dùng module này trực tiếp. **Cần Team Nam xác nhận khi bắt đầu R02-T01**: nên hợp nhất `atomic_write.py` vào `atomic_json_file.py` (Team Hoa đổi 4 import) hay giữ 2 module riêng (rủi ro trôi giữa 2 cách ghi atomic).
-> 2. **`WorkspaceRepository`/`ConversationRepository`/`FileWorkspaceService` chưa có nơi gọi thật** — giống tình trạng `ProviderRegistry` của Team Duy ở R03. Mọi call site sản xuất (`ui/workspace_tab.py`, `ui/folder_tab.py`, `state.py`, task executors) vẫn dùng trực tiếp `core/projects.py`/`core/history.py`/`core/tools.py::execute_tool` — các class mới là seam cho tầng application ở EPIC sau (R07/R08), chưa nối dây.
-> 3. **R06-T04 phạm vi thực tế khác một chút so với mô tả gốc**: bug không nằm ở `ui/workspace_tab.py::_load_current` (hàm đó chỉ *set* `config._project_history_dir`, không tự đọc lại nó) mà ở `ui/chat_panel.py::_persist_session` — nơi một turn chạy ngầm đọc SỐNG giá trị đó lúc turn xong. Đã sửa đúng điểm đọc, có test Qt offscreen thật (`tests/integration/test_history_dir_race.py`), nhưng chưa đổi kiến trúc `_load_current` như plan gốc gợi ý (dùng session id thay biến toàn cục) — việc đó cần tách `ChatPanel`/`WorkspaceTab` sâu hơn, thuộc phạm vi R08 (UI/Application Separation).
-> 4. R05/R06 xong toàn bộ — Team Hoa chờ chỉ đạo cho **R07** (Scheduling & Workflow Runtime, phối hợp Team Nam) hoặc merge/review trước khi tiếp tục.
-
----
-
 ## 📌 PHẦN 1: CHECKLIST CHI TIẾT THEO 10 EPIC (R01 ➔ R10)
 
 ### 🔹 EPIC R01: Architecture Foundation & Characterization (Nền Tảng Kiến Trúc & Test Bảo Vệ)
@@ -104,15 +27,15 @@
 * **Mục tiêu**: Khóa DTO, dựng fakes/test doubles chạy offline không phụ thuộc Qt/mạng, thiết lập script chặn vi phạm kiến trúc.
 
 - [x] **R01-T01 (Team Duy)**: Viết Architecture ADR định rõ ranh giới các tầng ➔ `docs/architecture/ADR-001-layered-architecture.md`  
-  *Start: `2026-08-21 09:56` | End: `2026-08-21 10:00`*
+  *Start: `2026-08-21 18:23` | End: `2026-08-21 18:24`*
 - [x] **R01-T02 (Team Duy)**: Xây dựng `FakeProvider` và `FakeToolExecutor` chạy offline từ `providers/base.py` ➔ `tests/fakes/fake_provider.py` & `tests/fakes/fake_tool_executor.py`  
-  *Start: `2026-08-21 10:00` | End: `2026-08-21 10:02`*
+  *Start: `2026-08-21 18:24` | End: `2026-08-21 18:26`*
 - [x] **R01-T03 (Team Duy)**: Viết script quét tĩnh chặn code mới trong `domain/` và `application/` import `PySide6` ➔ `scripts/check_imports.py`  
-  *Start: `2026-08-21 09:58` | End: `2026-08-21 10:05`*
+  *Start: `2026-08-21 18:26` | End: `2026-08-21 18:28`*
 - [x] **R01-T04 (Team Duy)**: Viết Characterization Tests cho `core/chat_agent.py::run_cowork` ➔ `tests/characterization/test_run_cowork.py`  
-  *Start: `2026-08-21 10:02` | End: `2026-08-21 10:04`*
+  *Start: `2026-08-21 18:28` | End: `2026-08-21 18:32`*
 - [x] **R01-T05 (Team Duy)**: Lập danh mục và phân loại mã nguồn dormant/dead code ➔ `docs/architecture/dormant-code.md`  
-  *Start: `2026-08-21 10:04` | End: `2026-08-21 10:05`*
+  *Start: `2026-08-21 18:32` | End: `2026-08-21 18:35`*
 
 ---
 
@@ -139,18 +62,18 @@
 * **Team chịu trách nhiệm**: 🔵 **Team Duy** (Chủ trì)
 * **Mục tiêu**: Hợp nhất logic routing bị phân tán thành `RoutingApplicationService` độc lập Qt; chuẩn hóa catalog nhà cung cấp.
 
-- [x] **R03-T01 (Team Duy)**: Xây dựng bộ Contract Tests chuẩn hóa cho các Provider từ `providers/base.py` ➔ `tests/contracts/test_providers.py`  
-  *Start: `2026-08-21 10:10` | End: `2026-08-21 10:12`*
-- [x] **R03-T02 (Team Duy)**: Xây dựng `ProviderDescriptor` và `ProviderRegistry` tập trung từ `providers/factory.py` ➔ `domain/models/provider_descriptor.py` & `infrastructure/providers/provider_registry.py`  
-  *Start: `2026-08-21 10:06` | End: `2026-08-21 10:10`*
-- [x] **R03-T03 (Team Duy)**: Xây dựng `RoutingApplicationService` độc lập với Qt từ `core/routing/` ➔ `application/model_routing/routing_application_service.py`  
-  *Start: `2026-08-21 10:12` | End: `2026-08-21 10:15`*
-- [x] **R03-T04 (Team Duy)**: Di chuyển luồng gọi routing từ `ui/chat_panel.py#L638` sang `RoutingApplicationService`  
-  *Start: `2026-08-21 10:17` | End: `2026-08-21 10:20`*
-- [x] **R03-T05 (Team Duy)**: Di chuyển luồng gọi routing từ `ui/co4e_tab.py` và `ui/folder_tab.py` sang `RoutingApplicationService`  
-  *Start: `2026-08-21 10:20` | End: `2026-08-21 10:22`*
-- [x] **R03-T06 (Team Duy)**: Tách logic ghi nhận token usage ra khỏi Provider, chuyển thành `UsageEventSink` ➔ `infrastructure/telemetry/usage_sink.py`  
-  *Start: `2026-08-21 10:15` | End: `2026-08-21 10:17`*
+- [ ] **R03-T01 (Team Duy)**: Xây dựng bộ Contract Tests chuẩn hóa cho các Provider từ `providers/base.py` ➔ `tests/contracts/test_providers.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R03-T02 (Team Duy)**: Xây dựng `ProviderDescriptor` và `ProviderRegistry` tập trung từ `providers/factory.py` ➔ `domain/models/provider_descriptor.py` & `infrastructure/providers/provider_registry.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R03-T03 (Team Duy)**: Xây dựng `RoutingApplicationService` độc lập với Qt từ `core/routing/` ➔ `application/model_routing/routing_application_service.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R03-T04 (Team Duy)**: Di chuyển luồng gọi routing từ `ui/chat_panel.py#L638` sang `RoutingApplicationService`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R03-T05 (Team Duy)**: Di chuyển luồng gọi routing từ `ui/co4e_tab.py` và `ui/folder_tab.py` sang `RoutingApplicationService`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R03-T06 (Team Duy)**: Tách logic ghi nhận token usage ra khỏi Provider, chuyển thành `UsageEventSink` ➔ `infrastructure/telemetry/usage_sink.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
 
 ---
 
@@ -158,16 +81,16 @@
 * **Team chịu trách nhiệm**: 🔵 **Team Duy** (Chủ trì)
 * **Mục tiêu**: Đóng gói input turn chat thành `ConversationExecutionRequest` bất biến, điều phối vòng đời qua `ConversationApplicationService` và phát sinh sự kiện `AgentEvent` có định kiểu.
 
-- [x] **R04-T01 (Team Duy)**: Định nghĩa immutable dataclass `ConversationExecutionRequest` ➔ `domain/agents/conversation_execution_request.py`  
-  *Start: `2026-08-21 10:23` | End: `2026-08-21 10:25`*
-- [x] **R04-T02 (Team Duy)**: Chuẩn hóa các sự kiện `AgentEvent` (TextChunk, ToolCallStarted, ToolCallResult, Error) ➔ `domain/agents/agent_event.py`  
-  *Start: `2026-08-21 10:22` | End: `2026-08-21 10:23`*
-- [x] **R04-T03 (Team Duy)**: Xây dựng `ConversationApplicationService` điều phối thực thi từ `core/chat_agent.py` ➔ `application/conversations/conversation_application_service.py`  
-  *Start: `2026-08-21 10:25` | End: `2026-08-21 10:27`*
-- [x] **R04-T04 (Team Duy)**: Di chuyển `ui/cowork_tab.py::build_job` sang sử dụng `ConversationExecutionRequest`  
-  *Start: `2026-08-21 10:27` | End: `2026-08-21 10:31`*
-- [x] **R04-T05 (Team Duy)**: Di chuyển `core/task_executors.py` sang dùng chung `ConversationApplicationService`  
-  *Start: `2026-08-21 10:28` | End: `2026-08-21 10:30`*
+- [ ] **R04-T01 (Team Duy)**: Định nghĩa immutable dataclass `ConversationExecutionRequest` ➔ `domain/agents/conversation_execution_request.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R04-T02 (Team Duy)**: Chuẩn hóa các sự kiện `AgentEvent` (TextChunk, ToolCallStarted, ToolCallResult, Error) ➔ `domain/agents/agent_event.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R04-T03 (Team Duy)**: Xây dựng `ConversationApplicationService` điều phối thực thi từ `core/chat_agent.py` ➔ `application/conversations/conversation_application_service.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R04-T04 (Team Duy)**: Di chuyển `ui/cowork_tab.py::build_job` sang sử dụng `ConversationExecutionRequest`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R04-T05 (Team Duy)**: Di chuyển `core/task_executors.py` sang dùng chung `ConversationApplicationService`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
 
 ---
 
@@ -175,16 +98,16 @@
 * **Team chịu trách nhiệm**: 🟢 **Team Hoa** (Chủ trì) + Phối hợp Team Duy
 * **Mục tiêu**: Bóc tách monolithic `core/tools.py`, đưa toàn bộ Built-in tools, MCP tools và Connectors qua `ToolPolicyGateway` kiểm tra quyền phân tầng.
 
-- [x] **R05-T01 (Team Hoa)**: Định nghĩa `ToolDescriptor`, `ToolCapability` (read/write/execute/network) ➔ `domain/tools/tool_descriptor.py` & `domain/tools/tool_registry.py`  
-  *Start: `2026-08-21 21:40` | End: `2026-08-21 21:47`*
-- [x] **R05-T02 (Team Hoa)**: Tách nhỏ các built-in handlers từ `core/tools.py` ➔ `infrastructure/filesystem/file_tools.py`, `command_tools.py`, `fetch_tools.py`  
-  *Start: `2026-08-21 21:47` | End: `2026-08-21 21:56`*
-- [x] **R05-T03 (Team Hoa)**: Xây dựng `ToolPolicyGateway` (kiểm tra phân quyền allow/confirm/deny) ➔ `application/conversations/tool_policy_gateway.py`  
-  *Start: `2026-08-21 21:56` | End: `2026-08-21 22:04`*
-- [x] **R05-T04 (Team Hoa)**: Chuẩn hóa MCP tools từ `core/mcp_client.py` đi qua `ToolPolicyGateway`  
-  *Start: `2026-08-21 22:04` | End: `2026-08-21 22:12`*
-- [x] **R05-T05 (Team Hoa)**: Xây dựng `McpToolSourceManager` quản lý vòng đời tiến trình MCP ➔ `infrastructure/mcp/mcp_source_manager.py`  
-  *Start: `2026-08-21 22:12` | End: `2026-08-21 22:19`*
+- [ ] **R05-T01 (Team Hoa)**: Định nghĩa `ToolDescriptor`, `ToolCapability` (read/write/execute/network) ➔ `domain/tools/tool_descriptor.py` & `domain/tools/tool_registry.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R05-T02 (Team Hoa)**: Tách nhỏ các built-in handlers từ `core/tools.py` ➔ `infrastructure/filesystem/file_tools.py`, `command_tools.py`, `fetch_tools.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R05-T03 (Team Hoa)**: Xây dựng `ToolPolicyGateway` (kiểm tra phân quyền allow/confirm/deny) ➔ `application/conversations/tool_policy_gateway.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R05-T04 (Team Hoa)**: Chuẩn hóa MCP tools từ `core/mcp_client.py` đi qua `ToolPolicyGateway`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R05-T05 (Team Hoa)**: Xây dựng `McpToolSourceManager` quản lý vòng đời tiến trình MCP ➔ `infrastructure/mcp/mcp_source_manager.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
 
 ---
 
@@ -192,16 +115,16 @@
 * **Team chịu trách nhiệm**: 🟢 **Team Hoa** (Chủ trì)
 * **Mục tiêu**: Xóa bỏ biến toàn cục `state.py::active_project_id`, đóng gói workspace per-turn thành `WorkspaceSession` bất biến, bảo vệ an toàn đường dẫn sandbox.
 
-- [x] **R06-T01 (Team Hoa)**: Định nghĩa `WorkspaceSession` chứa snapshot project id, workspace root ➔ `domain/workspaces/workspace_session.py`  
-  *Start: `2026-08-21 22:19` | End: `2026-08-21 22:24`*
-- [x] **R06-T02 (Team Hoa)**: Xây dựng `WorkspaceRepository` từ `core/projects.py` & `ConversationRepository` từ `core/history.py` ➔ `infrastructure/persistence/json/workspace_repository_impl.py`  
-  *Start: `2026-08-21 22:24` | End: `2026-08-21 22:35`*
-- [x] **R06-T03 (Team Hoa)**: Xây dựng `ExecutionWorkspace` quản lý output/scratch files ➔ `infrastructure/filesystem/execution_workspace.py`  
-  *Start: `2026-08-21 22:35` | End: `2026-08-21 22:40`*
-- [x] **R06-T04 (Team Hoa)**: Khắc phục race condition trong `ui/workspace_tab.py::_load_current`  
-  *Start: `2026-08-21 22:40` | End: `2026-08-21 22:50`*
-- [x] **R06-T05 (Team Hoa)**: Xây dựng `FileWorkspaceService` xử lý thao tác file cho File Explorer và AI File Editor ➔ `application/workspaces/file_workspace_service.py`  
-  *Start: `2026-08-21 22:50` | End: `2026-08-21 22:57`*
+- [ ] **R06-T01 (Team Hoa)**: Định nghĩa `WorkspaceSession` chứa snapshot project id, workspace root ➔ `domain/workspaces/workspace_session.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R06-T02 (Team Hoa)**: Xây dựng `WorkspaceRepository` từ `core/projects.py` & `ConversationRepository` từ `core/history.py` ➔ `infrastructure/persistence/json/workspace_repository_impl.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R06-T03 (Team Hoa)**: Xây dựng `ExecutionWorkspace` quản lý output/scratch files ➔ `infrastructure/filesystem/execution_workspace.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R06-T04 (Team Hoa)**: Khắc phục race condition trong `ui/workspace_tab.py::_load_current`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
+- [ ] **R06-T05 (Team Hoa)**: Xây dựng `FileWorkspaceService` xử lý thao tác file cho File Explorer và AI File Editor ➔ `application/workspaces/file_workspace_service.py`  
+  *Start: `____-__-__ __:__` | End: `____-__-__ __:__`*
 
 ---
 
@@ -306,16 +229,16 @@
 
 | Ngày | Task Cần Hoàn Thành | Start Time | End Time | Trạng Thái |
 | :--- | :--- | :---: | :---: | :---: |
-| **21/08 (T6)** | Khóa DTO `ConversationExecutionRequest`, `AgentEvent`; Xây dựng `FakeProvider`, `FakeToolExecutor` | `2026-08-21 09:56` | `2026-08-21 10:25` | [x] |
-| **22-23/08 (T7-CN)** | Chuẩn hóa `ProviderDescriptor`, `ProviderRegistry`; Wrap OpenAI, Anthropic, Ollama, FPT Gateway; Viết Contract Tests | `2026-08-21 10:06` | `2026-08-21 10:12` | [x] ⚠️ registry chưa nối vào `state.build_provider_for` |
-| **24/08 (T2)** | Xây dựng `RoutingApplicationService` độc lập Qt; Tách `ComposerWidget` & `AttachmentPicker` | `2026-08-21 10:12` | `2026-08-21 10:15` | [~] RoutingApplicationService xong; tách widget thuộc R08 |
-| **25/08 (T3)** | Xây dựng `ConversationApplicationService`; Tách `ChatHistoryWidget` và bubble renderer | `2026-08-21 10:25` | `2026-08-21 10:27` | [~] Service xong; tách widget thuộc R08 |
+| **21/08 (T6)** | Khóa DTO `ConversationExecutionRequest`, `AgentEvent`; Xây dựng `FakeProvider`, `FakeToolExecutor` | `2026-08-21 18:23` | `2026-08-21 18:35` | [x] |
+| **22-23/08 (T7-CN)** | Chuẩn hóa `ProviderDescriptor`, `ProviderRegistry`; Wrap OpenAI, Anthropic, Ollama, FPT Gateway; Viết Contract Tests | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
+| **24/08 (T2)** | Xây dựng `RoutingApplicationService` độc lập Qt; Tách `ComposerWidget` & `AttachmentPicker` | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
+| **25/08 (T3)** | Xây dựng `ConversationApplicationService`; Tách `ChatHistoryWidget` và bubble renderer | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
 | **26/08 (T4)** | Nối stream `AgentEvent` sang Chat History; Tách `AudioRecorderWidget` | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
 | **27/08 (T5)** | Tách `ChatOutputPanel` & File Watcher; Lắp ráp container `ChatPanel` và `Floating HelpAgent` | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
-| **28/08 (T6)** | Xóa copy routing cũ trong `ui/chat_panel.py`; Fix circular import `model_pricing` ↔ `usage_tracker` | `2026-08-21 10:17` | `2026-08-21 10:22` | [~] 3 bản copy routing đã gỡ; circular import chưa xử lý |
-| **29/08 (T7)** | Viết suite integration test cho toàn bộ luồng Chat (`tests/integration/test_chat_flow.py`) | `2026-08-21 10:35` | `2026-08-21 10:52` | [~] 25 integration test tại `tests/integration/{test_cowork_turn_flow,test_task_executor_flow,test_routing_surfaces}.py` |
-| **30/08 (CN)** | 🔍 **Chủ trì CASAN Check 3**: Chạy `python scripts/check_imports.py` đảm bảo 0 import `PySide6` trong domain & application | `2026-08-21 09:58` | `2026-08-21 10:05` | [x] PASS |
-| **31/08 (T2)** | **Chủ trì EPIC R10**: Viết Contributor Recipes, chạy E2E Smoke Test (`tests/e2e/test_smoke.py`) và merge PR cuối cùng | `____-__-__ __:__` | `____-__-__ __:__` | [ ] chờ 3 team hoàn tất |
+| **28/08 (T6)** | Xóa copy routing cũ trong `ui/chat_panel.py`; Fix circular import `model_pricing` ↔ `usage_tracker` | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
+| **29/08 (T7)** | Viết suite integration test cho toàn bộ luồng Chat (`tests/integration/test_chat_flow.py`) | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
+| **30/08 (CN)** | 🔍 **Chủ trì CASAN Check 3**: Chạy `python scripts/check_imports.py` đảm bảo 0 import `PySide6` trong domain & application | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
+| **31/08 (T2)** | **Chủ trì EPIC R10**: Viết Contributor Recipes, chạy E2E Smoke Test (`tests/e2e/test_smoke.py`) và merge PR cuối cùng | `____-__-__ __:__` | `____-__-__ __:__` | [ ] |
 
 ---
 

docs/refactor/bug.md

@@ -0,0 +1,155 @@
+# NHẬT KÝ THEO DÕI VÀ PHÒNG NGỪA LỖI TÁI CẤU TRÚC (BUG & LESSONS LEARNED LOG)
+## DỰ ÁN: COWORK LOCAL (COWORK-LOCAL BAMBOO)
+
+Tài liệu này dùng để ghi nhận **toàn bộ các lỗi, xung đột kiến trúc và sự cố phát sinh** trong suốt quá trình refactoring của cả 3 team (Team Duy, Team Nam, Team Hoa).
+
+> [!IMPORTANT]
+> ### 🛡️ NGUYÊN TẮC VÀNG VỀ QUẢN TRỊ CHẤT LƯỢNG (ZERO RECURRENCE):
+> 1. **Ghi nhận ngay lập tức**: Khi gặp bất kỳ lỗi nào (Syntax, Circular Import, Type Error, Test Failure, Thread Freeze, Data Corruption), kỹ sư/AI phải ghi ngay vào tài liệu này trước khi tiếp tục task.
+> 2. **Phân tích nguyên nhân gốc rễ (Root Cause)**: Không chỉ sửa phần ngọn mà phải giải thích rõ bản chất vì sao lỗi xảy ra.
+> 3. **Rút ra quy tắc phòng ngừa (Prevention Rule)**: Đặt ra nguyên tắc kỹ thuật để **TUYỆT ĐỐI KHÔNG TÁI PHẠM** ở các task tiếp theo.
+> 4. **Checklist đầu vào**: Trước khi bắt đầu bất kỳ task mới nào, kỹ sư/AI **bắt buộc phải đọc lại toàn bộ file này**.
+
+---
+
+## 📌 BẢNG TỔNG HỢP CÁC LỖI ĐÃ PHÁT HIỆN & KHẮC PHỤC
+
+| Bug ID | Ngày Phát Hiện | Phân Hệ / File Bị Ảnh Hưởng | Loại Lỗi | Trạng Thái | Team Phụ Trách |
+| :--- | :---: | :--- | :--- | :---: | :---: |
+| `BUG-001` | 2026-08-20 | `core/model_pricing.py` ↔ `core/usage_tracker.py` | Circular Dependency | 🟡 Đã có giải pháp (R09) | Team Duy & Team Nam |
+| `BUG-002` | 2026-08-20 | `core/agent_security.py` ↔ `core/agent_security_alert.py` | Circular Dependency | 🟡 Đã có giải pháp (R09) | Team Nam |
+| `BUG-003` | 2026-08-20 | `state.py::active_project_id` & `ui/workspace_tab.py` | Race Condition / Global State Leak | 🟡 Đã có giải pháp (R06) | Team Hoa |
+| `BUG-004` | 2026-08-20 | `core/task_scheduler.py` ↔ `PySide6.QtCore.QTimer` | Architecture Violation (Qt in Domain/App) | 🟡 Đã có giải pháp (R07) | Team Hoa |
+| `BUG-005` | 2026-08-20 | `ui/chat_panel.py#L638`, `ui/co4e_tab.py`, `ui/folder_tab.py` | Code Duplication (Copy Routing Logic) | 🟡 Đã có giải pháp (R03) | Team Duy |
+| `BUG-006` | 2026-08-21 | `scripts/check_imports.py` | UnicodeEncodeError (Windows CP932 console emoji) | 🟢 Đã khắc phục (R01) | Team Duy |
+| `BUG-007` | 2026-08-21 | `platform/` ➔ `infrastructure/platform/` | Standard Library Shadowing (`import platform`) | 🟢 Đã khắc phục (R01) | Team Duy |
+
+---
+
+## 🔍 CHI TIẾT TỪNG LỖI & QUY TẮC PHÒNG NGỪA
+
+---
+
+### 🔴 `BUG-001`: Circular Import giữa Module Định Giá (`model_pricing.py`) và Theo Dõi Token (`usage_tracker.py`)
+
+* **Phân hệ**: `core/model_pricing.py` & `core/usage_tracker.py`
+* **Triệu chứng (Symptom)**: Lỗi `ImportError: cannot import name 'ModelPricing' from partially initialized module` khi khởi động ứng dụng hoặc chạy test độc lập.
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - `model_pricing.py` import `UsageTracker` để cập nhật dữ liệu tiêu thụ.
+  - Ngược lại, `usage_tracker.py` import `ModelPricing` để tính toán chi phí theo từng model ID.
+* **Giải pháp khắc phục (Resolution)**:
+  - Tách Data Transfer Object (DTO) `ModelPricing` sang tầng Domain thuần túy `domain/models/model_pricing.py`.
+  - Cả `model_pricing.py` và `usage_tracker.py` đều import DTO từ `domain/models/`, chuyển quan hệ thành 1 chiều (Dependency Inversion).
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Không bao giờ để 2 service hoặc 2 module nghiệp vụ import lẫn nhau. Mọi cấu trúc dữ liệu dùng chung (DTO/Value Object/Event) **phải được đặt tại tầng `domain/`**.
+
+---
+
+### 🔴 `BUG-002`: Circular Import giữa An Ninh Agent (`agent_security.py`) và Cảnh Báo (`agent_security_alert.py`)
+
+* **Phân hệ**: `core/agent_security.py` & `core/agent_security_alert.py`
+* **Triệu chứng (Symptom)**: Lỗi khởi tạo vòng tròn khi runtime bắn ra alert sự kiện bảo mật.
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Module security vừa kiểm tra policy vừa khởi tạo trực tiếp instance alert dialog, trong khi alert dialog lại import ngược lại rule security để hiển thị chi tiết mã lỗi.
+* **Giải pháp khắc phục (Resolution)**:
+  - Tách sự kiện cảnh báo thành Event DTO `SecurityAlertEvent` tại `domain/security/security_event.py`.
+  - Tầng Security chỉ phát ra Event (`emit_event`), tầng Presentation/UI tự lắng nghe Event để render Dialog.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Logic an ninh và xử lý nghiệp vụ không bao giờ được gọi trực tiếp UI Dialog. Luôn giao tiếp thông qua cơ chế Event-Driven (`AgentEvent`, `SecurityEvent`).
+
+---
+
+### 🔴 `BUG-003`: Xung Đột Race Condition do Sử Dụng Biến Toàn Cục `active_project_id` trong `state.py`
+
+* **Phân hệ**: `state.py`, `ui/workspace_tab.py`, Scheduled Task Runners
+* **Triệu chứng (Symptom)**: Khi task scheduler chạy ngầm hoặc người dùng chuyển tab nhanh, file bị ghi nhầm vào thư mục dự án khác với dự án đang hiển thị trên màn hình.
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Ứng dụng đọc và ghi trực tiếp vào biến toàn cục `AppContext.active_project_id` từ nhiều luồng khác nhau mà không có cơ chế snapshot ngữ cảnh.
+* **Giải pháp khắc phục (Resolution)**:
+  - Xóa bỏ việc đọc biến toàn cục. Mỗi lần khởi chạy turn hoặc task, tạo một snapshot bất biến `WorkspaceSession(project_id, root_path, allowed_paths)`.
+  - Luồng ngầm chỉ thao tác trên `WorkspaceSession` được truyền vào từ lúc khởi tạo.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Tuyệt đối không dùng biến toàn cục (Global State / Singletons có trạng thái thay đổi) để điều khiển luồng thực thi nền. Mọi ngữ cảnh phải được truyền tường minh qua DTO snapshot.
+
+---
+
+### 🔴 `BUG-004`: Vi Phạm Ranh Giới Kiến Trúc Khi Import `PySide6.QtCore.QTimer` trong Domain / Scheduling Engine
+
+* **Phân hệ**: `core/task_scheduler.py#L20`
+* **Triệu chứng (Symptom)**: Không thể viết Unit Test cho thuật toán tính toán lịch chạy (cron/interval) trên môi trường CI/CD (GitHub Actions / Linux Server headless) nếu thiếu driver màn hình X11/Wayland hoặc chưa cài `PySide6`.
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Động cơ lập lịch bị gắn chặt cứng với `QTimer` của framework Qt thay vì tách riêng logic tính toán thời gian.
+* **Giải pháp khắc phục (Resolution)**:
+  - Tách thuật toán tính lịch sang `domain/tasks/schedule_calculator.py` (Pure Python 100%).
+  - Tạo `platform/qt/qt_scheduler_clock.py` làm adapter bọc `QTimer` cho app chạy thật, và `tests/fakes/fake_clock.py` cho unit test.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Tầng Domain và Application tuyệt đối không import thư viện GUI (`PySide6`, `PyQt`). Luôn bọc các thành phần phụ thuộc framework bên ngoài qua Adapter Interface.
+
+---
+
+### 🔴 `BUG-005`: Nhân Bản Mã Nguồn (Code Duplication) Logic Routing Mô Hình AI tại Nhiều Màn Hình
+
+* **Phân hệ**: `ui/chat_panel.py#L638`, `ui/co4e_tab.py`, `ui/folder_tab.py`
+* **Triệu chứng (Symptom)**: Khi cập nhật thêm model provider mới (như FPT Gateway hay Claude 3.7), phải sửa code thủ công ở 3 file UI khác nhau; phát sinh sai lệch quy tắc fallback giữa các màn hình.
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Thiếu một tầng Application Service tập trung, dẫn đến việc lập trình viên copy-paste hàm chọn model từ `ChatPanel` sang các tab khác.
+* **Giải pháp khắc phục (Resolution)**:
+  - Xây dựng `application/model_routing/routing_application_service.py` duy nhất, cung cấp API `route_request(request) -> ModelRouteDecision`.
+  - Mọi màn hình UI chỉ gọi service này, không tự viết lại logic kiểm tra key hay fallback.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Nghiệp vụ dùng chung giữa các màn hình phải được đưa vào `application/` services. Không bao giờ viết logic nghiệp vụ trực tiếp trong các file Widget UI.
+
+---
+
+### 🟢 `BUG-006`: `UnicodeEncodeError` khi in Emojis trên Console Windows (CP932/CP1252)
+
+* **Phân hệ / File**: `scripts/check_imports.py`
+* **Triệu chứng (Symptom)**:
+  ```text
+  Traceback (most recent call last):
+    File "scripts/check_imports.py", line 127, in main
+      print(f"\U0001f6e1\ufe0f  Running Clean Architecture Import Guard...")
+  UnicodeEncodeError: 'cp932' codec can't encode character '\U0001f6e1' in position 0: illegal multibyte sequence
+  ```
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Trên hệ điều hành Windows sử dụng locale tiếng Nhật (mã trang CP932) hoặc tiếng Anh (CP1252), `sys.stdout` mặc định không hỗ trợ các ký tự Unicode/Emoji ngoài bảng mã, dẫn đến crash khi in log dòng lệnh.
+* **Giải pháp khắc phục (Resolution)**:
+  - Tự động bọc lại `sys.stdout` và `sys.stderr` bằng `io.TextIOWrapper` với `encoding="utf-8"` và `errors="replace"`.
+  - Thay thế các emoji phức tạp bằng các tag văn bản ASCII chuẩn hóa như `[Clean Arch Guard]`, `[PASS]`, `[FAIL]`.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Mọi script CLI (`scripts/*.py`) phải có cơ chế cấu hình `utf-8` stream wrapper và ưu tiên sử dụng text tags (`[INFO]`, `[WARN]`, `[ERROR]`) thay vì emoji Unicode trực tiếp để đảm bảo chạy mượt mà trên mọi môi trường Windows đa ngôn ngữ.
+
+---
+
+### 🟢 `BUG-007`: Xung Đột Tên Thư Mục Trùng Với Standard Library (`platform/` Shadowing `import platform`)
+
+* **Phân hệ / File**: `platform/` ➔ Chuyển thành `infrastructure/platform/`
+* **Triệu chứng (Symptom)**:
+  ```text
+  INTERNALERROR> File "_pytest/terminal.py", line 853: verinfo = platform.python_version()
+  INTERNALERROR> AttributeError: module 'platform' has no attribute 'python_version'
+  ```
+* **Nguyên nhân gốc rễ (Root Cause)**:
+  - Khi tạo một package ở thư mục gốc có tên trùng với module thư viện chuẩn của Python (`platform`, `email`, `test`, `asyncio`, `logging`), Python trên `sys.path` sẽ ưu tiên import thư mục local thay vì thư viện chuẩn của Python runtime, dẫn đến crash toàn bộ pytest runner và các thư viện bên thứ ba.
+* **Giải pháp khắc phục (Resolution)**:
+  - Xóa bỏ package `platform/` ở root.
+  - Đưa adapter Qt Scheduler Clock vào đúng vị trí hạ tầng: `infrastructure/platform/qt/`.
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: Tuyệt đối không đặt tên package/thư mục ở root trùng với tên các module built-in của Python (`platform`, `logging`, `types`, `time`, `io`, `os`, `sys`). Mọi platform adapter phải nằm trong `infrastructure/platform/` hoặc `platform_adapters/`.
+
+---
+
+## 📝 MẪU GHI NHẬN BUG MỚI (BUG REPORT TEMPLATE)
+
+Khi gặp bất kỳ bug mới nào trong quá trình làm việc, hãy sao chép khối mẫu sau và điền vào cuối tài liệu:
+
+```markdown
+### 🔴 `BUG-XXX`: [Tóm tắt ngắn gọn tên lỗi]
+
+* **Phân hệ / File**: `[Đường dẫn file bị lỗi]`
+* **Triệu chứng (Symptom)**: `[Mô tả hiện tượng lỗi, paste thông báo traceback hoặc kết quả test fail]`
+* **Nguyên nhân gốc rễ (Root Cause)**: `[Giải thích tại sao lỗi lại xảy ra]`
+* **Giải pháp khắc phục (Resolution)**: `[Mô tả cách sửa, file DTO/Service tạo mới hoặc cách refactor]`
+* **Quy tắc phòng ngừa (Prevention Rule - TUYỆT ĐỐI KHÔNG TÁI PHẠM)**:
+  > **Quy tắc**: `[Nguyên tắc kỹ thuật cụ thể để không bao giờ tái phạm lỗi này]`
+```

domain/__init__.py

@@ -1,12 +1 @@
-"""Domain layer - pure Python entities, value objects and events.
-
-The innermost layer of the 4-tier architecture (see
-``docs/architecture/ADR-001-layered-architecture.md``). Modules here describe
-WHAT the application is about - a turn of conversation, a model candidate, an
-agent event - and depend on nothing but the standard library.
-
-Hard rule (ADR-001 I1/I2, enforced by ``scripts/check_imports.py``): no imports
-of PySide6/PyQt, and no imports from ``application/``, ``infrastructure/``,
-``presentation/`` or the legacy ``core/``/``ui/`` packages. That is what keeps
-this layer testable in milliseconds and reusable from a headless scheduler.
-"""
+"""Domain Layer: Pure Python domain entities, value objects, and events."""

domain/agents/__init__.py

@@ -1,48 +1 @@
-"""Domain entities for one agent turn: the request snapshot and the typed event
-stream it produces (EPIC R04)."""
-
-from .agent_event import (
-    AgentEvent,
-    AssistantDoneEvent,
-    ErrorEvent,
-    HistoryReadyEvent,
-    NoticeEvent,
-    OutputsAddedEvent,
-    OutputsRemovedEvent,
-    PlanUpdatedEvent,
-    ReasoningChunkEvent,
-    TextChunkEvent,
-    ToolCallFinishedEvent,
-    ToolCallStartedEvent,
-    ToolOutputEvent,
-    TurnCompletedEvent,
-    collect_text,
-    event_from_dict,
-    tool_calls,
-)
-from .conversation_execution_request import (
-    ConversationExecutionRequest,
-    new_turn_id,
-)
-
-__all__ = [
-    "ConversationExecutionRequest",
-    "new_turn_id",
-    "AgentEvent",
-    "TextChunkEvent",
-    "ReasoningChunkEvent",
-    "AssistantDoneEvent",
-    "PlanUpdatedEvent",
-    "ToolCallStartedEvent",
-    "ToolOutputEvent",
-    "ToolCallFinishedEvent",
-    "OutputsAddedEvent",
-    "OutputsRemovedEvent",
-    "NoticeEvent",
-    "HistoryReadyEvent",
-    "TurnCompletedEvent",
-    "ErrorEvent",
-    "event_from_dict",
-    "collect_text",
-    "tool_calls",
-]
+"""Domain agents package: turn requests, agent events, and role definitions."""

domain/agents/agent_event.py

@@ -1,370 +0,0 @@
-"""AgentEvent - the typed event stream one agent turn produces (R04-T02).
-
-Today the turn engine talks to its caller through untyped dicts::
-
-    emit({"type": "tool_result", "id": tc_id, "name": name,
-          "ok": result.get("ok", False), "output": result.get("output", "")})
-
-and every consumer re-discovers the vocabulary by reading the producer. There
-are eleven such shapes across ``core/chat_agent.py``, ``core/code_agent.py`` and
-``core/task_executors.py``; a consumer that misspells ``"tool_result"`` or reads
-``"result"`` instead of ``"output"`` fails silently, at runtime, only for the
-tool path that triggers it.
-
-This module makes the vocabulary explicit. Each event is a frozen dataclass, so:
-
-* the set of possible events is enumerable (see :data:`EVENT_TYPES`);
-* a field name typo is an ``AttributeError`` at the point of use, not a silently
-  missing chat bubble;
-* an event can cross a thread boundary safely - it cannot be mutated after the
-  producer hands it over, which is exactly what the Qt-signal seam needs.
-
-Bridging with the legacy dicts is deliberate and two-way: :func:`event_from_dict`
-adapts what ``run_cowork`` emits today, and :meth:`AgentEvent.to_dict` renders an
-event back into the legacy shape so existing widgets keep working untouched
-while the presentation layer migrates screen by screen (EPIC R08).
-
-Pure domain code: stdlib only, no Qt, no I/O.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass, field
-from typing import Any, Dict, List, Mapping, Optional, Sequence, Tuple
-
-
-@dataclass(frozen=True)
-class AgentEvent:
-    """Base class for everything a turn can report.
-
-    ``type`` is the legacy string tag, kept as a class attribute so the bridge
-    functions can round-trip an event without a separate mapping table.
-    """
-
-    type: str = field(init=False, default="event")
-
-    def to_dict(self) -> Dict[str, Any]:
-        """Render into the legacy ``emit()`` dict shape."""
-        return {"type": self.type}
-
-
-# --------------------------------------------------------------------------- #
-# Assistant output
-# --------------------------------------------------------------------------- #
-@dataclass(frozen=True)
-class TextChunkEvent(AgentEvent):
-    """One fragment of the visible answer, as it streams in."""
-
-    delta: str
-    type: str = field(init=False, default="text")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "delta": self.delta}
-
-
-@dataclass(frozen=True)
-class ReasoningChunkEvent(AgentEvent):
-    """One fragment of the model's PRIVATE reasoning.
-
-    Drives the "Thinking" indicator only. Consumers must never append this to
-    the answer or persist it into conversation history - keeping it a distinct
-    type is what makes that mistake hard to make by accident.
-    """
-
-    delta: str
-    type: str = field(init=False, default="reasoning")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "delta": self.delta}
-
-
-@dataclass(frozen=True)
-class AssistantDoneEvent(AgentEvent):
-    """One assistant message finished. A turn with tool calls emits this once
-    per step, not once per turn - see :class:`TurnCompletedEvent`."""
-
-    content: str = ""
-    type: str = field(init=False, default="assistant_done")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "content": self.content}
-
-
-# --------------------------------------------------------------------------- #
-# Planning
-# --------------------------------------------------------------------------- #
-@dataclass(frozen=True)
-class PlanUpdatedEvent(AgentEvent):
-    """The agent rewrote its plan (the ``update_plan`` tool)."""
-
-    steps: Tuple[Dict[str, Any], ...] = ()
-    type: str = field(init=False, default="plan_set")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "steps": [dict(s) for s in self.steps]}
-
-
-# --------------------------------------------------------------------------- #
-# Tool lifecycle
-# --------------------------------------------------------------------------- #
-@dataclass(frozen=True)
-class ToolCallStartedEvent(AgentEvent):
-    """A tool call is about to run, with the preview shown to the user.
-
-    Maps the legacy ``tool_proposed`` event. "Proposed" was a misnomer: by the
-    time it is emitted the call is already going to run unless a permission gate
-    rejects it, and the gate reports that as a finished call with ``ok=False``.
-    """
-
-    call_id: str
-    name: str
-    args: Dict[str, Any] = field(default_factory=dict)
-    preview: Optional[Dict[str, Any]] = None
-    type: str = field(init=False, default="tool_proposed")
-
-    def to_dict(self) -> Dict[str, Any]:
-        out: Dict[str, Any] = {"type": self.type, "id": self.call_id,
-                               "name": self.name, "args": dict(self.args)}
-        if self.preview is not None:
-            out["preview"] = dict(self.preview)
-        return out
-
-
-@dataclass(frozen=True)
-class ToolOutputEvent(AgentEvent):
-    """A line of live output from a running tool (command stdout, for example)."""
-
-    call_id: str
-    name: str
-    delta: str
-    type: str = field(init=False, default="tool_output")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "id": self.call_id, "name": self.name,
-                "delta": self.delta}
-
-
-@dataclass(frozen=True)
-class ToolCallFinishedEvent(AgentEvent):
-    """A tool call ended, successfully or not.
-
-    ``ok=False`` covers every failure mode alike - the tool raised, the sandbox
-    blocked it, or the user rejected it at the permission gate - because the
-    consumer's job is the same in all three: show the failure and let the model
-    react to it.
-    """
-
-    call_id: str
-    name: str
-    ok: bool = False
-    output: str = ""
-    path: str = ""            # file the tool wrote, when it wrote one
-    produced: Tuple[str, ...] = ()   # extra artefacts (e.g. a generator's outputs)
-    type: str = field(init=False, default="tool_result")
-
-    def to_dict(self) -> Dict[str, Any]:
-        out: Dict[str, Any] = {"type": self.type, "id": self.call_id, "name": self.name,
-                               "ok": self.ok, "output": self.output}
-        if self.path:
-            out["path"] = self.path
-        if self.produced:
-            out["produced"] = list(self.produced)
-        return out
-
-
-# --------------------------------------------------------------------------- #
-# Output folder
-# --------------------------------------------------------------------------- #
-@dataclass(frozen=True)
-class OutputsAddedEvent(AgentEvent):
-    """Files appeared in the turn's output folder."""
-
-    paths: Tuple[str, ...] = ()
-    type: str = field(init=False, default="outputs_added")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "paths": list(self.paths)}
-
-
-@dataclass(frozen=True)
-class OutputsRemovedEvent(AgentEvent):
-    """Files were cleaned up from the turn's output folder (intermediates)."""
-
-    paths: Tuple[str, ...] = ()
-    type: str = field(init=False, default="outputs_removed")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "paths": list(self.paths)}
-
-
-@dataclass(frozen=True)
-class NoticeEvent(AgentEvent):
-    """A UI-visible aside that is not part of the model's answer.
-
-    Three producers today, all reachable from a normal turn:
-    ``core/agent_security.py`` (a request or command blocked by the security
-    layer), ``core/context_budget.py`` (the conversation was auto-compressed)
-    and the attachment readers (a file that could not be processed, plus live
-    "reading page X/Y" progress).
-
-    ``level`` selects how the UI renders it: ``"progress"`` updates the thinking
-    indicator in place, anything else becomes a warning bubble. Dropping these
-    would silently hide security warnings from the user, which is why the type
-    exists rather than being folded into TextChunkEvent.
-    """
-
-    text: str
-    level: str = "info"
-    type: str = field(init=False, default="notice")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "level": self.level, "text": self.text}
-
-
-@dataclass(frozen=True)
-class HistoryReadyEvent(AgentEvent):
-    """A history session exists for this run and can be opened."""
-
-    session_id: str
-    type: str = field(init=False, default="history_ready")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "session_id": self.session_id}
-
-
-# --------------------------------------------------------------------------- #
-# Turn lifecycle - emitted by the application service, not by the legacy engine
-# --------------------------------------------------------------------------- #
-@dataclass(frozen=True)
-class TurnCompletedEvent(AgentEvent):
-    """The whole turn finished: no more events will follow.
-
-    New in R04. The legacy engine has no end-of-turn signal at all, so every
-    consumer infers "done" from the worker thread finishing - which is why a
-    cancelled turn and a failed turn look identical to the UI today.
-    """
-
-    content: str = ""
-    cancelled: bool = False
-    type: str = field(init=False, default="turn_completed")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "content": self.content, "cancelled": self.cancelled}
-
-
-@dataclass(frozen=True)
-class ErrorEvent(AgentEvent):
-    """The turn failed. ``recoverable`` marks errors the user can act on
-    (pick another model, shorten the prompt) rather than a hard outage."""
-
-    message: str
-    recoverable: bool = False
-    type: str = field(init=False, default="error")
-
-    def to_dict(self) -> Dict[str, Any]:
-        return {"type": self.type, "message": self.message,
-                "recoverable": self.recoverable}
-
-
-# The legacy tag -> event class map. Also the authoritative list of what a turn
-# can emit, which is what makes an exhaustive consumer possible for the first time.
-EVENT_TYPES: Dict[str, type] = {
-    "text": TextChunkEvent,
-    "reasoning": ReasoningChunkEvent,
-    "assistant_done": AssistantDoneEvent,
-    "plan_set": PlanUpdatedEvent,
-    "tool_proposed": ToolCallStartedEvent,
-    "tool_start": ToolCallStartedEvent,
-    "tool_output": ToolOutputEvent,
-    "tool_result": ToolCallFinishedEvent,
-    "outputs_added": OutputsAddedEvent,
-    "outputs_removed": OutputsRemovedEvent,
-    "notice": NoticeEvent,
-    "history_ready": HistoryReadyEvent,
-    "turn_completed": TurnCompletedEvent,
-    "error": ErrorEvent,
-}
-
-
-def event_from_dict(payload: Mapping[str, Any]) -> Optional[AgentEvent]:
-    """Adapt one legacy ``emit()`` dict into a typed event.
-
-    Returns ``None`` for an unknown tag instead of raising: the legacy engine is
-    still being refactored and may grow an event before this module knows about
-    it. Dropping an unrecognised event degrades the UI by one missing bubble;
-    raising here would abort a turn that had otherwise succeeded.
-    """
-    kind = str(payload.get("type", ""))
-    cls = EVENT_TYPES.get(kind)
-    if cls is None:
-        return None
-
-    if cls is TextChunkEvent or cls is ReasoningChunkEvent:
-        return cls(delta=str(payload.get("delta", "")))
-    if cls is AssistantDoneEvent:
-        return AssistantDoneEvent(content=str(payload.get("content", "")))
-    if cls is PlanUpdatedEvent:
-        return PlanUpdatedEvent(steps=tuple(payload.get("steps") or ()))
-    if cls is ToolCallStartedEvent:
-        return ToolCallStartedEvent(
-            call_id=str(payload.get("id", "")), name=str(payload.get("name", "")),
-            args=dict(payload.get("args") or {}), preview=payload.get("preview"),
-        )
-    if cls is ToolOutputEvent:
-        return ToolOutputEvent(call_id=str(payload.get("id", "")),
-                               name=str(payload.get("name", "")),
-                               delta=str(payload.get("delta", "")))
-    if cls is ToolCallFinishedEvent:
-        return ToolCallFinishedEvent(
-            call_id=str(payload.get("id", "")), name=str(payload.get("name", "")),
-            ok=bool(payload.get("ok", False)), output=str(payload.get("output", "")),
-            path=str(payload.get("path", "") or ""),
-            produced=tuple(payload.get("produced") or ()),
-        )
-    if cls is OutputsAddedEvent or cls is OutputsRemovedEvent:
-        return cls(paths=tuple(str(p) for p in (payload.get("paths") or ())))
-    if cls is NoticeEvent:
-        return NoticeEvent(text=str(payload.get("text", "")),
-                           level=str(payload.get("level", "info")))
-    if cls is HistoryReadyEvent:
-        return HistoryReadyEvent(session_id=str(payload.get("session_id", "")))
-    if cls is TurnCompletedEvent:
-        return TurnCompletedEvent(content=str(payload.get("content", "")),
-                                  cancelled=bool(payload.get("cancelled", False)))
-    return ErrorEvent(message=str(payload.get("message", "")),
-                      recoverable=bool(payload.get("recoverable", False)))
-
-
-def collect_text(events: Sequence[AgentEvent]) -> str:
-    """Join every :class:`TextChunkEvent` - the visible answer, reasoning excluded.
-
-    Provided here so no consumer has to re-derive "which events are the answer",
-    the question the untyped dicts made easy to get wrong.
-    """
-    return "".join(e.delta for e in events if isinstance(e, TextChunkEvent))
-
-
-def tool_calls(events: Sequence[AgentEvent]) -> List[ToolCallFinishedEvent]:
-    """Every finished tool call, in order - for audit views and assertions."""
-    return [e for e in events if isinstance(e, ToolCallFinishedEvent)]
-
-
-__all__ = [
-    "AgentEvent",
-    "TextChunkEvent",
-    "ReasoningChunkEvent",
-    "AssistantDoneEvent",
-    "PlanUpdatedEvent",
-    "ToolCallStartedEvent",
-    "ToolOutputEvent",
-    "ToolCallFinishedEvent",
-    "OutputsAddedEvent",
-    "OutputsRemovedEvent",
-    "NoticeEvent",
-    "HistoryReadyEvent",
-    "TurnCompletedEvent",
-    "ErrorEvent",
-    "EVENT_TYPES",
-    "event_from_dict",
-    "collect_text",
-    "tool_calls",
-]

domain/agents/conversation_execution_request.py

@@ -1,192 +0,0 @@
-"""ConversationExecutionRequest - an immutable snapshot of one turn (R04-T01).
-
-``ui/cowork_tab.py::build_job`` currently builds a closure that reads widget
-state from inside the worker thread::
-
-    def job(worker):
-        provider = self.build_provider()      # reads combo boxes
-        extra_tools, extra_exec = self.ctx.build_mcp_tools()
-        proj_ctx = project_context_text(load_project(project_id))
-        ...
-
-Everything that closure touches can change while the turn is running: the user
-can pick another model, switch workspace, or edit the project instructions. The
-turn then runs on a mixture of old and new state, and which mixture depends on
-thread timing - the class of bug that reproduces once a week and never in a test.
-
-This value object is the fix: the presentation layer captures everything a turn
-needs ON THE UI THREAD, at submit time, into one frozen object. Whatever happens
-to the widgets afterwards, the turn keeps running on the state the user actually
-submitted.
-
-Pure domain code: stdlib only, no Qt, no filesystem access. Paths are held as
-strings, not ``Path`` objects, so the snapshot stays trivially serialisable -
-which is what will let a turn be queued, replayed or logged later.
-"""
-from __future__ import annotations
-
-import uuid
-from dataclasses import dataclass, field, replace
-from typing import Any, Dict, List, Mapping, Optional, Sequence, Tuple
-
-# Default tool-use budget for an interactive turn, and the higher ceiling a
-# run-to-completion step (a Co4E flow step) is allowed. Same numbers
-# ``core.chat_agent.run_cowork`` defaults to - kept here so the policy is
-# visible in the request rather than buried in a function signature.
-DEFAULT_MAX_STEPS = 30
-DEFAULT_COMPLETION_MAX_STEPS = 200
-
-
-def new_turn_id() -> str:
-    """A fresh turn id. Short and random: it only has to be unique within a
-    session's lifetime, and it shows up in log lines humans read."""
-    return uuid.uuid4().hex[:12]
-
-
-@dataclass(frozen=True)
-class ConversationExecutionRequest:
-    """Everything one agent turn needs, captured at submit time.
-
-    Attributes:
-        prompt: the user's message for this turn (already assembled, including
-            any attachment text the UI inlined).
-        messages: the full conversation to send, oldest first. Held as a tuple
-            so the snapshot cannot be mutated after capture; use
-            :meth:`message_list` to get the mutable copy the engine expects.
-        output_dir: this turn's OWN folder. Each turn writes into an isolated
-            directory so parallel turns cannot clobber each other's files.
-        session_id: the conversation this turn belongs to.
-        turn_id: unique per turn, for logs and for matching events to a turn.
-        surface: which screen submitted it ("cowork", "co4e", "ai_edit", "task").
-        provider / model: what to run on, already resolved (routing included).
-            Empty ``model`` means "the provider's configured default".
-        title: conversation title, used to name generated files.
-        project_id / project_context: the workspace and its shared instructions,
-            snapshotted so a mid-turn workspace switch cannot change them.
-        agent_role: audit-log attribution for every tool call this turn makes.
-        allowed_tools: permission scope. ``None`` means "all enabled tools";
-            a list restricts the ADVERTISED catalogue, so a read-only step
-            literally cannot be offered a writing tool.
-        max_steps / run_to_completion / completion_max_steps: tool-use budget.
-        enforce_rules: run the security rulebase. Co4E sandboxed runs disable it.
-        confirm_commands: ask before run_command/install_package (permission gate).
-        metadata: free-form extras a caller wants carried along (never
-            interpreted here) - e.g. a scheduled task's id.
-    """
-
-    prompt: str
-    messages: Tuple[Mapping[str, Any], ...] = ()
-    output_dir: str = ""
-    session_id: str = ""
-    turn_id: str = field(default_factory=new_turn_id)
-    surface: str = "cowork"
-    provider: str = ""
-    model: str = ""
-    title: str = ""
-    project_id: str = ""
-    project_context: str = ""
-    agent_role: str = ""
-    allowed_tools: Optional[Tuple[str, ...]] = None
-    max_steps: int = DEFAULT_MAX_STEPS
-    run_to_completion: bool = False
-    completion_max_steps: int = DEFAULT_COMPLETION_MAX_STEPS
-    enforce_rules: bool = True
-    confirm_commands: bool = False
-    metadata: Mapping[str, Any] = field(default_factory=dict)
-
-    # -- construction helpers ------------------------------------------- #
-    @classmethod
-    def create(cls, prompt: str, messages: Optional[Sequence[Mapping[str, Any]]] = None,
-               **kwargs: Any) -> "ConversationExecutionRequest":
-        """Build a request from ordinary mutable inputs.
-
-        The messages list is copied element by element, so a later append by the
-        caller (the chat panel keeps appending to its own list) cannot reach
-        inside a request that is already running.
-        """
-        snapshot = tuple(dict(m) for m in (messages or ()))
-        allowed = kwargs.pop("allowed_tools", None)
-        return cls(prompt=prompt, messages=snapshot,
-                   allowed_tools=tuple(allowed) if allowed is not None else None,
-                   **kwargs)
-
-    def with_messages(self, messages: Sequence[Mapping[str, Any]]
-                      ) -> "ConversationExecutionRequest":
-        """A copy carrying a different message list, everything else unchanged.
-
-        Used when a caller assembles the system prompt or trims history after
-        building the request - it must produce a NEW snapshot rather than mutate
-        the one a turn may already be running on.
-        """
-        return replace(self, messages=tuple(dict(m) for m in messages))
-
-    def with_model(self, provider: str, model: str) -> "ConversationExecutionRequest":
-        """A copy pinned to another provider/model - how a routing switch is
-        applied without touching the user's saved settings."""
-        return replace(self, provider=provider, model=model)
-
-    # -- accessors ------------------------------------------------------ #
-    def message_list(self) -> List[Dict[str, Any]]:
-        """A fresh mutable copy of the messages, for the engine to append to.
-
-        The legacy engine mutates the list it is given (it inserts the system
-        prompt and appends assistant/tool messages). Handing it a copy is what
-        keeps this snapshot immutable in practice and not just by declaration.
-        """
-        return [dict(m) for m in self.messages]
-
-    @property
-    def effective_max_steps(self) -> int:
-        """The tool-use ceiling actually in force for this turn."""
-        return self.completion_max_steps if self.run_to_completion else self.max_steps
-
-    @property
-    def has_output_dir(self) -> bool:
-        """True when this turn may write files."""
-        return bool(self.output_dir)
-
-    def allows_tool(self, name: str) -> bool:
-        """Whether ``name`` is inside this turn's permission scope.
-
-        ``update_plan`` is always allowed: it has no side effects and drives the
-        Plan panel, so scoping it out would silently break the UI rather than
-        restrict a capability.
-        """
-        if self.allowed_tools is None:
-            return True
-        return name == "update_plan" or name in self.allowed_tools
-
-    def describe(self) -> str:
-        """Compact one-line identity for log lines."""
-        target = f"{self.provider}/{self.model}" if self.model else self.provider or "default"
-        return f"turn={self.turn_id} surface={self.surface} model={target}"
-
-    def to_dict(self) -> Dict[str, Any]:
-        """JSON-safe projection, for logging a turn or persisting it for replay."""
-        return {
-            "turn_id": self.turn_id,
-            "session_id": self.session_id,
-            "surface": self.surface,
-            "prompt": self.prompt,
-            "message_count": len(self.messages),
-            "output_dir": self.output_dir,
-            "provider": self.provider,
-            "model": self.model,
-            "title": self.title,
-            "project_id": self.project_id,
-            "agent_role": self.agent_role,
-            "allowed_tools": list(self.allowed_tools) if self.allowed_tools is not None else None,
-            "max_steps": self.effective_max_steps,
-            "run_to_completion": self.run_to_completion,
-            "enforce_rules": self.enforce_rules,
-            "confirm_commands": self.confirm_commands,
-            "metadata": dict(self.metadata),
-        }
-
-
-__all__ = [
-    "ConversationExecutionRequest",
-    "new_turn_id",
-    "DEFAULT_MAX_STEPS",
-    "DEFAULT_COMPLETION_MAX_STEPS",
-]

domain/models/__init__.py

@@ -1,5 +1 @@
-"""Domain models: provider/model catalogue value objects (EPIC R03)."""
-
-from .provider_descriptor import ProviderCapability, ProviderDescriptor
-
-__all__ = ["ProviderDescriptor", "ProviderCapability"]
+"""Domain models package: provider descriptors, model pricing, and routing metadata."""

domain/models/provider_descriptor.py

@@ -1,171 +0,0 @@
-"""ProviderDescriptor - the declarative catalogue entry for one model provider (R03-T02).
-
-Today the knowledge of "what a provider is" is scattered across three places
-that must be edited together and can silently drift apart:
-
-* ``providers/factory.py::_REGISTRY``  - name -> implementation class
-* ``config.py::DEFAULT_CONFIG["providers"]`` - default base_url / model / api_key
-* ``config.py::PROVIDER_LABELS``       - the human label shown in Settings
-
-Adding a provider means remembering all three; forgetting one produces a
-provider that exists but has no label, or a label with no implementation. This
-value object folds those facts into a single immutable description that the
-registry (``infrastructure/providers/provider_registry.py``) and the UI can both
-read, so a new provider is declared once.
-
-Pure domain code: stdlib only, no Qt, no network, no config access. It describes
-a provider; building one is infrastructure's job.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass, field
-from enum import Enum
-from typing import Any, Dict, FrozenSet, List, Mapping, Optional, Tuple
-
-
-class ProviderCapability(str, Enum):
-    """What a provider can do, as advertised by its descriptor.
-
-    Kept as a closed enum rather than free-form strings so a typo
-    (``"vison"``) fails at import time instead of silently disabling a feature
-    at runtime. Inherits ``str`` so existing dict/JSON code that compares against
-    plain strings keeps working during the migration.
-    """
-
-    STREAMING = "streaming"      # can stream answer fragments through on_text
-    TOOLS = "tools"              # can be given a ToolSpec catalogue and call tools
-    VISION = "vision"            # accepts image content blocks (see providers/base.py)
-    REASONING = "reasoning"      # emits a separate private "thinking" stream
-    MODEL_LISTING = "model_listing"  # list_models() returns a real catalogue
-
-
-@dataclass(frozen=True)
-class ProviderDescriptor:
-    """An immutable description of one provider the app can talk to.
-
-    Attributes:
-        id: the config key, e.g. ``"openai_compat"``. Also the ``provider`` half
-            of a routing candidate key (``provider/model_id``).
-        label: human-readable name for Settings and the model picker.
-        protocol: which wire format this provider speaks. Several ids share one
-            protocol - ``ollama``, ``github_copilot`` and ``codex`` are all
-            OpenAI-compatible endpoints - which is exactly why protocol and id
-            must be separate fields.
-        default_model: the model used when the user has not chosen one.
-        capabilities: what the provider supports (see :class:`ProviderCapability`).
-        requires_api_key: whether an empty ``api_key`` makes it unusable.
-        requires_base_url: whether an empty ``base_url`` makes it unusable.
-        local: True when the endpoint runs on the user's own machine. Routing
-            treats local models as zero-cost, and the security layer treats them
-            as not leaving the machine, so this is a real behavioural flag and
-            not just documentation.
-        notes: free-form remark shown in Settings (e.g. "paste a Copilot token").
-    """
-
-    id: str
-    label: str
-    protocol: str
-    default_model: str = ""
-    capabilities: FrozenSet[ProviderCapability] = field(default_factory=frozenset)
-    requires_api_key: bool = True
-    requires_base_url: bool = True
-    local: bool = False
-    notes: str = ""
-
-    # -- capability queries ---------------------------------------------- #
-    def supports(self, capability: ProviderCapability) -> bool:
-        """True when this provider advertises ``capability``."""
-        return capability in self.capabilities
-
-    @property
-    def supports_vision(self) -> bool:
-        """Mirrors ``providers.base.Provider.supports_vision`` so callers can ask
-        the descriptor (no instance, no network) before building a provider."""
-        return self.supports(ProviderCapability.VISION)
-
-    @property
-    def supports_tools(self) -> bool:
-        """True when this provider can run an agent turn with tools. A provider
-        without it can still chat, but must never be routed a tool-using task."""
-        return self.supports(ProviderCapability.TOOLS)
-
-    def capability_names(self) -> List[str]:
-        """Capabilities as sorted plain strings - the shape the routing layer's
-        ``required_capabilities`` filter and the assessment store both use."""
-        return sorted(c.value for c in self.capabilities)
-
-    # -- configuration validation ---------------------------------------- #
-    def missing_settings(self, conf: Mapping[str, Any]) -> List[str]:
-        """Which required config keys are absent or blank in ``conf``.
-
-        Returned as a list (not a bool) so Settings can tell the user exactly
-        what to fill in, instead of a generic "not configured". A provider that
-        needs nothing returns an empty list.
-        """
-        missing: List[str] = []
-        if self.requires_api_key and not str(conf.get("api_key", "") or "").strip():
-            missing.append("api_key")
-        if self.requires_base_url and not str(conf.get("base_url", "") or "").strip():
-            missing.append("base_url")
-        return missing
-
-    def is_configured(self, conf: Mapping[str, Any]) -> bool:
-        """True when ``conf`` carries everything this provider needs to run."""
-        return not self.missing_settings(conf)
-
-    def resolve_model(self, conf: Optional[Mapping[str, Any]] = None,
-                      requested: str = "") -> str:
-        """Pick the model id for a call: explicit request, else configured, else
-        this descriptor's default.
-
-        Centralised here because the same three-step fallback is currently
-        re-implemented at every call site (chat panel, Co4E, AI-edit, scheduler),
-        and each of them gets the precedence subtly different.
-        """
-        if requested:
-            return requested
-        configured = str((conf or {}).get("model", "") or "").strip()
-        return configured or self.default_model
-
-    def describe(self, conf: Optional[Mapping[str, Any]] = None) -> str:
-        """One-line summary for logs and the Settings row, e.g.
-        ``"anthropic:claude-sonnet-4-6 (Anthropic Claude)"``."""
-        return f"{self.id}:{self.resolve_model(conf)} ({self.label})"
-
-    def candidate_key(self, model_id: str) -> str:
-        """The ``provider/model_id`` identity the routing layer keys on.
-
-        Defined here so the domain owns the format; ``core.routing.models`` has
-        its own ``candidate_key()`` helper producing the identical string, and
-        keeping them equal is what lets the new registry and the existing
-        assessment store share one keyspace during the migration.
-        """
-        return f"{self.id}/{model_id}"
-
-    def to_dict(self) -> Dict[str, Any]:
-        """JSON-safe projection, for persisting a catalogue snapshot or sending
-        the descriptor to a UI layer that must not import domain types."""
-        return {
-            "id": self.id,
-            "label": self.label,
-            "protocol": self.protocol,
-            "default_model": self.default_model,
-            "capabilities": self.capability_names(),
-            "requires_api_key": self.requires_api_key,
-            "requires_base_url": self.requires_base_url,
-            "local": self.local,
-            "notes": self.notes,
-        }
-
-
-def split_candidate_key(key: str) -> Tuple[str, str]:
-    """Inverse of :meth:`ProviderDescriptor.candidate_key`.
-
-    Splits on the FIRST ``/`` only: some gateways expose model ids that contain
-    a slash (``org/model``), and splitting on the last one would corrupt them.
-    """
-    provider, _, model_id = key.partition("/")
-    return provider, model_id
-
-
-__all__ = ["ProviderCapability", "ProviderDescriptor", "split_candidate_key"]

domain/security/__init__.py

@@ -0,0 +1 @@
+"""Domain security package: security policies, alert events, and permission types."""

domain/tasks/__init__.py

@@ -0,0 +1 @@
+"""Domain tasks package: task definitions and deterministic schedule calculators."""

domain/tools/__init__.py

@@ -1,18 +1 @@
-"""Domain entities for tool risk classification and lookup (EPIC R05)."""
-
-from .tool_descriptor import ToolCapability, ToolDescriptor
-from .tool_registry import (
-    BUILT_IN_CAPABILITIES,
-    UNKNOWN_SOURCE_CAPABILITIES,
-    ToolRegistry,
-    default_registry,
-)
-
-__all__ = [
-    "ToolCapability",
-    "ToolDescriptor",
-    "ToolRegistry",
-    "BUILT_IN_CAPABILITIES",
-    "UNKNOWN_SOURCE_CAPABILITIES",
-    "default_registry",
-]
+"""Domain tools package: tool descriptors, capability scopes, and registry interfaces."""

domain/tools/tool_descriptor.py

@@ -1,86 +0,0 @@
-"""ToolCapability / ToolDescriptor - the risk-tagged catalogue entry for one
-tool the agent loop can call (R05-T01).
-
-Today a tool is just a name inside ``core/tools.py::TOOL_SPECS`` (a
-``providers.base.ToolSpec`` — name/description/JSON-schema parameters, with
-no notion of risk) plus a hand-written membership test wherever gating is
-needed: ``core/tools.py::WRITE_TOOLS``, ``core/code_agent.py``'s
-``WRITE_TOOLS | MS365_WRITE_TOOLS``, and ``core/chat_agent.py``'s literal
-``name in ("run_command", "install_package")``. Three call sites, three
-independently-maintained lists, and a new tool (or an MCP/connector tool,
-which has no list membership at all - see ``core/mcp_client.py``) is gated
-only if someone remembers to add it everywhere.
-
-``ToolDescriptor`` makes the risk an attribute of the tool itself, declared
-once, so ``application/conversations/tool_policy_gateway.py`` (R05-T03) can
-decide ALLOW/CONFIRM/DENY from data instead of a growing set of literal
-tuples.
-
-Pure domain code: stdlib only, no Qt, no I/O. ``to_spec``/``from_spec`` are
-the only place this module touches something outside domain/, and that
-something (``providers.base.ToolSpec``) is itself a plain dataclass with no
-further dependencies.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass, field
-from enum import Flag, auto
-from typing import Any, Dict
-
-from cowork_local.providers.base import ToolSpec
-
-
-class ToolCapability(Flag):
-    """What calling a tool can do to the machine or the network.
-
-    A ``Flag`` (not a plain ``Enum``) because a single tool can combine risks
-    - ``install_package`` writes to the environment, runs pip as a
-    subprocess, AND needs network access. Composing three separate booleans
-    per call site is exactly the duplication this type replaces.
-    """
-
-    NONE = 0
-    READ = auto()
-    WRITE = auto()
-    EXECUTE = auto()
-    NETWORK = auto()
-
-
-@dataclass(frozen=True)
-class ToolDescriptor:
-    """An immutable description of one callable tool.
-
-    Attributes:
-        name: the identifier the model calls (``ToolSpec.name``).
-        description: shown to the model, unchanged from ``ToolSpec``.
-        parameters: JSON-Schema object for the call's arguments.
-        capabilities: the risk this tool carries - see :class:`ToolCapability`.
-    """
-
-    name: str
-    description: str
-    parameters: Dict[str, Any] = field(default_factory=dict)
-    capabilities: ToolCapability = ToolCapability.NONE
-
-    def has(self, capability: ToolCapability) -> bool:
-        """True when this tool carries (any bit of) ``capability``."""
-        return bool(self.capabilities & capability)
-
-    def to_spec(self) -> ToolSpec:
-        """Project back to the ``ToolSpec`` shape the model-facing catalogue
-        and the provider call actually use - risk tagging is metadata the
-        wire format has no room for."""
-        return ToolSpec(name=self.name, description=self.description,
-                        parameters=self.parameters)
-
-    @classmethod
-    def from_spec(cls, spec: ToolSpec,
-                  capabilities: ToolCapability = ToolCapability.NONE) -> "ToolDescriptor":
-        """Wrap an existing ``ToolSpec`` (built-in, MCP, or connector) with a
-        capability tag. The one place callers attach risk to a spec they did
-        not author themselves."""
-        return cls(name=spec.name, description=spec.description,
-                   parameters=spec.parameters, capabilities=capabilities)
-
-
-__all__ = ["ToolCapability", "ToolDescriptor"]

domain/tools/tool_registry.py

@@ -1,125 +0,0 @@
-"""ToolRegistry - the centralised catalogue every tool source registers into
-(R05-T01).
-
-Built-in file/command/fetch tools (``core/tools.py``), MCP server tools
-(``core/mcp_client.py``) and unified connectors (``core/ext_connectors.py``)
-each produce their own ``List[ToolSpec]`` today, concatenated ad-hoc by
-``core/tools.py::combine_tool_sources``. None of that concatenation carries
-risk information, which is exactly why an MCP tool call reaches
-``core/chat_agent.py`` with no ``ToolDescriptor`` to consult and skips the
-permission gate entirely (the gap R05-T04 closes).
-
-``ToolRegistry`` is the one place a :class:`~domain.tools.tool_descriptor.ToolDescriptor`
-is looked up by name, so a policy gateway - or anything else that needs to ask
-"what can this tool do" - has a single source of truth instead of re-deriving
-it from a spec list.
-
-Pure domain code: stdlib only, no Qt, no I/O.
-"""
-from __future__ import annotations
-
-from typing import Dict, Iterable, List, Optional
-
-from cowork_local.providers.base import ToolSpec
-
-from .tool_descriptor import ToolCapability, ToolDescriptor
-
-
-class ToolRegistry:
-    """An in-memory, name-keyed catalogue of :class:`ToolDescriptor`.
-
-    Deliberately mutable and unordered-by-name-only: a turn builds one
-    registry from whichever tool sources it has (built-ins + whatever MCP
-    servers/connectors are enabled), so re-registering the same name simply
-    replaces the previous descriptor rather than raising - the same
-    "last one wins" behaviour ``combine_tool_sources`` already has for
-    duplicate tool names across sources.
-    """
-
-    def __init__(self, descriptors: Optional[Iterable[ToolDescriptor]] = None) -> None:
-        self._by_name: Dict[str, ToolDescriptor] = {}
-        for descriptor in descriptors or ():
-            self.register(descriptor)
-
-    def register(self, descriptor: ToolDescriptor) -> None:
-        self._by_name[descriptor.name] = descriptor
-
-    def get(self, name: str) -> Optional[ToolDescriptor]:
-        return self._by_name.get(name)
-
-    def all(self) -> List[ToolDescriptor]:
-        return list(self._by_name.values())
-
-    def specs(self) -> List[ToolSpec]:
-        """Every registered descriptor, projected back to ``ToolSpec`` - the
-        shape the provider call and the model-facing catalogue need."""
-        return [d.to_spec() for d in self._by_name.values()]
-
-    def capabilities_for(self, name: str) -> ToolCapability:
-        """The capability set for ``name``, or ``NONE`` for an unknown tool.
-
-        Returning ``NONE`` rather than raising lets a policy gateway treat an
-        unregistered tool the same way as one with no declared risk - the
-        gateway's DENY-on-unknown-name rule is a deliberate, separate check,
-        not something this lookup should pre-empt.
-        """
-        descriptor = self._by_name.get(name)
-        return descriptor.capabilities if descriptor is not None else ToolCapability.NONE
-
-    def __contains__(self, name: str) -> bool:
-        return name in self._by_name
-
-    def __len__(self) -> int:
-        return len(self._by_name)
-
-
-# --------------------------------------------------------------------------- #
-# Default capability map for this app's built-in tools (core/tools.py).
-# Kept here, next to the registry, rather than inside core/tools.py itself -
-# core/ is the legacy engine layer being strangled, not where new domain facts
-# should accumulate.
-# --------------------------------------------------------------------------- #
-_CAP = ToolCapability
-BUILT_IN_CAPABILITIES: Dict[str, ToolCapability] = {
-    "read_file": _CAP.READ,
-    "list_dir": _CAP.READ,
-    "write_file": _CAP.WRITE,
-    "edit_file": _CAP.WRITE,
-    "run_command": _CAP.EXECUTE,
-    "install_package": _CAP.WRITE | _CAP.EXECUTE | _CAP.NETWORK,
-    "fetch_url": _CAP.NETWORK,
-    "jira_search": _CAP.NETWORK,
-    "jira_get_issue": _CAP.NETWORK,
-    # Advertised by every engine but has no filesystem/process/network effect
-    # of its own - it only drives the Plan panel (see core/chat_agent.py).
-    "update_plan": _CAP.NONE,
-    "save_file": _CAP.WRITE,
-}
-
-# Tools with no standard, self-declared risk metadata (every MCP server tool,
-# every unified connector) are tagged with this conservative default - see
-# R05-T04. Better to over-gate an unknown remote tool than to silently let it
-# through as READ-only.
-UNKNOWN_SOURCE_CAPABILITIES: ToolCapability = _CAP.WRITE | _CAP.EXECUTE | _CAP.NETWORK
-
-
-def default_registry(specs: Iterable[ToolSpec]) -> ToolRegistry:
-    """Build a registry from ``core/tools.py``'s own ``TOOL_SPECS`` (plus
-    ``save_file``/``update_plan``, which the engines add separately), using
-    :data:`BUILT_IN_CAPABILITIES`. A spec with no entry in that map falls back
-    to :data:`UNKNOWN_SOURCE_CAPABILITIES` - the same conservative default
-    applied to MCP/connector tools, so a built-in nobody has classified yet
-    fails safe instead of silently ungated."""
-    registry = ToolRegistry()
-    for spec in specs:
-        capability = BUILT_IN_CAPABILITIES.get(spec.name, UNKNOWN_SOURCE_CAPABILITIES)
-        registry.register(ToolDescriptor.from_spec(spec, capability))
-    return registry
-
-
-__all__ = [
-    "ToolRegistry",
-    "BUILT_IN_CAPABILITIES",
-    "UNKNOWN_SOURCE_CAPABILITIES",
-    "default_registry",
-]

domain/workspaces/__init__.py

@@ -1,5 +1 @@
-"""Domain entities for workspace/project isolation (EPIC R06)."""
-
-from .workspace_session import WorkspaceSession
-
-__all__ = ["WorkspaceSession"]
+"""Domain workspaces package: immutable WorkspaceSession definitions."""

domain/workspaces/workspace_session.py

@@ -1,96 +0,0 @@
-"""WorkspaceSession - an immutable snapshot of which project a turn belongs
-to and where it may touch the filesystem (R06-T01).
-
-``state.py::AppContext.active_project_id`` is a single mutable field read by
-every background worker thread. ``ui/workspace_tab.py::_load_current`` writes
-it (and the related ``config._project_history_dir``) on the UI thread the
-moment the user switches projects - while a turn already running on a
-worker thread may read either field mid-switch and end up acting on the
-OTHER project's workspace/history for the rest of its run (the race
-R06-T04 fixes).
-
-The fix, same shape as R04's ``ConversationExecutionRequest``: capture the
-workspace facts a turn needs ONCE, on the thread that knows which project is
-selected, into one frozen object. Whatever the user does to the UI afterwards,
-the turn keeps using the workspace it was handed at submit time.
-
-Pure domain code: stdlib only, no Qt, no network. It does touch ``Path`` (not
-plain strings, unlike ``ConversationExecutionRequest``) because its whole job
-is path-containment checking - a snapshot with no room to answer "is this
-path mine" would not replace what ``ToolContext.resolve`` currently does
-inline.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass, field
-from pathlib import Path
-from typing import Tuple
-
-
-@dataclass(frozen=True)
-class WorkspaceSession:
-    """Everything a turn needs to know about ITS workspace, fixed at the
-    moment it was submitted.
-
-    Attributes:
-        project_id: the project this turn belongs to (``""`` when no project
-            is selected - e.g. the Code tab, which has no project concept).
-        workspace_root: the project's sandbox root (``Project.workspace_dir()``).
-        sandbox_dir: the ``.scratch`` subtree inside ``workspace_root`` used for
-            generator/helper scripts, never a final deliverable (see
-            ``infrastructure/filesystem/file_tools.py::_flatten_rel``).
-        allowed_paths: every root a tool call may read/write under. Almost
-            always just ``(workspace_root,)``; a project with a custom
-            ``output_dir`` outside the managed workspace tree still resolves
-            to exactly one root - the tuple exists so a future caller (e.g. a
-            step scoped to a shared input folder) can widen it without a
-            shape change.
-    """
-
-    project_id: str
-    workspace_root: Path
-    sandbox_dir: Path
-    allowed_paths: Tuple[Path, ...] = field(default_factory=tuple)
-
-    def __post_init__(self) -> None:
-        if not self.allowed_paths:
-            object.__setattr__(self, "allowed_paths", (self.workspace_root,))
-
-    @classmethod
-    def from_project(cls, project) -> "WorkspaceSession":
-        """Build a session from a ``core.projects.Project``. ``project`` is
-        typed loosely (not imported) so this module has no dependency on
-        ``core/`` - the caller (``core/projects.py`` itself, or
-        ``application/conversations``) already has the Project in hand."""
-        root = Path(project.workspace_dir())
-        return cls(project_id=project.project_id, workspace_root=root,
-                   sandbox_dir=root / ".scratch", allowed_paths=(root,))
-
-    @classmethod
-    def unscoped(cls, workspace_root: Path) -> "WorkspaceSession":
-        """A session for callers with no project concept (e.g. the Code tab,
-        which sandboxes to a plain folder rather than a ``Project``)."""
-        root = Path(workspace_root)
-        return cls(project_id="", workspace_root=root, sandbox_dir=root / ".scratch")
-
-    def is_allowed(self, path: Path) -> bool:
-        """True when ``path`` resolves inside one of :attr:`allowed_paths`.
-
-        Same containment rule as ``ToolContext.resolve`` (an exact root match
-        or a real descendant), but side-effect-free: it reports the answer
-        instead of raising, so a caller (``FileWorkspaceService``, R06-T05)
-        can decide what "not allowed" means for its own UI instead of
-        catching a ``ToolError``.
-        """
-        try:
-            resolved = Path(path).expanduser().resolve()
-        except OSError:
-            return False
-        for allowed in self.allowed_paths:
-            root = Path(allowed).resolve()
-            if resolved == root or root in resolved.parents:
-                return True
-        return False
-
-
-__all__ = ["WorkspaceSession"]

infrastructure/__init__.py

@@ -1,7 +1 @@
-"""Infrastructure layer - adapters to the outside world.
-
-Concrete implementations of what the inner layers only describe: HTTP calls to
-model gateways, the OS keyring, the filesystem, subprocesses, telemetry sinks.
-May import ``domain/`` (to speak its types) and third-party libraries, but never
-``presentation/``/``ui/``.
-"""
+"""Infrastructure Layer: External system adapters, persistence, and SDK clients."""

infrastructure/config/__init__.py

@@ -0,0 +1 @@
+"""Infrastructure config package: ConfigRepository and typed settings facades."""

infrastructure/filesystem/__init__.py

@@ -1,6 +1 @@
-"""Filesystem/process/network tool adapters split out of ``core/tools.py``
-(EPIC R05) and the sandbox execution context they share."""
-
-from .tool_context import CancelFn, ToolContext, ToolError
-
-__all__ = ["CancelFn", "ToolContext", "ToolError"]
+"""Infrastructure filesystem package: Tool handlers (file, command, fetch tools) and execution workspace."""

infrastructure/filesystem/command_tools.py

@@ -1,110 +0,0 @@
-"""Command tools - run_command, install_package (R05-T02).
-
-Moved verbatim out of ``core/tools.py`` (see ``file_tools.py`` for why). These
-two are the ones today's hand-written permission gate in
-``core/chat_agent.py`` singles out by literal name
-(``name in ("run_command", "install_package")``) — R05-T03 replaces that
-tuple with a capability lookup, but the tools themselves are unchanged here.
-"""
-from __future__ import annotations
-
-import os
-from pathlib import Path
-from typing import Any, Dict, Optional
-
-from .tool_context import CancelFn, ToolContext
-
-COMMAND_TIMEOUT = 120  # seconds
-
-_SNAPSHOT_SKIP = {".git", "__pycache__", "node_modules", ".scratch", ".venv",
-                  ".idea", ".mypy_cache", ".pytest_cache"}
-
-
-def _snapshot(workdir: Path) -> Dict[str, Any]:
-    """Map of file path -> (mtime, size) under the workdir (noise dirs skipped)."""
-    snap: Dict[str, Any] = {}
-    try:
-        for dirpath, dirnames, filenames in os.walk(str(workdir)):
-            dirnames[:] = [d for d in dirnames if d not in _SNAPSHOT_SKIP]
-            for fn in filenames:
-                full = os.path.join(dirpath, fn)
-                try:
-                    st = os.stat(full)
-                    snap[full] = (st.st_mtime_ns, st.st_size)
-                except OSError:
-                    pass
-            if len(snap) > 5000:
-                return snap
-    except OSError:
-        pass
-    return snap
-
-
-def _sandbox_python(ctx: ToolContext, cancel: Optional[CancelFn] = None,
-                     on_output=None) -> Optional[str]:
-    """Lazily create/reuse this ctx's project sandbox venv (Code tab only —
-    ``ctx.sandbox``); returns its python path, or None to use the app's own."""
-    if not ctx.sandbox:
-        return None
-    from cowork_local.core.deps import ensure_project_venv
-
-    py = ensure_project_venv(ctx.workdir, cancel=cancel, on_output=on_output)
-    return str(py) if py else None
-
-
-def run_command(ctx: ToolContext, args: Dict[str, Any],
-                 cancel: Optional[CancelFn] = None,
-                 on_output=None) -> Dict[str, Any]:
-    from cowork_local.core.deps import network_blocked_env, run_cancellable, sandbox_env
-    from cowork_local.core.sandbox_manager import ExecutionConfig, SandboxManager
-    from cowork_local.security.command_risk_classifier import classify_command
-
-    command = str(args.get("command", "")).strip()
-    if not command:
-        return {"ok": False, "output": "Empty command."}
-
-    # --- Security validation pipeline ---
-    risk = classify_command(command, is_cowork_mode=ctx.flatten_writes)
-    if risk.blocked:
-        denial = "Command blocked by security policy: " + "; ".join(risk.reasons)
-        return {"ok": False, "output": denial}
-
-    # Route through SandboxManager for risk-based isolation
-    mgr = SandboxManager(ExecutionConfig(
-        enabled=True,
-        block_network_by_default=ctx.block_network,
-        is_cowork_mode=ctx.flatten_writes,
-    ))
-    sandbox_result = mgr.run(
-        command=command,
-        workdir=str(ctx.workdir),
-        block_network=ctx.block_network,
-        timeout_sec=COMMAND_TIMEOUT,
-        cancel=cancel,
-    )
-    # Sandbox ALWAYS executes (never double-run). Return its result directly.
-    if sandbox_result.get("sandbox") == "blocked":
-        return {"ok": False, "output": sandbox_result.get("stderr", "Command blocked")}
-    out = sandbox_result.get("stdout", "").strip() or "(no output)"
-    err = sandbox_result.get("stderr", "")
-    rc = sandbox_result.get("returncode", -1)
-    if err:
-        out = f"{out}\n{err}" if out else err
-    return {"ok": sandbox_result.get("ok", False), "output": f"[exit {rc}]\n{out}"}
-
-
-def install_package(ctx: ToolContext, args: Dict[str, Any],
-                     cancel: Optional[CancelFn] = None,
-                     on_output=None) -> Dict[str, Any]:
-    from cowork_local.core.deps import pip_install
-
-    package = str(args.get("package", "")).strip()
-    if not package:
-        return {"ok": False, "output": "No package specified."}
-    python = _sandbox_python(ctx, cancel, on_output)
-    ok, detail = pip_install(package, cancel=cancel, on_output=on_output, python=python)
-    head = f"Installed {package}." if ok else f"Could not install {package}."
-    return {"ok": ok, "output": f"{head}\n{detail}"}
-
-
-__all__ = ["COMMAND_TIMEOUT", "run_command", "install_package", "_snapshot"]

infrastructure/filesystem/execution_workspace.py

@@ -1,80 +0,0 @@
-"""ExecutionWorkspace - the output folder vs. the scratch folder for one
-turn, as two distinct properties instead of a name convention (R06-T03).
-
-Today the ``.scratch`` subtree is a special case buried inside
-``_flatten_rel`` (``infrastructure/filesystem/file_tools.py``): a generator
-script writes there, the deliverable lands in the output root, and
-``core/chat_agent.py`` cleans ``.scratch`` up after the turn — but nothing
-NAMES "the scratch folder" as a thing; every call site re-derives
-``workdir / ".scratch"`` (or checks ``Path(rel).parts[0] == ".scratch"``) by
-hand. This class gives that convention one home.
-
-It does not change WHERE files land - ``workspace_root/.scratch`` stays
-exactly what it always was. It exists so a caller (an application service,
-R06-T05's ``FileWorkspaceService``, or a future turn-cleanup step) can ask
-for "the output dir" / "the scratch dir" instead of hand-building the path
-and hoping the convention hasn't drifted.
-"""
-from __future__ import annotations
-
-import shutil
-from dataclasses import dataclass
-from pathlib import Path
-
-from cowork_local.domain.workspaces import WorkspaceSession
-
-SCRATCH_DIRNAME = ".scratch"
-
-
-@dataclass(frozen=True)
-class ExecutionWorkspace:
-    """The two folders a turn actually writes to, derived from a
-    :class:`WorkspaceSession`.
-
-    ``output_dir`` is always the session's ``workspace_root`` itself, not a
-    per-turn subfolder - Cowork's whole design is that every deliverable lands
-    directly in the one configured Output folder (see
-    ``infrastructure/filesystem/file_tools.py::_flatten_rel``'s docstring).
-    ``scratch_dir`` is the SAME flat ``workspace_root/.scratch`` every turn on
-    that workspace already shares today (``core/chat_agent.py``'s
-    ``_cleanup_cowork_intermediates`` operates on that exact path) - this
-    class does not introduce per-turn namespacing that doesn't exist in the
-    engine yet, only names the existing convention.
-
-    ``turn_id`` is kept as metadata for callers that want to attribute a
-    workspace to the turn that used it (logging, future per-turn scratch
-    namespacing); it does not affect either path today.
-    """
-
-    session: WorkspaceSession
-    turn_id: str
-
-    @property
-    def output_dir(self) -> Path:
-        return self.session.workspace_root
-
-    @property
-    def scratch_dir(self) -> Path:
-        return self.session.workspace_root / SCRATCH_DIRNAME
-
-    def ensure_dirs(self) -> None:
-        """Create both folders if they don't exist yet. Callers that only
-        need one (most do) can skip this and let ``write_file`` create parents
-        on demand, same as today."""
-        self.output_dir.mkdir(parents=True, exist_ok=True)
-        self.scratch_dir.mkdir(parents=True, exist_ok=True)
-
-    def cleanup_scratch(self) -> None:
-        """Unconditionally remove the scratch subtree.
-
-        Coarser than ``core/chat_agent.py::_cleanup_cowork_intermediates``,
-        which rescues any real deliverable a generator script wrote INSIDE
-        ``.scratch`` before wiping it - that rescue logic stays there. This
-        is for callers that only need "make the scratch folder go away"
-        (e.g. before starting a fresh run) and know it holds nothing worth
-        saving.
-        """
-        shutil.rmtree(self.scratch_dir, ignore_errors=True)
-
-
-__all__ = ["ExecutionWorkspace", "SCRATCH_DIRNAME"]

infrastructure/filesystem/fetch_tools.py

@@ -1,55 +0,0 @@
-"""Fetch tools - fetch_url, jira_search, jira_get_issue (R05-T02).
-
-Moved verbatim out of ``core/tools.py`` (see ``file_tools.py`` for why). The
-network access these three carry is exactly what the ``ToolCapability.NETWORK``
-tag added in R05-T01/domain/tools/tool_registry.py describes.
-"""
-from __future__ import annotations
-
-from typing import Any, Dict
-
-from .tool_context import ToolContext
-
-
-def fetch_url(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    """Fetch a URL's text content (web page / online document / SharePoint-
-    OneDrive share link) via link_fetch — the same parser task-link attachments
-    use. Honors the Sandbox Security Layer's "Block network" policy."""
-    url = str(args.get("url", "")).strip()
-    if not url:
-        return {"ok": False, "output": "fetch_url: 'url' is required."}
-    if not url.lower().startswith(("http://", "https://")):
-        return {"ok": False, "output": f"fetch_url: not an http(s) URL: {url}"}
-    if not ctx.allow_url_fetch:
-        return {"ok": False,
-                "output": ("fetch_url: URL fetching is turned off in Settings → Security "
-                           "(\"Allow the agent to fetch URLs\").")}
-    # A pasted Jira issue link on the CONNECTED Jira host is read via the
-    # authenticated API (so private issues resolve, not a login page). Public
-    # links / any other URL fall through to the normal fetcher below.
-    from cowork_local.core import jira_tool
-    if jira_tool.is_jira_issue_url(ctx.jira, url):
-        return {"ok": True, "output": jira_tool.get_issue_by_url(ctx.jira, url)}
-    from cowork_local.core.link_fetch import fetch_link_preview
-
-    return {"ok": True, "output": fetch_link_preview(url)}
-
-
-def jira_search(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    from cowork_local.core import jira_tool
-
-    out = jira_tool.search(ctx.jira, str(args.get("jql", "")),
-                           int(args.get("max_results", 25) or 25))
-    return {"ok": not out.lower().startswith(("jira is not configured", "jira search failed")),
-            "output": out}
-
-
-def jira_get_issue(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    from cowork_local.core import jira_tool
-
-    out = jira_tool.get_issue(ctx.jira, str(args.get("key", "")))
-    return {"ok": not out.lower().startswith(("jira is not configured", "could not fetch")),
-            "output": out}
-
-
-__all__ = ["fetch_url", "jira_search", "jira_get_issue"]

infrastructure/filesystem/file_tools.py

@@ -1,136 +0,0 @@
-"""File tools - read_file, list_dir, write_file, edit_file (R05-T02).
-
-Moved verbatim out of ``core/tools.py``, whose ``execute_tool`` used to
-dispatch to these via a hand-written if/elif chain over every tool name it
-knew about. Splitting the built-in handlers into per-concern modules
-(this one, ``command_tools.py``, ``fetch_tools.py``) means adding a tool no
-longer means growing that one function; ``core/tools.py::execute_tool`` now
-looks the name up in a dict built from these modules instead.
-
-Behavior is unchanged from before the split - this is a pure move, not a
-rewrite. Every existing characterization/contract test that exercises
-read_file/write_file/edit_file/list_dir through ``core.tools.execute_tool``
-still exercises the exact same code, just imported from here.
-"""
-from __future__ import annotations
-
-import ast
-from pathlib import Path
-from typing import Any, Dict
-
-from .tool_context import ToolContext
-
-MAX_READ_BYTES = 200_000
-
-
-def _flatten_rel(rel: str) -> str:
-    """Collapse a sub-folder path down to a bare filename so the file lands in the
-    workdir root — EXCEPT the ``.scratch`` sandbox subtree, which is preserved.
-
-    Used by the Cowork agent (flatten_writes=True) so it can never create a
-    per-session / per-chat / per-task output sub-folder: every deliverable stays
-    directly in the single configured Output folder."""
-    parts = Path(rel).parts
-    if parts and parts[0] == ".scratch":
-        return rel  # temporary sandbox is allowed (and cleaned up afterwards)
-    return Path(rel).name or rel
-
-
-def _check_python_syntax(target: Path, content: str) -> str:
-    """Return a short warning if ``content`` is invalid Python, else ''.
-
-    Catches syntax errors the instant a .py file is written/edited — before the
-    agent wastes a whole run_command round-trip just to get the same error back
-    from a traceback."""
-    if target.suffix.lower() not in (".py", ".pyw"):
-        return ""
-    try:
-        ast.parse(content, filename=str(target))
-        return ""
-    except SyntaxError as exc:
-        return f"\n⚠ Syntax error at line {exc.lineno}: {exc.msg} — fix this before running the file."
-
-
-def read_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    target = ctx.resolve(str(args.get("path", "")))
-    if not target.exists():
-        return {"ok": False, "output": f"File not found: {args.get('path')}"}
-    data = target.read_bytes()[:MAX_READ_BYTES]
-    text = data.decode("utf-8", errors="replace")
-    return {"ok": True, "output": text}
-
-
-def list_dir(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    rel = str(args.get("path", ".") or ".")
-    target = ctx.resolve(rel)
-    # A missing/not-yet-created path is NOT a tool failure — report it as an
-    # ordinary result so the agent can create it or pick another path and keep
-    # going. Returning ok=False here surfaced a false "tool failed: list_dir" in
-    # Co4E flows and could stall a step on a recoverable situation.
-    if not target.exists():
-        return {"ok": True, "output": f"(path '{rel}' does not exist yet — create it or use another path)"}
-    if target.is_file():
-        return {"ok": True, "output": f"('{rel}' is a file, not a directory)"}
-    entries = []
-    for child in sorted(target.iterdir(), key=lambda p: (p.is_file(), p.name.lower())):
-        marker = "/" if child.is_dir() else ""
-        entries.append(f"{child.name}{marker}")
-    return {"ok": True, "output": "\n".join(entries) or "(empty folder)"}
-
-
-def write_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    rel = str(args.get("path", ""))
-    if ctx.flatten_writes:
-        rel = _flatten_rel(rel)
-    target = ctx.resolve(rel)
-    content = str(args.get("content", ""))
-    target.parent.mkdir(parents=True, exist_ok=True)
-    # A .xlsx is a binary package — build a REAL workbook from the content
-    # (CSV/TSV/Markdown-table/JSON) rather than writing raw text (which corrupts it).
-    if target.suffix.lower() in (".xlsx", ".xlsm"):
-        from cowork_local.core import xlsx_write
-        if xlsx_write.build_xlsx_from_text(target, content):
-            return {"ok": True, "path": str(target),
-                    "output": f"Wrote spreadsheet {rel} ({target.name})."}
-        return {"ok": False, "output": "Could not build the .xlsx (openpyxl unavailable) — "
-                "write a .csv instead, or use a generator script."}
-    target.write_text(content, encoding="utf-8")
-    warning = _check_python_syntax(target, content)
-    return {"ok": True, "path": str(target),
-            "output": f"Wrote {len(content)} chars to {rel}.{warning}"}
-
-
-def edit_file(ctx: ToolContext, args: Dict[str, Any]) -> Dict[str, Any]:
-    """Replace an exact snippet inside an existing file (precise patch edit)."""
-    rel = str(args.get("path", ""))
-    if ctx.flatten_writes:
-        rel = _flatten_rel(rel)
-    target = ctx.resolve(rel)
-    if not target.exists():
-        return {"ok": False,
-                "output": f"File not found: {rel} — use write_file to create it."}
-    old = str(args.get("old_string", ""))
-    new = str(args.get("new_string", ""))
-    replace_all = bool(args.get("replace_all", False))
-    if not old:
-        return {"ok": False, "output": "old_string is empty — provide the exact text to replace."}
-    try:
-        text = target.read_text(encoding="utf-8", errors="replace")
-    except OSError as exc:
-        return {"ok": False, "output": f"Could not read file: {exc}"}
-    count = text.count(old)
-    if count == 0:
-        return {"ok": False, "output": ("old_string not found. Read the file and copy the exact "
-                                        "text to replace, including indentation/whitespace.")}
-    if count > 1 and not replace_all:
-        return {"ok": False, "output": (f"old_string appears {count} times — add surrounding "
-                                        "context to make it unique, or set replace_all=true.")}
-    updated = text.replace(old, new) if replace_all else text.replace(old, new, 1)
-    target.write_text(updated, encoding="utf-8")
-    n = count if replace_all else 1
-    warning = _check_python_syntax(target, updated)
-    return {"ok": True,
-            "output": f"Edited {args.get('path')} ({n} replacement{'' if n == 1 else 's'}).{warning}"}
-
-
-__all__ = ["MAX_READ_BYTES", "read_file", "list_dir", "write_file", "edit_file"]

infrastructure/filesystem/tool_context.py

@@ -1,62 +0,0 @@
-"""ToolContext / ToolError / CancelFn - the sandboxed execution context every
-built-in tool runs against (moved out of ``core/tools.py`` in R05-T02).
-
-Kept as its own leaf module (no dependency on any sibling in this package) so
-``file_tools.py``, ``command_tools.py`` and ``fetch_tools.py`` can each import
-it without creating an import cycle back through ``core/tools.py``, which
-itself re-exports ``ToolContext``/``ToolError`` from here for the existing
-callers (``core/chat_agent.py``, ``core/code_agent.py``,
-``core/task_executors.py``) that do ``from .tools import ToolContext``.
-"""
-from __future__ import annotations
-
-from dataclasses import dataclass
-from pathlib import Path
-from typing import Any, Callable, Dict, Optional
-
-CancelFn = Callable[[], bool]
-
-
-class ToolError(Exception):
-    pass
-
-
-@dataclass
-class ToolContext:
-    workdir: Path
-    flatten_writes: bool = False  # Cowork: force every write into the workdir root
-    sandbox: bool = False  # Code tab: isolate run_command/install_package into <workdir>/.venv
-    # Sandbox Security Layer — Settings' "Resource Limits" (cpu_percent/memory_mb/
-    # disk_mb), applied to every run_command/install_package this context runs.
-    # None (default) = no limits, matching pre-existing behavior.
-    resource_limits: Optional[Dict[str, float]] = None
-    # Sandbox Security Layer — Settings' "Block network for agent commands"
-    # (policy-level, see deps.py::network_blocked_env). False (default) =
-    # unrestricted, matching pre-existing behavior.
-    block_network: bool = False
-    # Whether the fetch_url tool may read URLs — SEPARATE from block_network
-    # (reading a web page/share link for info is safe; running networked shell
-    # commands is the risk). Defaults True; set from agent_security.allow_url_fetch.
-    allow_url_fetch: bool = True
-    # Jira read connector config (base_url/email/api_token) — None disables the
-    # jira_* tools' ability to connect. Populated from config.data["jira"].
-    jira: Optional[Dict[str, Any]] = None
-
-    def resolve(self, rel: str) -> Path:
-        """Resolve ``rel`` inside the workdir, rejecting escapes."""
-        if rel in ("", "."):
-            return self.workdir
-        candidate = (self.workdir / rel).expanduser()
-        try:
-            resolved = candidate.resolve()
-        except OSError as exc:
-            raise ToolError(f"Invalid path: {rel} ({exc})")
-        root = self.workdir.resolve()
-        if resolved != root and root not in resolved.parents:
-            raise ToolError(
-                f"Refused: '{rel}' is outside the working folder ({root})."
-            )
-        return resolved
-
-
-__all__ = ["CancelFn", "ToolError", "ToolContext"]

infrastructure/mcp/__init__.py

@@ -1,5 +1 @@
-"""MCP server connection lifecycle management (EPIC R05)."""
-
-from .mcp_source_manager import McpToolSourceManager
-
-__all__ = ["McpToolSourceManager"]
+"""Infrastructure MCP package: McpToolSourceManager and child process lifecycle."""

infrastructure/mcp/mcp_source_manager.py

@@ -1,113 +0,0 @@
-"""McpToolSourceManager - the MCP server connection lifecycle, extracted out
-of ``state.py::AppContext`` (R05-T05).
-
-Today ``AppContext.build_mcp_tools`` inlines all of this: a ``_mcp_connections``
-dict, a ``_conn_lock`` guarding check-then-create against concurrent turns (a
-Cowork tab, a Co4E flow and a Scheduled Task can all call it at once), and a
-"start it, cache it, skip it on failure" loop repeated for both the
-admin-configured servers AND the built-in MS365 server
-(``_ms365_builtin_connection``). None of that logic touches Qt; it was only
-ever inline because ``AppContext`` is where the config lived.
-
-This class owns the SAME cache/lock/start-or-skip behavior as a standalone,
-directly testable object — ``AppContext`` becomes a thin caller (one instance
-per app, same as it holds one ``RoutingApplicationService``).
-
-Pure Python: no Qt. It DOES touch the network/filesystem via
-``core.mcp_client.McpServerConnection`` (a subprocess + asyncio loop), which is
-exactly what makes it infrastructure rather than domain.
-"""
-from __future__ import annotations
-
-import threading
-from typing import Dict, List, Optional
-
-from cowork_local.core.mcp_client import McpServerConnection
-
-
-class McpToolSourceManager:
-    """Caches and supervises one :class:`McpServerConnection` per server name.
-
-    ``connection_factory`` defaults to ``McpServerConnection`` itself; tests
-    substitute a fake so no real subprocess is spawned (see
-    ``tests/unit/test_mcp_source_manager.py``).
-    """
-
-    def __init__(self, connection_factory=McpServerConnection) -> None:
-        self._connections: Dict[str, McpServerConnection] = {}
-        self._lock = threading.Lock()
-        self._connection_factory = connection_factory
-
-    def ensure(self, name: str, command: str, args: Optional[List[str]] = None,
-               env: Optional[Dict[str, str]] = None) -> Optional[McpServerConnection]:
-        """Return a live connection for ``name``, starting one if there is
-        none cached or the cached one's subprocess has died.
-
-        Serialized under one lock so two turns racing to build their tool
-        list at the same moment share one subprocess per server instead of
-        each spawning their own (the bug this replaces:
-        ``AppContext._conn_lock``'s original docstring). Returns ``None`` -
-        never raises - when the server fails to start, matching the existing
-        "one broken server must not block the turn" behavior.
-        """
-        with self._lock:
-            existing = self._connections.get(name)
-            if existing is not None and existing.is_alive():
-                return existing
-            if existing is not None:
-                self._connections.pop(name, None)
-            connection = self._connection_factory(name, command, args or [], env)
-            try:
-                connection.start()
-            except Exception:  # noqa: BLE001 - one broken server must not block the turn
-                return None
-            self._connections[name] = connection
-            return connection
-
-    def get(self, name: str) -> Optional[McpServerConnection]:
-        """The cached connection for ``name``, without starting one."""
-        return self._connections.get(name)
-
-    def is_alive(self, name: str) -> bool:
-        connection = self._connections.get(name)
-        return connection is not None and connection.is_alive()
-
-    def restart(self, name: str, command: str, args: Optional[List[str]] = None,
-                env: Optional[Dict[str, str]] = None) -> Optional[McpServerConnection]:
-        """Force a fresh connection for ``name`` even if the cached one still
-        looks alive - for a server the caller knows is misbehaving."""
-        with self._lock:
-            self._connections.pop(name, None)
-        return self.ensure(name, command, args, env)
-
-    def stop(self, name: str) -> None:
-        """Stop and forget one connection - used when a server becomes
-        unavailable by configuration (e.g. MS365 signed out) rather than by
-        crashing."""
-        with self._lock:
-            connection = self._connections.pop(name, None)
-        if connection is not None:
-            try:
-                connection.stop()
-            except Exception:  # noqa: BLE001 - shutdown must never raise into the caller
-                pass
-
-    def active(self) -> List[McpServerConnection]:
-        """Every currently cached connection - what
-        ``core/mcp_client.py::build_mcp_tools`` merges tool specs from."""
-        return list(self._connections.values())
-
-    def stop_all(self) -> None:
-        """Terminate every connection's subprocess - called on app shutdown
-        so none of them linger as orphan processes."""
-        with self._lock:
-            connections = list(self._connections.values())
-            self._connections.clear()
-        for connection in connections:
-            try:
-                connection.stop()
-            except Exception:  # noqa: BLE001
-                pass
-
-
-__all__ = ["McpToolSourceManager"]

infrastructure/persistence/__init__.py

@@ -1 +1 @@
-"""Persistence adapters (EPIC R02/R06)."""
+"""Infrastructure persistence package."""

infrastructure/persistence/json/__init__.py

@@ -1,8 +1 @@
-"""JSON-file persistence adapters: crash-safe writes and the workspace/
-conversation repositories built on them (EPIC R06)."""
-
-from .atomic_write import write_json
-from .conversation_repository_impl import ConversationRepository
-from .workspace_repository_impl import WorkspaceRepository
-
-__all__ = ["write_json", "WorkspaceRepository", "ConversationRepository"]
+"""Infrastructure JSON persistence package: AtomicJsonFile and repositories."""

infrastructure/persistence/json/atomic_write.py

@@ -1,56 +0,0 @@
-"""write_json - crash-safe JSON writes (R06-T02).
-
-``core/projects.py::save_project`` and ``core/history.py``'s
-``save_conversation``/``rename_conversation``/``set_pinned`` all do a plain
-``path.write_text(json.dumps(...))`` today. That is two syscalls with a gap in
-between: a crash, a killed process, or a full disk between the truncate and
-the write leaves a half-written, unparseable JSON file - the NEXT read of
-that project/conversation then fails outright (``load_project`` /
-``load_conversation`` already treat a parse error as "missing", so this isn't
-even a loud failure - a project can silently vanish).
-
-``write_json`` fixes this the standard way: write the full content to a
-temporary file in the SAME directory (so the following replace is on one
-filesystem, not crossing a mount point), then atomically rename it over the
-target. Either the old file is still there, or the new one is fully there -
-never a partial one.
-
-Transitional note: EPIC R02 (Team Nam, ``docs/refactor/Refactoring_Checklist.md``
-R02-T01) plans a shared ``infrastructure/persistence/json/atomic_json_file.py``
-for the SAME purpose across the whole app (config, secrets, ...). This module
-is deliberately named differently and scoped to R06's two repositories only,
-so the two EPICs don't edit the same file in parallel; once R02-T01 lands,
-``WorkspaceRepository``/``ConversationRepository`` should switch to it and
-this module can go away.
-"""
-from __future__ import annotations
-
-import json
-import os
-import tempfile
-from pathlib import Path
-from typing import Any
-
-
-def write_json(path: Path, data: Any) -> None:
-    """Serialize ``data`` as indented UTF-8 JSON and write it to ``path``
-    atomically. Creates parent directories if needed."""
-    path = Path(path)
-    path.parent.mkdir(parents=True, exist_ok=True)
-    text = json.dumps(data, ensure_ascii=False, indent=2)
-    fd, tmp_name = tempfile.mkstemp(dir=str(path.parent), prefix=f".{path.name}.", suffix=".tmp")
-    try:
-        with os.fdopen(fd, "w", encoding="utf-8") as handle:
-            handle.write(text)
-            handle.flush()
-            os.fsync(handle.fileno())
-        os.replace(tmp_name, path)
-    except BaseException:
-        try:
-            os.unlink(tmp_name)
-        except OSError:
-            pass
-        raise
-
-
-__all__ = ["write_json"]

infrastructure/persistence/json/conversation_repository_impl.py

@@ -1,54 +0,0 @@
-"""ConversationRepository - an object-shaped, atomic-write-backed facade over
-``core/history.py`` (R06-T02). Same rationale as
-``workspace_repository_impl.py``: the module-level functions in
-``core/history.py`` are still what production code calls (they now write
-atomically themselves), this class is the seam for application-layer code
-that wants an object instead of a directory-parameterised function.
-"""
-from __future__ import annotations
-
-from pathlib import Path
-from typing import Any, Dict, List, Optional
-
-from cowork_local.config import HISTORY_DIR
-from cowork_local.core.history import (
-    delete_conversation,
-    list_conversations,
-    load_conversation,
-    new_session_id,
-    rename_conversation,
-    save_conversation,
-    set_pinned,
-)
-
-
-class ConversationRepository:
-    """CRUD + search over conversation JSON files, scoped to one
-    ``directory`` (defaults to the app's real ``HISTORY_DIR``)."""
-
-    def __init__(self, directory: Optional[Path] = None) -> None:
-        self._directory = Path(directory) if directory is not None else HISTORY_DIR
-
-    def new_session_id(self) -> str:
-        return new_session_id()
-
-    def save(self, kind: str, session_id: str, messages: List[Dict[str, Any]], **kwargs) -> Path:
-        return save_conversation(self._directory, kind, session_id, messages, **kwargs)
-
-    def load(self, path: Path) -> Dict[str, Any]:
-        return load_conversation(path)
-
-    def list(self, query: str = "") -> List[Dict[str, Any]]:
-        return list_conversations(self._directory, query)
-
-    def delete(self, path: Path) -> None:
-        delete_conversation(path)
-
-    def rename(self, path: Path, new_title: str) -> None:
-        rename_conversation(path, new_title)
-
-    def set_pinned(self, path: Path, pinned: bool) -> None:
-        set_pinned(path, pinned)
-
-
-__all__ = ["ConversationRepository"]

infrastructure/persistence/json/workspace_repository_impl.py

@@ -1,59 +0,0 @@
-"""WorkspaceRepository - an object-shaped, atomic-write-backed facade over
-``core/projects.py`` (R06-T02).
-
-``core/projects.py``'s module-level functions (``list_projects``,
-``load_project``, ``save_project``, ``new_project``, ``delete_project``) are
-still what every existing call site (``ui/workspace_tab.py``, ``state.py``,
-task executors) uses, and stay that way - they now write through
-:func:`atomic_write.write_json` themselves, so the durability fix applies
-whether or not a caller ever touches this class.
-
-This repository exists for the application layer (``application/workspaces``,
-R06-T05) to depend on an interface instead of reaching into ``core/`` -
-useful once code above ``core/`` starts being written against
-``domain``/``application`` seams instead of the legacy module functions. It
-is a thin pass-through today, not a re-implementation: same on-disk format,
-same directory, same functions underneath.
-"""
-from __future__ import annotations
-
-from pathlib import Path
-from typing import List, Optional
-
-from cowork_local.core.projects import (
-    PROJECTS_DIR,
-    Project,
-    delete_project,
-    list_projects,
-    load_project,
-    new_project,
-    save_project,
-)
-
-
-class WorkspaceRepository:
-    """CRUD over :class:`~cowork_local.core.projects.Project`, scoped to one
-    ``directory`` (defaults to the app's real ``PROJECTS_DIR``; tests pass a
-    ``tmp_path`` so nothing touches the user's real config folder)."""
-
-    def __init__(self, directory: Optional[Path] = None) -> None:
-        self._directory = directory or PROJECTS_DIR
-
-    def list(self) -> List[Project]:
-        return list_projects(self._directory)
-
-    def get(self, project_id: str) -> Optional[Project]:
-        return load_project(project_id, self._directory)
-
-    def save(self, project: Project) -> Path:
-        return save_project(project, self._directory)
-
-    def create(self, name: str, description: str = "", instructions: str = "",
-               output_dir: str = "") -> Project:
-        return new_project(name, description, instructions, output_dir, self._directory)
-
-    def delete(self, project_id: str) -> bool:
-        return delete_project(project_id, self._directory)
-
-
-__all__ = ["WorkspaceRepository"]

infrastructure/platform/__init__.py

@@ -0,0 +1 @@
+"""Infrastructure platform adapters package."""

infrastructure/platform/qt/__init__.py

@@ -0,0 +1 @@
+"""Infrastructure Qt platform adapters: QtSchedulerClock."""

infrastructure/providers/__init__.py

@@ -1,5 +1 @@
-"""Provider adapters and the central provider catalogue (EPIC R03)."""
-
-from .provider_registry import ProviderRegistry, default_registry
-
-__all__ = ["ProviderRegistry", "default_registry"]
+"""Infrastructure providers package: LLM provider adapters and ProviderRegistry."""

infrastructure/providers/provider_registry.py

@@ -1,207 +0,0 @@
-"""ProviderRegistry - the one place a provider is declared (R03-T02).
-
-Replaces the three-way split between ``providers/factory.py::_REGISTRY``,
-``config.py::DEFAULT_CONFIG["providers"]`` and ``config.py::PROVIDER_LABELS``
-with a single catalogue of :class:`ProviderDescriptor` objects plus the
-implementation class each one maps to.
-
-Adding a provider is now one entry in :data:`BUILT_IN_PROVIDERS` (declarative
-facts) and one line in :data:`_IMPLEMENTATIONS` (which class speaks that
-protocol) - see ``docs/governance/contributor-recipes.md`` (R10-T04).
-
-Migration note (strangler fig, ADR-001 section 4): this registry does not
-re-implement any provider. It builds the SAME classes ``providers/factory.py``
-builds, so both entry points stay behaviourally identical while call sites move
-over one at a time.
-"""
-from __future__ import annotations
-
-from typing import Any, Dict, Iterable, List, Mapping, Optional
-
-from cowork_local.domain.models.provider_descriptor import (
-    ProviderCapability,
-    ProviderDescriptor,
-)
-from cowork_local.providers.base import Provider, ProviderError
-
-_CAP = ProviderCapability
-
-# Every provider the app ships with, described once.
-#
-# The capability sets are deliberately conservative: a capability listed here is
-# one the adapter genuinely implements today. Claiming VISION for a provider
-# whose chat() cannot translate an image block would route an image turn into a
-# guaranteed failure, so an unimplemented capability must stay off the list.
-BUILT_IN_PROVIDERS: tuple = (
-    ProviderDescriptor(
-        id="openai_compat",
-        label="OpenAI-compatible (Internal Gateway)",
-        protocol="openai_compat",
-        default_model="gpt-4o-mini",
-        capabilities=frozenset({_CAP.STREAMING, _CAP.TOOLS, _CAP.VISION,
-                                _CAP.REASONING, _CAP.MODEL_LISTING}),
-        notes="Any endpoint speaking the OpenAI Chat Completions protocol.",
-    ),
-    ProviderDescriptor(
-        id="anthropic",
-        label="Anthropic Claude",
-        protocol="anthropic",
-        default_model="claude-sonnet-4-6",
-        capabilities=frozenset({_CAP.STREAMING, _CAP.TOOLS, _CAP.VISION,
-                                _CAP.MODEL_LISTING}),
-    ),
-    ProviderDescriptor(
-        id="ollama",
-        label="Ollama (local models)",
-        protocol="openai_compat",
-        default_model="llama3.1",
-        capabilities=frozenset({_CAP.STREAMING, _CAP.TOOLS, _CAP.REASONING,
-                                _CAP.MODEL_LISTING}),
-        # Ollama ignores the key, but the OpenAI client layer requires a value,
-        # so the default config ships a placeholder rather than an empty string.
-        requires_api_key=False,
-        local=True,
-        notes="Runs on this machine - no data leaves the device, no token cost.",
-    ),
-    ProviderDescriptor(
-        id="github_copilot",
-        label="GitHub Copilot",
-        protocol="openai_compat",
-        default_model="gpt-4o",
-        capabilities=frozenset({_CAP.STREAMING, _CAP.TOOLS, _CAP.MODEL_LISTING}),
-        notes="Paste a Copilot token as the API key.",
-    ),
-    ProviderDescriptor(
-        id="codex",
-        label="OpenAI (Codex / GPT)",
-        protocol="openai_compat",
-        default_model="gpt-4o-mini",
-        capabilities=frozenset({_CAP.STREAMING, _CAP.TOOLS, _CAP.VISION,
-                                _CAP.REASONING, _CAP.MODEL_LISTING}),
-    ),
-)
-
-
-def _implementations() -> Dict[str, type]:
-    """Protocol -> adapter class.
-
-    Imported lazily inside the function because ``providers/anthropic.py`` and
-    ``providers/openai_compat.py`` pull in ``requests`` at import time; keeping
-    that out of module import means a test that only inspects descriptors pays
-    no import cost at all.
-    """
-    from cowork_local.providers.anthropic import AnthropicProvider
-    from cowork_local.providers.openai_compat import OpenAICompatProvider
-
-    return {
-        "openai_compat": OpenAICompatProvider,
-        "anthropic": AnthropicProvider,
-    }
-
-
-class ProviderRegistry:
-    """Catalogue of known providers + the factory that instantiates them.
-
-    Intentionally holds no config and no app context: it is a pure lookup table
-    plus a build step, so it can be constructed in a test with a custom
-    descriptor list and no application running.
-    """
-
-    def __init__(self, descriptors: Optional[Iterable[ProviderDescriptor]] = None) -> None:
-        # Dict preserves declaration order (Python 3.7+), which is the order
-        # Settings lists providers in - so the catalogue order is data, not luck.
-        self._by_id: Dict[str, ProviderDescriptor] = {
-            d.id: d for d in (descriptors if descriptors is not None else BUILT_IN_PROVIDERS)
-        }
-
-    # -- catalogue queries ------------------------------------------------ #
-    def ids(self) -> List[str]:
-        """Known provider ids, in declaration order."""
-        return list(self._by_id)
-
-    def all(self) -> List[ProviderDescriptor]:
-        """Every descriptor, in declaration order."""
-        return list(self._by_id.values())
-
-    def get(self, provider_id: str) -> Optional[ProviderDescriptor]:
-        """The descriptor for ``provider_id``, or None when unknown.
-
-        Returns None rather than raising because the caller is often reacting to
-        a config file that may name a provider from a newer version; the UI
-        should be able to skip it, not crash.
-        """
-        return self._by_id.get(provider_id)
-
-    def require(self, provider_id: str) -> ProviderDescriptor:
-        """Like :meth:`get` but raises :class:`ProviderError` when unknown.
-
-        Same error type ``providers/factory.py::build_provider`` already raises,
-        so callers that migrate to the registry keep their existing except clause.
-        """
-        descriptor = self._by_id.get(provider_id)
-        if descriptor is None:
-            known = ", ".join(self._by_id) or "(none)"
-            raise ProviderError(f"Unsupported provider: {provider_id} (known: {known})")
-        return descriptor
-
-    def labels(self) -> Dict[str, str]:
-        """``{id: label}`` - the drop-in replacement for ``config.PROVIDER_LABELS``."""
-        return {d.id: d.label for d in self._by_id.values()}
-
-    def supporting(self, capability: ProviderCapability) -> List[ProviderDescriptor]:
-        """Every descriptor advertising ``capability`` - used to answer "which
-        providers could serve this turn?" before any of them is built."""
-        return [d for d in self._by_id.values() if d.supports(capability)]
-
-    def configured(self, providers_conf: Mapping[str, Mapping[str, Any]]
-                   ) -> List[ProviderDescriptor]:
-        """Descriptors whose config section is complete enough to actually call.
-
-        ``providers_conf`` is ``AppConfig.data["providers"]``. Passing the raw
-        mapping (not the AppConfig object) keeps this layer independent of the
-        config implementation, which EPIC R02 is rewriting in parallel.
-        """
-        return [d for d in self._by_id.values()
-                if d.is_configured(providers_conf.get(d.id, {}) or {})]
-
-    # -- construction ----------------------------------------------------- #
-    def build(self, provider_id: str, conf: Mapping[str, Any],
-              model: str = "") -> Provider:
-        """Instantiate the adapter for ``provider_id``.
-
-        ``model`` overrides the configured model for this instance only - that is
-        how the routing layer runs one turn on a different model without mutating
-        the user's saved settings.
-        """
-        descriptor = self.require(provider_id)
-        impl = _implementations().get(descriptor.protocol)
-        if impl is None:  # pragma: no cover - only reachable via a bad descriptor
-            raise ProviderError(
-                f"Provider '{provider_id}' declares unknown protocol "
-                f"'{descriptor.protocol}'."
-            )
-        # Copy before mutating: conf is the caller's live config dict, and
-        # writing the routed model into it would silently change the user's
-        # saved default for every later turn.
-        resolved = dict(conf or {})
-        resolved["model"] = descriptor.resolve_model(conf, model)
-        instance = impl(resolved)
-        # The adapter class is shared by several ids (three of them are
-        # OpenAI-compatible), so its class-level `name` cannot identify which
-        # provider this is. Stamping the instance keeps usage records, audit
-        # entries and routing candidate keys attributed to the right provider.
-        instance.name = descriptor.id
-        return instance
-
-    def describe(self, provider_id: str, conf: Optional[Mapping[str, Any]] = None) -> str:
-        """One-line description used in logs and error messages."""
-        return self.require(provider_id).describe(conf)
-
-
-# Shared default instance. Callers that need the built-in catalogue use this
-# instead of constructing a registry each time; tests build their own with an
-# explicit descriptor list.
-default_registry = ProviderRegistry()
-
-
-__all__ = ["ProviderRegistry", "BUILT_IN_PROVIDERS", "default_registry"]

infrastructure/sandbox/__init__.py

@@ -0,0 +1 @@
+"""Infrastructure sandbox package: OS-specific sandbox capability adapters."""

infrastructure/telemetry/__init__.py

@@ -1,21 +1 @@
-"""Telemetry sinks: where token usage and turn metrics are recorded (EPIC R03)."""
-
-from .usage_sink import (
-    NullUsageSink,
-    RecordingUsageSink,
-    UsageEvent,
-    UsageEventSink,
-    UsageTrackerSink,
-    default_sink,
-    set_default_sink,
-)
-
-__all__ = [
-    "UsageEvent",
-    "UsageEventSink",
-    "UsageTrackerSink",
-    "NullUsageSink",
-    "RecordingUsageSink",
-    "default_sink",
-    "set_default_sink",
-]
+"""Infrastructure telemetry package: CanonicalAuditLogger and token usage sinks."""

infrastructure/telemetry/usage_sink.py

@@ -1,229 +0,0 @@
-"""UsageEventSink - where a turn's token usage goes (R03-T06).
-
-Today each provider records its own usage inline, in the middle of the streaming
-loop::
-
-    # providers/openai_compat.py
-    def _record_usage(self, messages, text_parts, tool_acc, usage_seen):
-        from ..core import usage_tracker as ut
-        ...
-        ut.record(self.name, self.model, ...)
-
-Three problems with that shape:
-
-1. **Hidden side effect.** ``chat()`` looks like a pure request/response call but
-   also writes to the Dashboard's store, so a test of a provider silently
-   appends rows to the developer's real usage history.
-2. **Duplicated estimation.** The "no usage block from the server, so estimate
-   at ~4 chars/token" fallback is copy-pasted per provider and can drift.
-3. **One hard-wired destination.** Usage can only ever go to
-   ``core.usage_tracker``; a run that wants to bill a workflow, or a test that
-   wants to assert on token counts, has nowhere to plug in.
-
-This module introduces the seam: providers build a :class:`UsageEvent` and hand
-it to a :class:`UsageEventSink`. Production wires :class:`UsageTrackerSink`
-(same destination, same numbers as before); tests wire
-:class:`RecordingUsageSink` or :class:`NullUsageSink`.
-"""
-from __future__ import annotations
-
-import logging
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional, Protocol, Sequence
-
-logger = logging.getLogger("cowork_local.telemetry")
-
-# Rough characters-per-token ratio used when the gateway sends no usage block.
-# Matches the constant behaviour of ``core.usage_tracker.estimate_tokens`` so
-# moving the estimation here does not change a single recorded number.
-_CHARS_PER_TOKEN = 4
-
-
-@dataclass(frozen=True)
-class UsageEvent:
-    """Token usage for exactly one provider round trip.
-
-    ``estimated`` marks a record derived from text length rather than reported by
-    the server. The Dashboard shows the two differently, and conflating them
-    would make cost figures look more precise than they are.
-    """
-
-    provider: str
-    model: str
-    input_tokens: int = 0
-    output_tokens: int = 0
-    cached_tokens: int = 0
-    estimated: bool = False
-
-    @property
-    def total_tokens(self) -> int:
-        """Input + output. Cached tokens are a subset of input, not an addition,
-        so adding them here would double-count a cache hit."""
-        return self.input_tokens + self.output_tokens
-
-    def to_dict(self) -> Dict[str, Any]:
-        """JSON-safe projection for logs and for sinks that persist raw events."""
-        return {
-            "provider": self.provider,
-            "model": self.model,
-            "input_tokens": self.input_tokens,
-            "output_tokens": self.output_tokens,
-            "cached_tokens": self.cached_tokens,
-            "estimated": self.estimated,
-        }
-
-
-class UsageEventSink(Protocol):
-    """Anything that can absorb a :class:`UsageEvent`.
-
-    Implementations MUST NOT raise: telemetry is observability, and a failure to
-    record usage must never abort the turn that produced it.
-    """
-
-    def record(self, event: UsageEvent) -> None:
-        """Absorb one usage event."""
-
-
-class NullUsageSink:
-    """Discards everything. The default for tests and headless tooling, so a
-    unit test never writes into the developer's real usage history."""
-
-    def record(self, event: UsageEvent) -> None:  # noqa: D102 - see protocol
-        return None
-
-
-class RecordingUsageSink:
-    """Keeps events in memory so a test can assert on what was recorded."""
-
-    def __init__(self) -> None:
-        self.events: List[UsageEvent] = []
-
-    def record(self, event: UsageEvent) -> None:  # noqa: D102 - see protocol
-        self.events.append(event)
-
-    @property
-    def total_tokens(self) -> int:
-        """Sum across every recorded event."""
-        return sum(e.total_tokens for e in self.events)
-
-
-class UsageTrackerSink:
-    """Forwards to ``core.usage_tracker`` - the Dashboard's store.
-
-    This is the production sink and the only place that still knows about the
-    legacy tracker module, which is what lets EPIC R10 replace the storage
-    without touching a single provider.
-    """
-
-    def __init__(self, tracker: Optional[Any] = None) -> None:
-        # Injectable for tests; imported lazily otherwise because the tracker
-        # touches the config directory at import time.
-        self._tracker = tracker
-
-    def _resolve(self) -> Any:
-        if self._tracker is None:
-            from cowork_local.core import usage_tracker
-
-            self._tracker = usage_tracker
-        return self._tracker
-
-    def record(self, event: UsageEvent) -> None:
-        """Write the event to the usage tracker, swallowing any failure.
-
-        The bare except mirrors the behaviour this replaces (each provider
-        already wrapped its ``ut.record`` call in ``try/except: pass``) but logs
-        at debug level instead of discarding the reason entirely, so a broken
-        Dashboard store can at least be diagnosed.
-        """
-        try:
-            self._resolve().record(
-                event.provider, event.model,
-                event.input_tokens, event.output_tokens, event.cached_tokens,
-                estimated=event.estimated,
-            )
-        except Exception:  # noqa: BLE001 - telemetry must never break a turn
-            logger.debug("usage sink: failed to record %s", event.to_dict(), exc_info=True)
-
-
-def estimate_tokens(text: str) -> int:
-    """Approximate token count for ``text`` (~4 characters per token).
-
-    Deliberately identical to ``core.usage_tracker.estimate_tokens`` so that
-    moving estimation into this layer changes no recorded number. Duplicated
-    rather than imported to keep this module free of the legacy dependency;
-    :class:`UsageTrackerSink` is the only bridge back to it.
-    """
-    return max(0, len(text or "") // _CHARS_PER_TOKEN)
-
-
-def estimated_event(provider: str, model: str, sent: str, received: str) -> UsageEvent:
-    """Build an estimated :class:`UsageEvent` from the raw text of a round trip.
-
-    Used when the gateway sends no usage block - most self-hosted OpenAI-compatible
-    servers and Ollama do not.
-    """
-    return UsageEvent(
-        provider=provider, model=model,
-        input_tokens=estimate_tokens(sent),
-        output_tokens=estimate_tokens(received),
-        cached_tokens=0,
-        estimated=True,
-    )
-
-
-def openai_usage_event(provider: str, model: str, usage: Dict[str, Any]) -> UsageEvent:
-    """Build a reported :class:`UsageEvent` from an OpenAI-style usage block."""
-    details = usage.get("prompt_tokens_details") or {}
-    return UsageEvent(
-        provider=provider, model=model,
-        input_tokens=int(usage.get("prompt_tokens", 0) or 0),
-        output_tokens=int(usage.get("completion_tokens", 0) or 0),
-        cached_tokens=int(details.get("cached_tokens", 0) or 0),
-        estimated=False,
-    )
-
-
-def anthropic_usage_event(provider: str, model: str, usage: Dict[str, Any]) -> UsageEvent:
-    """Build a reported :class:`UsageEvent` from Anthropic's usage accumulator.
-
-    Anthropic reports input tokens on ``message_start`` and output tokens on
-    ``message_delta``, so ``providers/anthropic.py`` accumulates them into a dict
-    keyed ``in``/``out``/``cache`` - this reads that shape.
-    """
-    return UsageEvent(
-        provider=provider, model=model,
-        input_tokens=int(usage.get("in", 0) or 0),
-        output_tokens=int(usage.get("out", 0) or 0),
-        cached_tokens=int(usage.get("cache", 0) or 0),
-        estimated=False,
-    )
-
-
-# The sink providers use unless one is injected. A module-level default keeps
-# the change to the provider classes to a single attribute, and lets a test swap
-# the destination process-wide with one monkeypatch.
-default_sink: UsageEventSink = UsageTrackerSink()
-
-
-def set_default_sink(sink: UsageEventSink) -> UsageEventSink:
-    """Replace the process-wide default sink; returns the previous one so a
-    caller (or fixture) can restore it."""
-    global default_sink
-    previous = default_sink
-    default_sink = sink
-    return previous
-
-
-__all__ = [
-    "UsageEvent",
-    "UsageEventSink",
-    "UsageTrackerSink",
-    "NullUsageSink",
-    "RecordingUsageSink",
-    "estimate_tokens",
-    "estimated_event",
-    "openai_usage_event",
-    "anthropic_usage_event",
-    "default_sink",
-    "set_default_sink",
-]

mcp_servers/project_context/__init__.py

@@ -0,0 +1,5 @@
+"""Provider-neutral Project Context MCP server template."""
+
+from .server import build_server, dispatch
+
+__all__ = ["build_server", "dispatch"]

mcp_servers/project_context/foundation.py

@@ -0,0 +1,106 @@
+"""Shared, stable boundary used by all Project Context tool work packages."""
+
+from __future__ import annotations
+
+from collections.abc import Callable
+from dataclasses import dataclass
+from datetime import datetime
+from typing import Any, Protocol
+
+from pydantic import AnyUrl, BaseModel, ConfigDict, Field
+
+
+class ContractModel(BaseModel):
+    """Strict immutable model so provider-specific fields cannot leak to the Agent."""
+
+    model_config = ConfigDict(extra="forbid", frozen=True)
+
+
+class IdentityContext(ContractModel):
+    actor_id: str = Field(min_length=1, max_length=256)
+    org_unit: str = Field(min_length=1, max_length=128)
+    customer: str = Field(min_length=1, max_length=128)
+    project: str = Field(min_length=1, max_length=128)
+    granted_scopes: frozenset[str]
+
+
+class SourceCitation(ContractModel):
+    system: str = Field(min_length=1, max_length=64)
+    url: AnyUrl
+    revision: str = Field(min_length=1, max_length=256)
+    retrieved_at: datetime
+
+
+@dataclass(frozen=True)
+class DispatchResult:
+    ok: bool
+    payload: dict[str, Any]
+
+
+class PolicyDecisionPoint(Protocol):
+    def decide(self, identity: IdentityContext, tool_name: str, project_id: str) -> bool: ...
+
+
+class CredentialResolver(Protocol):
+    def resolve(self, identity: IdentityContext, tool_name: str) -> Any: ...
+
+
+@dataclass(frozen=True)
+class ProjectContextRuntime:
+    identity: IdentityContext
+    policy: PolicyDecisionPoint
+    credential_resolver: CredentialResolver
+
+
+class ProviderError(RuntimeError):
+    """A provider failure with a caller-safe message and retry classification."""
+
+    def __init__(self, code: str, message: str, *, retryable: bool) -> None:
+        super().__init__(message)
+        self.code = code
+        self.safe_message = message
+        self.retryable = retryable
+
+
+ToolHandler = Callable[[ContractModel, Any], dict[str, Any]]
+
+
+@dataclass(frozen=True)
+class ToolTemplate:
+    name: str
+    description: str
+    input_model: type[ContractModel]
+    output_model: type[ContractModel]
+    handler: ToolHandler
+
+    def declaration(self) -> dict[str, Any]:
+        return {
+            "name": self.name,
+            "description": self.description,
+            "inputSchema": self.input_model.model_json_schema(),
+            "outputSchema": self.output_model.model_json_schema(),
+        }
+
+
+def error_result(
+    code: str,
+    *,
+    category: str,
+    retryable: bool,
+    message: str,
+    suggested_action: str,
+    correlation_id: str,
+) -> DispatchResult:
+    return DispatchResult(
+        ok=False,
+        payload={
+            "error": {
+                "code": code,
+                "category": category,
+                "retryable": retryable,
+                "message": message,
+                "suggested_action": suggested_action,
+                "correlation_id": correlation_id,
+            }
+        },
+    )

mcp_servers/project_context/providers/__init__.py

@@ -0,0 +1 @@
+"""One provider module per member-owned tool work package."""

mcp_servers/project_context/providers/change.py

@@ -0,0 +1,25 @@
+"""Provider boundary owned with get_project_change_context."""
+
+from __future__ import annotations
+
+from typing import Any, Protocol
+
+from ..foundation import IdentityContext, ProviderError
+
+
+class ChangeProvider(Protocol):
+    def get_change_context(self, **arguments: Any) -> dict[str, Any]: ...
+
+
+class UnconfiguredChangeProvider:
+    def get_change_context(self, **arguments: Any) -> dict[str, Any]:
+        raise ProviderError(
+            "UNAVAILABLE",
+            "The change provider is not configured for this environment.",
+            retryable=False,
+        )
+
+
+def build_provider(identity: IdentityContext) -> ChangeProvider:
+    """Replace only this factory when wiring the approved read-only Git adapter."""
+    return UnconfiguredChangeProvider()

mcp_servers/project_context/providers/issue.py

@@ -0,0 +1,25 @@
+"""Provider boundary owned with get_project_issue_context."""
+
+from __future__ import annotations
+
+from typing import Any, Protocol
+
+from ..foundation import IdentityContext, ProviderError
+
+
+class IssueProvider(Protocol):
+    def get_issue_context(self, **arguments: Any) -> dict[str, Any]: ...
+
+
+class UnconfiguredIssueProvider:
+    def get_issue_context(self, **arguments: Any) -> dict[str, Any]:
+        raise ProviderError(
+            "UNAVAILABLE",
+            "The issue provider is not configured for this environment.",
+            retryable=False,
+        )
+
+
+def build_provider(identity: IdentityContext) -> IssueProvider:
+    """Replace only this factory when wiring the approved read-only issue adapter."""
+    return UnconfiguredIssueProvider()

mcp_servers/project_context/providers/knowledge.py

@@ -0,0 +1,25 @@
+"""Provider boundary owned with search_project_knowledge."""
+
+from __future__ import annotations
+
+from typing import Any, Protocol
+
+from ..foundation import IdentityContext, ProviderError
+
+
+class KnowledgeProvider(Protocol):
+    def search_knowledge(self, **arguments: Any) -> dict[str, Any]: ...
+
+
+class UnconfiguredKnowledgeProvider:
+    def search_knowledge(self, **arguments: Any) -> dict[str, Any]:
+        raise ProviderError(
+            "UNAVAILABLE",
+            "The knowledge provider is not configured for this environment.",
+            retryable=False,
+        )
+
+
+def build_provider(identity: IdentityContext) -> KnowledgeProvider:
+    """Replace only this factory when wiring approved project retrieval."""
+    return UnconfiguredKnowledgeProvider()

mcp_servers/project_context/registry.py

@@ -0,0 +1,23 @@
+"""Immutable registry composed before member work starts to prevent merge conflicts."""
+
+from __future__ import annotations
+
+from types import MappingProxyType
+from typing import Any
+
+from .foundation import ToolTemplate
+from .tools.change_context import TOOL as CHANGE_CONTEXT_TOOL
+from .tools.issue_context import TOOL as ISSUE_CONTEXT_TOOL
+from .tools.knowledge_search import TOOL as KNOWLEDGE_SEARCH_TOOL
+
+TOOLS: tuple[ToolTemplate, ...] = (
+    ISSUE_CONTEXT_TOOL,
+    KNOWLEDGE_SEARCH_TOOL,
+    CHANGE_CONTEXT_TOOL,
+)
+TOOLS_BY_NAME = MappingProxyType({tool.name: tool for tool in TOOLS})
+TOOL_NAMES = tuple(tool.name for tool in TOOLS)
+
+
+def tool_declarations() -> list[dict[str, Any]]:
+    return [tool.declaration() for tool in TOOLS]

mcp_servers/project_context/runtime.py

@@ -0,0 +1,74 @@
+"""Fail-closed identity, policy, and provider resolution for the template server."""
+
+from __future__ import annotations
+
+import os
+import sys
+from collections.abc import Callable
+from dataclasses import dataclass
+from typing import Any
+
+from .foundation import IdentityContext, ProjectContextRuntime, ProviderError
+from .providers.change import build_provider as build_change_provider
+from .providers.issue import build_provider as build_issue_provider
+from .providers.knowledge import build_provider as build_knowledge_provider
+
+MINIMUM_PYTHON = (3, 11)
+
+
+def require_supported_python(version_info: tuple[int, ...] | None = None) -> None:
+    """Fail with an actionable message before the MCP server starts."""
+    current = version_info or tuple(sys.version_info[:3])
+    if current[:2] < MINIMUM_PYTHON:
+        raise RuntimeError(
+            "Project Context MCP requires Python 3.11 or newer; "
+            f"current runtime is {current[0]}.{current[1]}"
+        )
+
+
+@dataclass(frozen=True)
+class ProjectScopePolicy:
+    """Pilot policy: read scope and exact identity-bound project are both mandatory."""
+
+    def decide(self, identity: IdentityContext, tool_name: str, project_id: str) -> bool:
+        return "read" in identity.granted_scopes and project_id == identity.project
+
+
+PROVIDER_FACTORIES: dict[str, Callable[[IdentityContext], Any]] = {
+    "get_project_issue_context": build_issue_provider,
+    "search_project_knowledge": build_knowledge_provider,
+    "get_project_change_context": build_change_provider,
+}
+
+
+@dataclass(frozen=True)
+class ProjectProviderResolver:
+    def resolve(self, identity: IdentityContext, tool_name: str) -> Any:
+        factory = PROVIDER_FACTORIES.get(tool_name)
+        if factory is None:
+            raise ProviderError("NOT_FOUND", "The requested tool is not registered.", retryable=False)
+        return factory(identity)
+
+
+def _required_environment(name: str) -> str:
+    value = os.environ.get(name, "").strip()
+    if not value:
+        raise RuntimeError(f"Project Context MCP cannot start: required setting {name} is missing")
+    return value
+
+
+def default_runtime() -> ProjectContextRuntime:
+    """Build immutable runtime state; missing identity configuration fails at boot."""
+    require_supported_python()
+    identity = IdentityContext(
+        actor_id=_required_environment("COWORK_MCP_ACTOR_ID"),
+        org_unit=_required_environment("COWORK_MCP_ORG_UNIT"),
+        customer=_required_environment("COWORK_MCP_CUSTOMER"),
+        project=_required_environment("COWORK_MCP_PROJECT"),
+        granted_scopes=frozenset({"read"}),
+    )
+    return ProjectContextRuntime(
+        identity=identity,
+        policy=ProjectScopePolicy(),
+        credential_resolver=ProjectProviderResolver(),
+    )

mcp_servers/project_context/server.py

@@ -0,0 +1,142 @@
+"""Low-level MCP stdio adapter around the transport-agnostic Project Context core."""
+
+# ruff: noqa: UP045 -- Optional keeps the template importable with Pydantic on Python 3.9.
+
+from __future__ import annotations
+
+import json
+from typing import Any, Optional
+from uuid import uuid4
+
+from pydantic import ValidationError
+
+from .foundation import (
+    DispatchResult,
+    ProjectContextRuntime,
+    ProviderError,
+    error_result,
+)
+from .registry import TOOLS_BY_NAME, tool_declarations
+from .runtime import default_runtime, require_supported_python
+
+
+def dispatch(
+    name: str,
+    arguments: dict[str, Any],
+    runtime: ProjectContextRuntime,
+) -> DispatchResult:
+    """Validate → authorize → resolve provider → execute → validate output."""
+    correlation_id = str(uuid4())
+    tool = TOOLS_BY_NAME.get(name)
+    if tool is None:
+        return error_result(
+            "NOT_FOUND",
+            category="NOT_FOUND",
+            retryable=False,
+            message="The requested MCP tool is not registered.",
+            suggested_action="Refresh the tool list and choose one of the advertised tools.",
+            correlation_id=correlation_id,
+        )
+
+    try:
+        validated_input = tool.input_model.model_validate(arguments or {})
+    except ValidationError:
+        return error_result(
+            "INVALID_INPUT",
+            category="INVALID_INPUT",
+            retryable=False,
+            message="The tool arguments do not match the published input contract.",
+            suggested_action="Correct the required fields and value bounds, then call again.",
+            correlation_id=correlation_id,
+        )
+
+    project_id = str(validated_input.project_id)
+    if not runtime.policy.decide(runtime.identity, name, project_id):
+        return error_result(
+            "DENIED",
+            category="DENIED",
+            retryable=False,
+            message="The project is outside the caller's approved scope.",
+            suggested_action="Use an approved project or ask the project owner for access.",
+            correlation_id=correlation_id,
+        )
+
+    try:
+        provider = runtime.credential_resolver.resolve(runtime.identity, name)
+        raw_output = tool.handler(validated_input, provider)
+    except ProviderError as exc:
+        return error_result(
+            exc.code,
+            category=exc.code,
+            retryable=exc.retryable,
+            message=exc.safe_message,
+            suggested_action="Check the approved provider configuration and retry if allowed.",
+            correlation_id=correlation_id,
+        )
+    except Exception:  # noqa: BLE001 - provider failures must not crash or leak into the agent turn
+        return error_result(
+            "UPSTREAM_ERROR",
+            category="UPSTREAM_ERROR",
+            retryable=False,
+            message="The approved provider could not complete the request.",
+            suggested_action="Check the correlation ID in server logs; do not resend credentials.",
+            correlation_id=correlation_id,
+        )
+
+    try:
+        output_with_trace = {**raw_output, "correlation_id": correlation_id}
+        validated_output = tool.output_model.model_validate(output_with_trace)
+    except ValidationError:
+        return error_result(
+            "UPSTREAM_ERROR",
+            category="UPSTREAM_ERROR",
+            retryable=False,
+            message="The provider response did not match the published output contract.",
+            suggested_action="Fix the provider mapping before retrying the request.",
+            correlation_id=correlation_id,
+        )
+    return DispatchResult(ok=True, payload=validated_output.model_dump(mode="json"))
+
+
+def build_server(runtime: Optional[ProjectContextRuntime] = None):
+    from mcp import types
+    from mcp.server.lowlevel import Server
+
+    require_supported_python()
+    app_runtime = runtime or default_runtime()
+    app = Server("project_context")
+
+    @app.list_tools()
+    async def list_tools() -> list[types.Tool]:
+        return [types.Tool(**declaration) for declaration in tool_declarations()]
+
+    @app.call_tool()
+    async def call_tool(name: str, arguments: dict[str, Any]) -> types.CallToolResult:
+        result = dispatch(name, arguments or {}, app_runtime)
+        return types.CallToolResult(
+            content=[types.TextContent(
+                type="text",
+                text=json.dumps(result.payload, ensure_ascii=False, separators=(",", ":")),
+            )],
+            structuredContent=result.payload if result.ok else None,
+            isError=not result.ok,
+        )
+
+    return app
+
+
+def main() -> None:
+    import anyio
+    from mcp.server.stdio import stdio_server
+
+    app = build_server()
+
+    async def _run() -> None:
+        async with stdio_server() as (read, write):
+            await app.run(read, write, app.create_initialization_options())
+
+    anyio.run(_run)
+
+
+if __name__ == "__main__":
+    main()

mcp_servers/project_context/tools/__init__.py

@@ -0,0 +1 @@
+"""Independent tool modules; ownership is documented in the team guide."""

mcp_servers/project_context/tools/change_context.py

@@ -0,0 +1,55 @@
+"""Member C work package: get_project_change_context."""
+
+# ruff: noqa: UP045 -- Optional keeps Pydantic model evaluation compatible with Python 3.9.
+
+from __future__ import annotations
+
+from typing import Any, Literal, Optional
+
+from pydantic import Field
+
+from ..foundation import ContractModel, SourceCitation, ToolTemplate
+
+
+class ChangeContextInput(ContractModel):
+    project_id: str = Field(min_length=1, max_length=128)
+    change_id: str = Field(min_length=1, max_length=128)
+    detail: Literal["summary", "standard", "full"] = "standard"
+    cursor: Optional[str] = Field(default=None, max_length=2048)
+
+
+class ChangeContextOutput(ContractModel):
+    correlation_id: str
+    project_id: str
+    change_id: str
+    change_type: Literal["commit", "pull-request", "merge-request"]
+    title: str
+    state: str
+    summary: str
+    authors: tuple[str, ...]
+    files: tuple[str, ...]
+    commits: tuple[str, ...]
+    related_issues: tuple[str, ...]
+    source: SourceCitation
+    truncated: bool
+    returned: int = Field(ge=0)
+    remaining: int = Field(ge=0)
+    next_cursor: Optional[str] = None
+
+
+def _handle(arguments: ContractModel, provider: Any) -> dict[str, Any]:
+    request = ChangeContextInput.model_validate(arguments)
+    return provider.get_change_context(**request.model_dump())
+
+
+TOOL = ToolTemplate(
+    name="get_project_change_context",
+    description=(
+        "Returns provider-neutral context for one authorized commit, pull request, or merge request "
+        "with changed files, commits, related issues, and a pinned source. Use when an exact change "
+        "identifier is known. Do not use for issue details or free-text document search."
+    ),
+    input_model=ChangeContextInput,
+    output_model=ChangeContextOutput,
+    handler=_handle,
+)

mcp_servers/project_context/tools/issue_context.py

@@ -0,0 +1,59 @@
+"""Member A work package: get_project_issue_context."""
+
+# ruff: noqa: UP045 -- Optional keeps Pydantic model evaluation compatible with Python 3.9.
+
+from __future__ import annotations
+
+from typing import Any, Literal, Optional
+
+from pydantic import Field
+
+from ..foundation import ContractModel, SourceCitation, ToolTemplate
+
+
+class IssueContextInput(ContractModel):
+    project_id: str = Field(min_length=1, max_length=128)
+    issue_key: str = Field(min_length=1, max_length=128)
+    detail: Literal["summary", "standard", "full"] = "standard"
+    cursor: Optional[str] = Field(default=None, max_length=2048)
+
+
+class RelatedItem(ContractModel):
+    item_id: str
+    relation: str
+    title: str
+    url: str
+
+
+class IssueContextOutput(ContractModel):
+    correlation_id: str
+    project_id: str
+    issue_key: str
+    title: str
+    status: str
+    description: str
+    acceptance_criteria: tuple[str, ...]
+    related: tuple[RelatedItem, ...]
+    source: SourceCitation
+    truncated: bool
+    returned: int = Field(ge=0)
+    remaining: int = Field(ge=0)
+    next_cursor: Optional[str] = None
+
+
+def _handle(arguments: ContractModel, provider: Any) -> dict[str, Any]:
+    request = IssueContextInput.model_validate(arguments)
+    return provider.get_issue_context(**request.model_dump())
+
+
+TOOL = ToolTemplate(
+    name="get_project_issue_context",
+    description=(
+        "Returns one authorized work item's title, state, description, acceptance criteria, "
+        "related items, and pinned source. Use when an exact issue key is known. Do not use for "
+        "free-text knowledge search or Git change review."
+    ),
+    input_model=IssueContextInput,
+    output_model=IssueContextOutput,
+    handler=_handle,
+)

mcp_servers/project_context/tools/knowledge_search.py

@@ -0,0 +1,58 @@
+"""Member B work package: search_project_knowledge."""
+
+# ruff: noqa: UP045 -- Optional keeps Pydantic model evaluation compatible with Python 3.9.
+
+from __future__ import annotations
+
+from typing import Any, Literal, Optional
+
+from pydantic import Field
+
+from ..foundation import ContractModel, SourceCitation, ToolTemplate
+
+
+class KnowledgeSearchInput(ContractModel):
+    project_id: str = Field(min_length=1, max_length=128)
+    query: str = Field(min_length=2, max_length=1000)
+    detail: Literal["summary", "standard", "full"] = "standard"
+    top_k: int = Field(default=5, ge=1, le=20)
+    language: Optional[Literal["en", "ja", "vi"]] = None
+    cursor: Optional[str] = Field(default=None, max_length=2048)
+
+
+class KnowledgeItem(ContractModel):
+    document_id: str
+    chunk_id: str
+    title: str
+    excerpt: str
+    score: float = Field(ge=0, le=1)
+    source: SourceCitation
+
+
+class KnowledgeSearchOutput(ContractModel):
+    correlation_id: str
+    project_id: str
+    query: str
+    items: tuple[KnowledgeItem, ...]
+    truncated: bool
+    returned: int = Field(ge=0)
+    remaining: int = Field(ge=0)
+    next_cursor: Optional[str] = None
+
+
+def _handle(arguments: ContractModel, provider: Any) -> dict[str, Any]:
+    request = KnowledgeSearchInput.model_validate(arguments)
+    return provider.search_knowledge(**request.model_dump())
+
+
+TOOL = ToolTemplate(
+    name="search_project_knowledge",
+    description=(
+        "Searches approved knowledge for one authorized project and returns ranked excerpts with "
+        "pinned citations. Use for requirements, design notes, or runbooks when no exact issue is "
+        "known. Do not use for issue details or Git change review."
+    ),
+    input_model=KnowledgeSearchInput,
+    output_model=KnowledgeSearchOutput,
+    handler=_handle,
+)

mcp_servers/project_context_server.py

@@ -0,0 +1,9 @@
+"""Stable module entry point for ``python -m cowork_local.mcp_servers.project_context_server``."""
+
+from .project_context.server import build_server, dispatch, main
+
+__all__ = ["build_server", "dispatch", "main"]
+
+
+if __name__ == "__main__":
+    main()

presentation/__init__.py

@@ -0,0 +1 @@
+"""Presentation Layer: PySide6 UI widgets, dialogs, and shell views (<400 LOC per file)."""

presentation/chat/__init__.py

@@ -0,0 +1 @@
+"""Presentation chat package: ChatHistoryWidget, ComposerWidget, AttachmentPicker, AudioRecorderWidget, ChatOutputPanel."""

presentation/co4e/__init__.py

@@ -0,0 +1 @@
+"""Presentation Co4E package: Co4ECanvasWidget, NodePropertyPanel, RunControlWidget, Co4EChatView."""

presentation/dashboard/__init__.py

@@ -0,0 +1 @@
+"""Presentation dashboard package: TokenUsageCardWidget, UsageChartWidget, HabitsWidget."""

presentation/folder/__init__.py

@@ -0,0 +1 @@
+"""Presentation folder package: WorkspaceFileTree, DocumentPreviewManager, AiFileEditorDialog."""

presentation/graph/__init__.py

@@ -0,0 +1 @@
+"""Presentation graph package: StructureGraphView and GraphQaWidget."""

presentation/monitoring/__init__.py

@@ -0,0 +1 @@
+"""Presentation monitoring package: 8 modular sub-tab widgets."""

presentation/scheduling/__init__.py

@@ -0,0 +1 @@
+"""Presentation scheduling package: KanbanBoardWidget, CalendarViewWidget, AiTaskCreatorDialog."""

presentation/settings/__init__.py

@@ -0,0 +1 @@
+"""Presentation settings package: Section widgets for provider, connector, routing, and general settings."""

presentation/shell/__init__.py

@@ -0,0 +1 @@
+"""Presentation shell package: MainWindow shell, TrayManager, LifecycleCoordinator."""

providers/anthropic.py

@@ -292,19 +292,21 @@ class AnthropicProvider(Provider):
                 args = {"_raw": b["json"]}
             tool_calls.append({"id": b["id"], "name": b["name"], "arguments": args})
 
-        # Dashboard usage event — real counts from the stream's usage events
-        # (input arrives on message_start, output on message_delta), else a
-        # ~4 chars/token estimate. Delivery is the sink's job (R03-T06), so this
-        # only translates Anthropic's wire shape into a canonical UsageEvent.
-        from ..infrastructure.telemetry import usage_sink as telemetry
+        # Dashboard usage event — real counts from the stream's usage events,
+        # else a ~4 chars/token estimate. Never breaks the turn.
+        try:
+            from ..core import usage_tracker as ut
 
-        if usage_seen:
-            event = telemetry.anthropic_usage_event(self.name, self.model, usage_seen)
-        else:
-            sent = json.dumps(payload.get("messages", []), ensure_ascii=False)
-            got = "".join(text_parts) + "".join(b["json"] for b in blocks.values())
-            event = telemetry.estimated_event(self.name, self.model, sent, got)
-        self._emit_usage(event)
+            if usage_seen:
+                ut.record(self.name, self.model, usage_seen.get("in", 0),
+                          usage_seen.get("out", 0), usage_seen.get("cache", 0))
+            else:
+                sent = json.dumps(payload.get("messages", []), ensure_ascii=False)
+                got = "".join(text_parts) + "".join(b["json"] for b in blocks.values())
+                ut.record(self.name, self.model, ut.estimate_tokens(sent),
+                          ut.estimate_tokens(got), 0, estimated=True)
+        except Exception:  # noqa: BLE001
+            pass
 
         return {"role": "assistant", "content": "".join(text_parts), "tool_calls": tool_calls}
 

providers/base.py

@@ -224,12 +224,6 @@ class Provider:
         # silently swallowing the error — Settings' "Test connection" / "Load
         # models" surfaces this so "model won't load" has a concrete reason.
         self.last_error = ""
-        # Where this provider's token usage goes (R03-T06). None means "the
-        # process-wide default sink", resolved lazily in _emit_usage so that a
-        # test can swap the destination without rebuilding every provider.
-        # Set it per instance to bill one run somewhere else (a workflow, a
-        # scheduled task) without touching global state.
-        self.usage_sink = None
 
     def chat(
         self,
@@ -280,24 +274,6 @@ class Provider:
             return True, f"OK — {len(models)} model(s) available."
         return False, "No models returned. Check base_url/API key and network access."
 
-    # -- telemetry -----------------------------------------------------
-    def _emit_usage(self, event) -> None:
-        """Hand one ``UsageEvent`` to this provider's usage sink.
-
-        Never raises: recording how many tokens a turn cost must not be able to
-        fail the turn itself. Falls back to the process-wide default sink so
-        existing call sites keep reporting to the Dashboard exactly as before
-        (see infrastructure/telemetry/usage_sink.py)."""
-        try:
-            sink = self.usage_sink
-            if sink is None:
-                from ..infrastructure.telemetry import usage_sink as telemetry
-
-                sink = telemetry.default_sink
-            sink.record(event)
-        except Exception:  # noqa: BLE001 — telemetry is never worth a failed turn
-            pass
-
     # -- shared helpers ------------------------------------------------
     @staticmethod
     def _is_cancelled(cancel) -> bool:

providers/openai_compat.py

@@ -268,25 +268,22 @@ class OpenAICompatProvider(Provider):
     def _record_usage(self, messages, text_parts, tool_acc, usage_seen) -> None:
         """One Dashboard usage event per turn: real counts when the server's
         final chunk carried a "usage" block, a ~4 chars/token estimate
-        otherwise.
+        otherwise. Never breaks the turn."""
+        try:
+            from ..core import usage_tracker as ut
 
-        Building the event and delivering it are now separate concerns (R03-T06):
-        this method only translates THIS provider's wire shape into a canonical
-        ``UsageEvent``; where it ends up is the sink's decision, so a test can
-        assert on token counts without writing to the real Dashboard store."""
-        from ..infrastructure.telemetry import usage_sink as telemetry
-
-        if usage_seen:
-            event = telemetry.openai_usage_event(self.name, self.model, usage_seen)
-        else:
-            # No usage block from the gateway (self-hosted servers and Ollama
-            # never send one) - fall back to estimating from the raw text of
-            # both directions, tool-call arguments included since the model was
-            # billed for generating them.
-            sent = json.dumps(self._to_api_messages(messages), ensure_ascii=False)
-            got = "".join(text_parts) + "".join(s["args"] for s in tool_acc.values())
-            event = telemetry.estimated_event(self.name, self.model, sent, got)
-        self._emit_usage(event)
+            if usage_seen:
+                ut.record(self.name, self.model,
+                          usage_seen.get("prompt_tokens", 0),
+                          usage_seen.get("completion_tokens", 0),
+                          (usage_seen.get("prompt_tokens_details") or {}).get("cached_tokens", 0))
+            else:
+                sent = json.dumps(self._to_api_messages(messages), ensure_ascii=False)
+                got = "".join(text_parts) + "".join(s["args"] for s in tool_acc.values())
+                ut.record(self.name, self.model, ut.estimate_tokens(sent),
+                          ut.estimate_tokens(got), 0, estimated=True)
+        except Exception:  # noqa: BLE001
+            pass
 
     def list_models(self):
         self.last_error = ""

requirements (cloud copy).txt

@@ -1,9 +0,0 @@
-PySide6>=6.6
-pydantic>=2
-requests
-psutil
-pygments
-openpyxl
-python-pptx
-networkx
-pytest

scripts/check_imports.py

@@ -1,237 +1,164 @@
-#!/usr/bin/env python3
-"""CASAN Check 3 — Clean Architecture Guard (R01-T03).
+"""AST-based Static Analysis Guard for Clean Architecture Enforcement.
 
-Statically walks the AST of every Python file in the pure-Python layers and
-fails when a file imports something the layer is not allowed to depend on.
-
-Why AST instead of ``grep``: a regex over source text cannot tell an import
-apart from the same words appearing inside a docstring, a comment or a string
-literal (this repo has several docstrings that legitimately mention
-``PySide6``). ``ast`` sees only real ``import`` / ``from … import`` nodes, so
-the check has no false positives and needs no ``# noqa`` escape hatches.
-
-Rules enforced (see docs/architecture/ADR-001-layered-architecture.md):
-
-* **I1** ``domain/`` and ``application/`` must be 100% pure Python — no Qt.
-* **I2** ``domain/`` must not import ``application/``, ``infrastructure/``,
-  ``presentation/`` or the legacy ``ui/``.
-* **I3** ``application/`` must not import ``presentation/`` or ``ui/``.
-
-Usage::
-
-    python scripts/check_imports.py            # scan the whole repo
-    python scripts/check_imports.py domain     # scan one layer only
-
-Exit code is 0 when clean and 1 when at least one violation is found, so it
-can be wired straight into CI / ``scripts/run_quality_gate.py`` (R10-T02).
+Scans designated Python packages (such as `domain/` and `application/`) to ensure
+they remain 100% Pure Python and do not import presentation/GUI frameworks (PySide6, PyQt)
+or concrete application shells.
 """
 from __future__ import annotations
 
 import argparse
 import ast
+import io
 import sys
-from dataclasses import dataclass
 from pathlib import Path
-from typing import Dict, Iterable, List, Sequence, Tuple
+from typing import List, NamedTuple, Set
 
-# Repository root = parent of this scripts/ folder. Everything below is resolved
-# relative to it so the checker works no matter what the checkout folder is
-# named or which directory the developer runs it from.
-REPO_ROOT = Path(__file__).resolve().parents[1]
+# Ensure UTF-8 output on standard console streams across diverse Windows locales (CP932, etc.)
+if sys.stdout.encoding and sys.stdout.encoding.lower() not in ("utf-8", "utf8"):
+    try:
+        sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8", errors="replace")
+        sys.stderr = io.TextIOWrapper(sys.stderr.buffer, encoding="utf-8", errors="replace")
+    except Exception:
+        pass
 
-# The distribution package name. Absolute imports may be written either as
-# ``from cowork_local.ui import x`` or ``from ui import x`` depending on how the
-# module was reached; we normalise the prefix away so both spellings are caught.
-PACKAGE_NAME = "cowork_local"
 
-# Any import whose first dotted segment is one of these is a GUI toolkit.
-QT_ROOTS = frozenset({"PySide6", "PySide2", "PyQt5", "PyQt6", "shiboken6", "shiboken2"})
+class ImportViolation(NamedTuple):
+    file_path: Path
+    line_number: int
+    imported_module: str
+    rule_description: str
 
-# Per-layer rules: layer directory -> top-level package names it may not import.
-# Kept as a plain table so adding a layer later is a one-line change and the
-# rules stay readable next to the ADR they implement.
-LAYER_RULES: Dict[str, frozenset] = {
-    # I1 + I2: domain is the innermost layer and depends on nothing but stdlib.
-    "domain": frozenset({"application", "infrastructure", "presentation", "ui", "core"}),
-    # I1 + I3: application may use domain, but never anything that draws pixels.
-    "application": frozenset({"presentation", "ui"}),
+
+# Disallowed top-level package names in pure business/domain layers
+FORBIDDEN_MODULE_PREFIXES: Set[str] = {
+    "PySide6",
+    "PySide2",
+    "PyQt6",
+    "PyQt5",
+    "ui",
+    "app",
 }
 
-# Directories that are never production code and therefore never scanned.
-SKIP_DIRS = frozenset({".git", "__pycache__", ".pytest_cache", "tests", "build", "dist"})
+# Default directories that must strictly adhere to Clean Architecture
+DEFAULT_SCAN_DIRS: List[str] = [
+    "domain",
+    "application",
+]
 
 
-@dataclass(frozen=True)
-class Violation:
-    """One forbidden import, carrying enough context to fix it without grepping."""
+class ArchitectureImportVisitor(ast.NodeVisitor):
+    """AST visitor that checks all Import and ImportFrom statements against forbidden prefixes."""
 
-    path: Path
-    line: int
-    imported: str
-    rule: str
+    def __init__(self, file_path: Path, forbidden: Set[str]) -> None:
+        self.file_path = file_path
+        self.forbidden = forbidden
+        self.violations: List[ImportViolation] = []
 
-    def render(self) -> str:
-        """Format as ``file:line: message`` — the shape editors turn into a
-        clickable link, so a CI failure lands the developer on the exact line."""
-        rel = self.path.relative_to(REPO_ROOT).as_posix()
-        # ASCII-only on purpose: this line is printed to a console that may run a
-        # legacy code page (cp932 on the team's Windows boxes), where a non-ASCII
-        # dash raises UnicodeEncodeError and would crash the gate on the very
-        # failure path it exists to report.
-        return f"{rel}:{self.line}: imports '{self.imported}' - {self.rule}"
+    def visit_Import(self, node: ast.Import) -> None:
+        # Check direct `import x, y` statements
+        for alias in node.names:
+            root_module = alias.name.split(".")[0]
+            if root_module in self.forbidden:
+                self.violations.append(
+                    ImportViolation(
+                        file_path=self.file_path,
+                        line_number=node.lineno,
+                        imported_module=alias.name,
+                        rule_description=f"Direct import of GUI/shell module '{alias.name}' is prohibited.",
+                    )
+                )
+        self.generic_visit(node)
+
+    def visit_ImportFrom(self, node: ast.ImportFrom) -> None:
+        # Check `from x import y` statements
+        if node.module:
+            root_module = node.module.split(".")[0]
+            if root_module in self.forbidden:
+                self.violations.append(
+                    ImportViolation(
+                        file_path=self.file_path,
+                        line_number=node.lineno,
+                        imported_module=node.module,
+                        rule_description=f"Import from GUI/shell module '{node.module}' is prohibited.",
+                    )
+                )
+        self.generic_visit(node)
 
 
-def iter_python_files(layer_dir: Path) -> Iterable[Path]:
-    """Yield every production ``.py`` file under ``layer_dir``.
-
-    Test files are excluded on purpose: a test for a pure-Python service is
-    allowed to import Qt (an integration test may need a headless widget), and
-    holding tests to the production rule would push people to disable the gate.
-    """
-    if not layer_dir.is_dir():
-        return
-    for path in sorted(layer_dir.rglob("*.py")):
-        # Reject a path as soon as ANY of its parent folder names is skippable,
-        # which also covers nested __pycache__ inside a sub-package.
-        if any(part in SKIP_DIRS for part in path.parts):
-            continue
-        yield path
-
-
-def module_parts(path: Path) -> List[str]:
-    """Dotted package path of ``path`` relative to the repo root, as a list.
-
-    ``domain/agents/agent_event.py``   -> ``["domain", "agents", "agent_event"]``
-    ``domain/agents/__init__.py``      -> ``["domain", "agents"]``
-
-    Needed to resolve *relative* imports: ``from ..models import X`` inside
-    ``domain/agents/foo.py`` really means ``domain.models``, and only the file's
-    own position tells us that.
-    """
-    rel = path.relative_to(REPO_ROOT)
-    parts = list(rel.parts)
-    if parts[-1] == "__init__.py":
-        parts.pop()
-    else:
-        parts[-1] = parts[-1][: -len(".py")]
-    return parts
-
-
-def resolve_relative(parts: Sequence[str], level: int, module: str) -> str:
-    """Turn a relative import into the absolute top-level package it points at.
-
-    ``level`` is the number of leading dots. Level 1 means "the package this
-    module lives in", so we drop the module's own name plus ``level - 1``
-    further parents. Returns the FIRST segment of the resolved path, because
-    the rules are expressed in terms of top-level layers.
-
-    Walking off the top of the tree (more dots than there are parents) yields
-    an empty string, which simply never matches a rule — a malformed import
-    like that is a syntax/packaging problem, not an architecture violation.
-    """
-    base = list(parts[:-1])  # the package containing this module
-    if level > 1:
-        drop = level - 1
-        if drop > len(base):
-            return ""
-        base = base[: len(base) - drop]
-    tail = module.split(".") if module else []
-    resolved = base + tail
-    return resolved[0] if resolved else ""
-
-
-def top_level(name: str) -> str:
-    """First dotted segment of an absolute import, with the distribution package
-    prefix stripped so ``cowork_local.ui.chat_panel`` and ``ui.chat_panel`` are
-    treated as the same dependency."""
-    segments = name.split(".")
-    if segments and segments[0] == PACKAGE_NAME:
-        segments = segments[1:]
-    return segments[0] if segments else ""
-
-
-def imported_roots(tree: ast.AST, parts: Sequence[str]) -> Iterable[Tuple[str, int, str]]:
-    """Yield ``(top_level_package, line_number, as_written)`` for every import.
-
-    ``as_written`` is kept so the error message shows what the developer
-    actually typed rather than the normalised root, which makes the violation
-    obvious at a glance.
-
-    ``ast.walk`` (not just the module body) is deliberate: this repo defers many
-    heavy imports into function bodies to keep app start-up fast, and a
-    function-local ``from PySide6 import QtWidgets`` breaks the layer exactly
-    the same way a top-level one does.
-    """
-    for node in ast.walk(tree):
-        if isinstance(node, ast.Import):
-            for alias in node.names:
-                yield top_level(alias.name), node.lineno, alias.name
-        elif isinstance(node, ast.ImportFrom):
-            if node.level:
-                written = "." * node.level + (node.module or "")
-                yield resolve_relative(parts, node.level, node.module or ""), node.lineno, written
-            else:
-                module = node.module or ""
-                yield top_level(module), node.lineno, module
-
-
-def check_file(path: Path, layer: str, banned: frozenset) -> List[Violation]:
-    """Collect every rule violation in one file.
-
-    A file that cannot be parsed is reported as a violation rather than skipped:
-    silently passing a file the checker could not read would make the gate lie.
-    """
+def scan_file(file_path: Path, forbidden: Set[str]) -> List[ImportViolation]:
+    """Parse a single Python file into AST and return all detected architecture import violations."""
     try:
-        tree = ast.parse(path.read_text(encoding="utf-8"), filename=str(path))
+        source_code = file_path.read_text(encoding="utf-8")
+        tree = ast.parse(source_code, filename=str(file_path))
     except (SyntaxError, UnicodeDecodeError) as exc:
-        return [Violation(path, getattr(exc, "lineno", 0) or 0, "<unparseable>",
-                          f"cannot be parsed by the architecture guard ({exc})")]
+        print(f"[Syntax/Read Warning] Could not parse {file_path}: {exc}", file=sys.stderr)
+        return []
 
-    parts = module_parts(path)
-    out: List[Violation] = []
-    for root, lineno, written in imported_roots(tree, parts):
-        if root in QT_ROOTS:
-            out.append(Violation(path, lineno, written,
-                                 f"'{layer}/' must be 100% pure Python (ADR-001 I1)"))
-        elif root in banned:
-            out.append(Violation(path, lineno, written,
-                                 f"'{layer}/' must not depend on '{root}/' (ADR-001 I2/I3)"))
-    return out
+    visitor = ArchitectureImportVisitor(file_path, forbidden)
+    visitor.visit(tree)
+    return visitor.violations
 
 
-def run(layers: Sequence[str]) -> List[Violation]:
-    """Scan the requested layers and return every violation found, in file order."""
-    found: List[Violation] = []
-    for layer in layers:
-        banned = LAYER_RULES[layer]
-        for path in iter_python_files(REPO_ROOT / layer):
-            found.extend(check_file(path, layer, banned))
-    return found
+def scan_directory(dir_path: Path, forbidden: Set[str]) -> List[ImportViolation]:
+    """Recursively scan all Python files in a directory."""
+    violations: List[ImportViolation] = []
+    if not dir_path.exists():
+        return violations
+
+    for py_file in dir_path.rglob("*.py"):
+        if py_file.is_file() and "__pycache__" not in py_file.parts:
+            violations.extend(scan_file(py_file, forbidden))
+
+    return violations
 
 
-def main(argv: Sequence[str] | None = None) -> int:
+def main() -> int:
+    """CLI entry point for CI/pre-commit quality gate checks."""
     parser = argparse.ArgumentParser(
-        description="CASAN Check 3 - Clean Architecture Guard (see ADR-001).")
-    parser.add_argument(
-        "layers", nargs="*", choices=sorted(LAYER_RULES) or None, default=None,
-        help="Layers to scan (default: every layer with a rule).",
+        description="Clean Architecture Import Guard: Verifies zero GUI/Qt dependencies in domain/app layers."
     )
-    args = parser.parse_args(argv)
-    layers = args.layers or sorted(LAYER_RULES)
+    parser.add_argument(
+        "--paths",
+        nargs="*",
+        default=DEFAULT_SCAN_DIRS,
+        help="Paths or directories to scan (defaults to 'domain' and 'application')",
+    )
+    parser.add_argument(
+        "--root",
+        default=".",
+        help="Root workspace directory",
+    )
+    args = parser.parse_args()
 
-    violations = run(layers)
-    scanned = sum(1 for layer in layers for _ in iter_python_files(REPO_ROOT / layer))
+    root_dir = Path(args.root).resolve()
+    all_violations: List[ImportViolation] = []
 
-    if violations:
-        print(f"FAIL - {len(violations)} architecture violation(s) in {scanned} file(s):\n")
-        for v in violations:
-            print("  " + v.render())
-        # Point at the rationale instead of just the rule id, so someone hitting
-        # this for the first time knows where the decision was made.
-        print("\nSee docs/architecture/ADR-001-layered-architecture.md")
+    print(f"[Clean Arch Guard] Scanning root: {root_dir}")
+
+    for target in args.paths:
+        target_path = (root_dir / target).resolve()
+        if not target_path.exists():
+            # If the layer directory does not exist yet (during early migration), skip cleanly
+            print(f"[Clean Arch Guard] Directory '{target}' does not exist yet (skipped).")
+            continue
+
+        if target_path.is_file():
+            all_violations.extend(scan_file(target_path, FORBIDDEN_MODULE_PREFIXES))
+        else:
+            all_violations.extend(scan_directory(target_path, FORBIDDEN_MODULE_PREFIXES))
+
+    if all_violations:
+        print("\n[FAIL] CLEAN ARCHITECTURE VIOLATIONS DETECTED:")
+        print("=" * 70)
+        for v in all_violations:
+            rel_path = v.file_path.relative_to(root_dir) if v.file_path.is_relative_to(root_dir) else v.file_path
+            print(f"  • {rel_path}:{v.line_number} -> Forbidden import: '{v.imported_module}'")
+            print(f"    Reason: {v.rule_description}")
+        print("=" * 70)
+        print(f"Total Violations: {len(all_violations)}")
         return 1
 
-    print(f"PASS - 0 Qt imports in {', '.join(layers)} ({scanned} file(s) scanned)")
+    print("\n[PASS] CLEAN ARCHITECTURE CHECK: 0 forbidden imports detected.")
     return 0
 
 

state.py

@@ -6,7 +6,6 @@ import time
 from typing import TYPE_CHECKING, Optional, Tuple
 
 from .config import AppConfig
-from .infrastructure.mcp import McpToolSourceManager
 
 
 def resolve_agent_default(
@@ -39,30 +38,21 @@ class AppContext:
     def __init__(self, config: AppConfig):
         self.config = config
         self.started_at = time.time()   # for Monitoring's Sandbox Details "Created"/"Uptime"
-        # Admin-configured MCP servers + the built-in MS365 server (R05-T05):
-        # connection caching/lifecycle (check-then-create, restart, shutdown)
-        # now lives in McpToolSourceManager, extracted so it is testable
-        # without an AppContext/Qt. See its docstring for why the check-then-
-        # create race matters — several turns (multiple Cowork tabs, parallel
-        # Co4E flows, scheduled tasks) can call build_mcp_tools() at once.
-        self._mcp_manager = McpToolSourceManager()
+        self._mcp_connections: dict = {}  # server name -> McpServerConnection
         self._ext_connections: dict = {}  # connector id -> McpServerConnection (mcp_stdio mode only)
-        # Guards ``_ext_connections`` only now — unified Connectors (CAD/CAE/
-        # MS365/Other) aren't covered by McpToolSourceManager (R05-T05 scoped
-        # to MCP servers), so this cache still needs its own check-then-create
-        # lock, the same race McpToolSourceManager guards against internally.
+        # Guards the two connection caches above. build_mcp_tools() runs on EVERY
+        # chat turn's own AgentWorker thread, so several turns (multiple Cowork
+        # tabs, parallel Co4E flows, scheduled tasks) can enter it at once. The
+        # cache is populated check-then-create ("conn is None → spawn → store");
+        # without this lock two concurrent turns both see None and each spawns a
+        # subprocess for the SAME server — one leaks as an orphan and the wrong
+        # object may be handed out. The lock makes connection setup atomic; the
+        # provider/HTTP path itself is already thread-safe (a fresh provider per
+        # call, module-level `requests`, MCP calls multiplexed on the server's
+        # own event loop), so concurrent model calls never needed serializing.
         self._conn_lock = threading.Lock()
         self._routing_service = None       # lazy RoutingService (Auto Model Routing)
-        # Lazy RoutingApplicationService (R03-T03) — the Qt-free decision layer
-        # every chat surface now routes through. Wraps _routing_service, which
-        # stays the scoring/ranking engine underneath.
-        self._routing_application = None
         self._routing_lock = threading.Lock()
-        # A SEPARATE lock for the application service: building it calls
-        # routing(), which takes _routing_lock. threading.Lock is not
-        # reentrant, so sharing one lock across both accessors deadlocks the
-        # first caller instead of just serialising them.
-        self._routing_app_lock = threading.Lock()
         # The workspace (project) currently selected in the Workspace screen.
         # Per-workspace modes (routing + auto-run) resolve against THIS project
         # so each workspace keeps its own modes. Updated by WorkspaceTab on
@@ -89,28 +79,16 @@ class AppContext:
         workspace keep its own routing mode."""
         project = self._current_project()
         if project is not None:
-            # Validated through the single mode vocabulary (R03-T03) rather
-            # than a literal tuple, so a workspace can store any mode the
-            # routing service understands - including "fallback", whose
-            # on-screen toggle arrives in EPIC R08.
-            from .application.model_routing import is_valid_mode, normalize_mode
-
             mode = (project.routing_modes or {}).get(surface, "")
-            # Only a RECOGNISED override wins; an empty or corrupt value falls
-            # through to the global setting, exactly as before. Validation goes
-            # through the routing vocabulary (R03-T03) instead of a literal
-            # tuple, so a new mode works everywhere the moment it is defined.
-            if is_valid_mode(mode):
-                return normalize_mode(mode)
+            if mode in ("off", "auto", "manual"):
+                return mode
         return self.config.routing_mode_for(surface)
 
     def set_project_routing_mode(self, surface: str, mode: str) -> None:
         """Persist a surface's routing mode for the ACTIVE workspace. With no
         workspace selected, falls back to the global setting so behaviour
         outside a project stays global."""
-        from .application.model_routing import normalize_mode
-
-        mode = normalize_mode(mode)
+        mode = mode if mode in ("off", "auto", "manual") else "off"
         project = self._current_project()
         if project is None:
             self.config.set_routing_mode_for(surface, mode)
@@ -166,34 +144,6 @@ class AppContext:
                     self._routing_service = RoutingService(self)
         return self._routing_service
 
-    def routing_application(self):
-        """The shared :class:`RoutingApplicationService` (R03-T03).
-
-        This is what UI code should call: it owns the Off/Auto/Manual/Fallback
-        policy, the confirm handshake and the never-raise guarantee, while
-        :meth:`routing` remains the scoring engine underneath. Chat, Co4E and
-        AI-Edit all go through this one object, so a change to routing policy is
-        made once instead of three times.
-
-        Built lazily and memoised for the same reason as :meth:`routing`: the
-        pending-switch registry and assessment store must be shared app-wide."""
-        if self._routing_application is None:
-            # Resolve the engine BEFORE taking this lock: routing() takes
-            # _routing_lock, and nesting the two acquisitions is what makes the
-            # ordering fragile in the first place.
-            engine = self.routing()
-            with self._routing_app_lock:
-                if self._routing_application is None:
-                    from .application.model_routing import RoutingApplicationService
-
-                    self._routing_application = RoutingApplicationService(
-                        engine,
-                        # Per-workspace mode lookup, so each workspace keeps its
-                        # own routing behaviour (see project_routing_mode).
-                        mode_reader=self.project_routing_mode,
-                    )
-        return self._routing_application
-
     def build_active_provider(self):
         """Construct the currently selected provider (called inside workers)."""
         return self.build_provider_for(self.config.active_provider)
@@ -238,41 +188,48 @@ class AppContext:
         if not self.config.connect_external:
             return [], None
         from .core.ext_connectors import build_ext_connector_tools
+        from .core.mcp_client import McpServerConnection
         from .core.mcp_client import build_mcp_tools as _merge_mcp_tools
         from .core.tools import combine_tool_sources
 
-        # R05-T05: connection caching/check-then-create for admin-configured
-        # servers + the MS365 builtin now lives in McpToolSourceManager (its
-        # own lock guards the race — see its docstring).
-        active = []
-        for entry in self.config.mcp_servers:
-            if not entry.get("enabled", True):
-                continue
-            name = entry.get("name", "")
-            command = entry.get("command", "")
-            if not name or not command:
-                continue
-            conn = self._mcp_manager.ensure(name, command, entry.get("args") or [],
-                                            entry.get("env") or None)
-            if conn is not None:
-                active.append(conn)
-        builtin = self._ms365_builtin_connection(skip={c.name for c in active})
-        if builtin is not None:
-            active.append(builtin)
-        mcp_tools, mcp_executor = _merge_mcp_tools(active)
-
-        # ``_ext_connections`` isn't covered by McpToolSourceManager (T05
-        # scoped to MCP servers) — still serialized under ``_conn_lock``.
+        # Serialize the check-then-create against the connection caches so
+        # concurrent turns share one subprocess per server instead of racing to
+        # spawn duplicates (see _conn_lock in __init__). The lock is held while
+        # connections are established (a one-time cost per server per app run);
+        # once warm, every turn just finds the cached connection and returns.
         with self._conn_lock:
+            active = []
+            for entry in self.config.mcp_servers:
+                if not entry.get("enabled", True):
+                    continue
+                name = entry.get("name", "")
+                command = entry.get("command", "")
+                if not name or not command:
+                    continue
+                conn = self._mcp_connections.get(name)
+                if conn is None:
+                    conn = McpServerConnection(name, command, entry.get("args") or [],
+                                               entry.get("env") or None)
+                    try:
+                        conn.start()
+                    except Exception:  # noqa: BLE001 - one broken server must not block the turn
+                        continue
+                    self._mcp_connections[name] = conn
+                active.append(conn)
+            builtin = self._ms365_builtin_connection(skip={c.name for c in active})
+            if builtin is not None:
+                active.append(builtin)
+            mcp_tools, mcp_executor = _merge_mcp_tools(active)
+
             ext = self.config.ext_connectors
             all_connectors = [*ext.get("cad", []), *ext.get("cae", []),
                               *ext.get("ms365", []), *ext.get("other", [])]
             ext_tools, ext_executor = build_ext_connector_tools(all_connectors, self._ext_connections)
 
-        # Locally-synced OneDrive/SharePoint (no sign-in) — reads/writes the
-        # OneDrive-desktop-synced folders directly, gated on ms365.connectors.
-        from .core.ms365_local import build_ms365_local_tools
-        local_tools, local_executor = build_ms365_local_tools(self.config)
+            # Locally-synced OneDrive/SharePoint (no sign-in) — reads/writes the
+            # OneDrive-desktop-synced folders directly, gated on ms365.connectors.
+            from .core.ms365_local import build_ms365_local_tools
+            local_tools, local_executor = build_ms365_local_tools(self.config)
 
         return combine_tool_sources((mcp_tools, mcp_executor), (ext_tools, ext_executor),
                                     (local_tools, local_executor))
@@ -302,20 +259,36 @@ class AppContext:
         import sys
         from pathlib import Path
 
+        from .core.mcp_client import McpServerConnection
+
         name = self._MS365_BUILTIN
         if name in skip:
             return None
         if not self._ms365_available():
-            self._mcp_manager.stop(name)
+            stale = self._mcp_connections.pop(name, None)
+            if stale is not None:
+                try:
+                    stale.stop()
+                except Exception:  # noqa: BLE001
+                    pass
             return None
-        # The subprocess must import cowork_local even in a from-source run
-        # (PYTHONPATH=src) — prepend this package's parent dir explicitly.
-        env = dict(os.environ)
-        src_root = str(Path(__file__).resolve().parent.parent)
-        env["PYTHONPATH"] = (src_root + os.pathsep + env["PYTHONPATH"]
-                             if env.get("PYTHONPATH") else src_root)
-        return self._mcp_manager.ensure(
-            name, sys.executable, ["-m", "cowork_local.mcp_servers.ms365_server"], env)
+        conn = self._mcp_connections.get(name)
+        if conn is None:
+            # The subprocess must import cowork_local even in a from-source run
+            # (PYTHONPATH=src) — prepend this package's parent dir explicitly.
+            env = dict(os.environ)
+            src_root = str(Path(__file__).resolve().parent.parent)
+            env["PYTHONPATH"] = (src_root + os.pathsep + env["PYTHONPATH"]
+                                 if env.get("PYTHONPATH") else src_root)
+            conn = McpServerConnection(
+                name, sys.executable,
+                ["-m", "cowork_local.mcp_servers.ms365_server"], env)
+            try:
+                conn.start()
+            except Exception:  # noqa: BLE001 - MS365 down must not block the turn
+                return None
+            self._mcp_connections[name] = conn
+        return conn
 
     def stop_mcp_connections(self) -> None:
         """Terminate every connected MCP server's subprocess (incl. External
@@ -323,6 +296,11 @@ class AppContext:
         them linger as orphan processes."""
         from .core.ext_connectors import stop_ext_connections
 
-        self._mcp_manager.stop_all()
         with self._conn_lock:
+            for conn in self._mcp_connections.values():
+                try:
+                    conn.stop()
+                except Exception:  # noqa: BLE001
+                    pass
+            self._mcp_connections.clear()
             stop_ext_connections(self._ext_connections)

tests/characterization/__init__.py

@@ -1,11 +0,0 @@
-"""Characterization tests: pin the CURRENT behaviour of legacy code (R01-T04).
-
-These are not specifications of what the code *should* do - they are a snapshot
-of what it *does* today, written before the refactor so that any behavioural
-drift introduced while moving logic into ``application/`` shows up as a failing
-test rather than as a bug report from a user.
-
-Rule for this folder: when a test here fails during the refactor, do not "fix"
-the test first. Decide deliberately whether the behaviour change is intended,
-and only then update the snapshot in the same commit as the change.
-"""

tests/characterization/test_run_cowork.py

@@ -1,288 +1,157 @@
-"""Characterization snapshot of ``core.chat_agent.run_cowork`` (R01-T04).
+"""Characterization tests for core/chat_agent.py (run_chat and run_cowork runtime seams).
 
-``run_cowork`` is the turn engine every Cowork surface funnels through (chat tab,
-Co4E flow steps, Schedule Task runs). EPIC R04 moves its orchestration into
-``application/conversations/conversation_application_service.py``; these tests
-lock down the observable contract BEFORE that move so the new service can be
-proven equivalent:
-
-* which system prompt ends up in ``messages``
-* which tools are advertised to the provider
-* the exact ``emit`` event sequence for a plain turn and for a tool turn
-* that ``save_file`` produces a real file in the turn's output folder
-* that ``cancel`` stops the loop without calling the provider
-
-Everything runs offline: :class:`FakeProvider` replaces the network and the two
-disk-backed prompt sources (skills, security rules) are stubbed to empty so the
-snapshot does not depend on the developer's own ``~/.cowork_local`` contents.
+These tests capture existing behavior as an executable baseline specification,
+ensuring that future refactoring to ConversationApplicationService does not alter
+core turn semantics, event emissions, or file handling.
 """
 from __future__ import annotations
 
 from pathlib import Path
 from typing import Any, Dict, List
 
-import pytest
-
 from cowork_local.core import chat_agent
-from tests.fakes import FakeProvider, FakeToolExecutor, ScriptedTurn
+from cowork_local.tests.fakes.fake_provider import FakeProvider
 
 
-@pytest.fixture
-def isolated_agent(monkeypatch, tmp_path: Path):
-    """Neutralise every ambient input ``run_cowork`` reads from the machine.
+def test_run_chat_characterization() -> None:
+    """Capture baseline behavior of run_chat: system prompt insertion, streaming, and message persistence."""
+    provider = FakeProvider()
+    provider.queue_response(content="Hello there!", chunks=["Hello ", "there!"])
 
-    Without this the snapshot would silently depend on whichever skills and
-    security rules the developer happens to have enabled locally, and on the
-    real audit log under ``~/.cowork_local`` - the test would then pass on one
-    laptop and fail on another for reasons unrelated to the code under test.
-    """
-    monkeypatch.setattr(chat_agent, "active_skills_text", lambda: "")
-    monkeypatch.setattr(chat_agent, "load_rules", lambda: "")
-    # audit_log is imported lazily inside run_cowork, so patch the module's own
-    # target directory rather than the name chat_agent sees.
-    from cowork_local.core import audit_log
+    messages: List[Dict[str, Any]] = [{"role": "user", "content": "Hi assistant"}]
+    emitted_events: List[Dict[str, Any]] = []
 
-    monkeypatch.setattr(audit_log, "AUDIT_DIR", tmp_path / "audit")
-    return tmp_path
+    def emit(event: Dict[str, Any]) -> None:
+        emitted_events.append(event)
+
+    result = chat_agent.run_chat(
+        provider=provider,
+        messages=messages,
+        emit=emit,
+    )
+
+    # 1. Verify system prompt was injected at position 0
+    assert messages[0]["role"] == "system"
+    assert "Cowork Local" in messages[0]["content"]
+
+    # 2. Verify returned assistant message
+    assert result["role"] == "assistant"
+    assert result["content"] == "Hello there!"
+
+    # 3. Verify assistant message was appended to messages list
+    assert messages[-1] == result
+
+    # 4. Verify emitted events sequence
+    text_deltas = [e["delta"] for e in emitted_events if e["type"] == "text"]
+    assert "".join(text_deltas) == "Hello there!"
+    assert any(e["type"] == "assistant_done" for e in emitted_events)
 
 
-def _run(provider, messages, out_dir: Path, **kwargs):
-    """Run one turn and return ``(returned_messages, emitted_events)``."""
-    events: List[Dict[str, Any]] = []
-    result = chat_agent.run_cowork(provider, messages, out_dir, events.append, **kwargs)
-    return result, events
+def test_run_cowork_save_file_characterization(tmp_path: Path) -> None:
+    """Capture baseline behavior of run_cowork: tool execution loop and file production."""
+    output_dir = tmp_path / "output"
+    output_dir.mkdir(parents=True, exist_ok=True)
+
+    provider = FakeProvider()
+    # Step 1: Model requests save_file tool
+    provider.queue_response(
+        content="Saving your requested report.",
+        tool_calls=[{
+            "id": "call_save_1",
+            "name": "save_file",
+            "arguments": {
+                "filename": "report.md",
+                "content": "# Executive Summary\nAll systems nominal.",
+            },
+        }],
+    )
+    # Step 2: Model finishes after tool result
+    provider.queue_response(
+        content="I have created report.md in your output directory.",
+        chunks=["I have created report.md in your output directory."],
+    )
+
+    messages: List[Dict[str, Any]] = [{"role": "user", "content": "Export report to markdown file"}]
+    emitted_events: List[Dict[str, Any]] = []
+
+    def emit(event: Dict[str, Any]) -> None:
+        emitted_events.append(event)
+
+    final_messages = chat_agent.run_cowork(
+        provider=provider,
+        messages=messages,
+        output_dir=output_dir,
+        emit=emit,
+        enforce_rules=False,
+    )
+
+    # 1. Verify file was created in output directory with expected content
+    created_file = output_dir / "report.md"
+    assert created_file.exists()
+    assert created_file.read_text(encoding="utf-8") == "# Executive Summary\nAll systems nominal."
+
+    # 2. Verify message history contains user -> assistant (tool_calls) -> tool -> assistant
+    roles = [m["role"] for m in final_messages]
+    assert "system" in roles
+    assert "user" in roles
+    assert "tool" in roles
+
+    # 3. Verify tool result message content
+    tool_msg = next(m for m in final_messages if m["role"] == "tool")
+    assert tool_msg["name"] == "save_file"
+    assert "Saved report.md" in tool_msg["content"]
 
 
-def _types(events: List[Dict[str, Any]]) -> List[str]:
-    """Event ``type`` values in order - the shape assertions read on."""
-    return [e.get("type") for e in events]
+def test_run_cowork_cancellation_characterization(tmp_path: Path) -> None:
+    """Capture cancellation behavior in run_cowork."""
+    output_dir = tmp_path / "output_cancel"
+    output_dir.mkdir(parents=True, exist_ok=True)
 
+    provider = FakeProvider()
+    provider.queue_response(content="Working...")
 
-# --------------------------------------------------------------------------- #
-# A plain answer with no tool calls
-# --------------------------------------------------------------------------- #
-def test_plain_turn_streams_text_and_appends_assistant_message(isolated_agent):
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="Hello there.")])
-    messages: List[Dict[str, Any]] = [{"role": "user", "content": "hi"}]
+    is_cancelled = True
 
-    result, events = _run(provider, messages, out_dir)
+    def check_cancel() -> bool:
+        return is_cancelled
 
-    # The loop ends as soon as the model stops calling tools: exactly one call.
-    assert provider.call_count == 1
-    # run_cowork mutates and returns the SAME list the caller passed in - callers
-    # (ui/cowork_tab.py::build_job) rely on this to persist conversation history.
-    assert result is messages
-    assert result[-1]["role"] == "assistant"
-    assert result[-1]["content"] == "Hello there."
-    assert _types(events) == ["text", "assistant_done"]
-    assert events[0]["delta"] == "Hello there."
-    assert events[-1]["content"] == "Hello there."
+    emitted_events: List[Dict[str, Any]] = []
+    messages: List[Dict[str, Any]] = [{"role": "user", "content": "Please start"}]
 
+    chat_agent.run_cowork(
+        provider=provider,
+        messages=messages,
+        output_dir=output_dir,
+        emit=lambda e: emitted_events.append(e),
+        cancel=check_cancel,
+        enforce_rules=False,
+    )
 
-def test_system_prompt_is_inserted_once_at_the_front(isolated_agent):
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="ok")])
-    messages: List[Dict[str, Any]] = [{"role": "user", "content": "hi"}]
-
-    result, _ = _run(provider, messages, out_dir)
-
-    assert result[0]["role"] == "system"
-    assert result[0]["content"].startswith("You are Cowork Local")
-    # Exactly one system message: a second turn on the same conversation must not
-    # stack another copy of the prompt (that would grow the context every turn).
-    assert sum(1 for m in result if m.get("role") == "system") == 1
-
-
-def test_caller_supplied_system_prompt_is_preserved(isolated_agent):
-    """A caller that already put a system message first keeps its own prompt.
-
-    Co4E flow steps depend on this to give a step its own persona instead of the
-    generic Cowork prompt.
-    """
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="ok")])
-    messages: List[Dict[str, Any]] = [
-        {"role": "system", "content": "CUSTOM PERSONA"},
-        {"role": "user", "content": "hi"},
-    ]
-
-    result, _ = _run(provider, messages, out_dir)
-
-    assert result[0]["content"] == "CUSTOM PERSONA"
-
-
-def test_reasoning_is_emitted_separately_and_never_joins_the_answer(isolated_agent):
-    """Reasoning drives the "Thinking" indicator only - it must not become part
-    of the assistant's content, otherwise a reasoning model's private chain of
-    thought would be persisted into conversation history."""
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="42", reasoning="let me think...")])
-
-    result, events = _run(provider, [{"role": "user", "content": "q"}], out_dir)
-
-    assert _types(events) == ["reasoning", "text", "assistant_done"]
-    assert result[-1]["content"] == "42"
-    assert "let me think" not in result[-1]["content"]
-
-
-def test_reasoning_only_reply_gets_a_placeholder_answer(isolated_agent):
-    """A model that returns only reasoning must not end the turn on a blank
-    bubble - headless callers (Schedule Task) read this content back as the
-    run's final answer and would otherwise write "(no output)"."""
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="", reasoning="thinking")])
-
-    result, events = _run(provider, [{"role": "user", "content": "q"}], out_dir)
-
-    assert result[-1]["content"].startswith("*(model returned only its reasoning")
-    assert "text" in _types(events)
-
-
-# --------------------------------------------------------------------------- #
-# Tool advertising
-# --------------------------------------------------------------------------- #
-def test_save_file_and_update_plan_are_always_advertised(isolated_agent):
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="ok")])
-
-    _run(provider, [{"role": "user", "content": "hi"}], out_dir)
-
-    advertised = provider.calls[0].tool_names
-    assert "save_file" in advertised
-    assert "update_plan" in advertised
-
-
-def test_allowed_tools_scopes_the_catalogue_but_keeps_update_plan(isolated_agent):
-    """``allowed_tools`` is the permission scope Co4E steps use: a read-only step
-    must literally not be offered a writing tool. ``update_plan`` survives the
-    filter because it has no side effects."""
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([ScriptedTurn(text="ok")])
-
-    _run(provider, [{"role": "user", "content": "hi"}], out_dir,
-         allowed_tools=["read_file"])
-
-    advertised = set(provider.calls[0].tool_names)
-    assert "save_file" not in advertised
-    assert "update_plan" in advertised
-
-
-def test_extra_tools_are_advertised_alongside_built_ins(isolated_agent):
-    out_dir = isolated_agent / "out"
-    executor = FakeToolExecutor(results={"ms365_send_mail": {"output": "sent"}})
-    provider = FakeProvider([ScriptedTurn(text="ok")])
-
-    _run(provider, [{"role": "user", "content": "hi"}], out_dir,
-         extra_tools=executor.specs(), extra_executor=executor)
-
-    assert "ms365_send_mail" in provider.calls[0].tool_names
-
-
-# --------------------------------------------------------------------------- #
-# Tool execution
-# --------------------------------------------------------------------------- #
-def test_save_file_writes_a_real_file_and_reports_it(isolated_agent):
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([
-        ScriptedTurn(tool_calls=[("save_file", {"filename": "note.md",
-                                                "content": "# Result\n"})]),
-        ScriptedTurn(text="Done."),
-    ])
-
-    result, events = _run(provider, [{"role": "user", "content": "make a note"}], out_dir)
-
-    written = [p for p in out_dir.iterdir() if p.is_file()]
-    assert len(written) == 1
-    assert written[0].read_text(encoding="utf-8") == "# Result\n"
-
-    assert _types(events) == [
-        "assistant_done",       # first turn: tool call only, no visible text
-        "tool_proposed",        # the diff preview shown in the chat
-        "tool_result",
-        "text",                 # second turn's answer
-        "assistant_done",
-    ]
-    assert events[2]["ok"] is True
-
-    # The tool result is fed back as a `tool` message so the model can react to it.
-    roles = [m["role"] for m in result]
-    assert roles == ["system", "user", "assistant", "tool", "assistant"]
-    assert result[3]["name"] == "save_file"
-
-
-def test_extra_tool_calls_are_routed_to_the_extra_executor(isolated_agent):
-    """MCP / Microsoft 365 tools bypass the built-in file+command handlers and go
-    to the caller-supplied executor instead."""
-    out_dir = isolated_agent / "out"
-    executor = FakeToolExecutor(results={"ms365_send_mail": {"ok": True, "output": "sent"}})
-    provider = FakeProvider([
-        ScriptedTurn(tool_calls=[("ms365_send_mail", {"to": "a@b.c"})]),
-        ScriptedTurn(text="Mail sent."),
-    ])
-
-    result, events = _run(provider, [{"role": "user", "content": "mail them"}], out_dir,
-                          extra_tools=executor.specs(), extra_executor=executor)
-
-    assert executor.call_names == ["ms365_send_mail"]
-    assert executor.args_for("ms365_send_mail") == [{"to": "a@b.c"}]
-    assert [e for e in events if e["type"] == "tool_result"][0]["output"] == "sent"
-    assert result[3] == {"role": "tool", "tool_call_id": result[3]["tool_call_id"],
-                         "name": "ms365_send_mail", "content": "sent"}
-
-
-def test_update_plan_drives_the_plan_panel_without_producing_a_file(isolated_agent):
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([
-        ScriptedTurn(tool_calls=[("update_plan", {"steps": [{"title": "step one"}]})]),
-        ScriptedTurn(text="Planned."),
-    ])
-
-    result, events = _run(provider, [{"role": "user", "content": "plan it"}], out_dir)
-
-    plan_events = [e for e in events if e["type"] == "plan_set"]
-    assert len(plan_events) == 1
-    assert plan_events[0]["steps"]
-    # No tool_proposed/tool_result bubbles for a plan update, and no file on disk.
-    assert "tool_proposed" not in _types(events)
-    assert list(out_dir.iterdir()) == []
-    assert result[3]["content"] == "Plan updated."
-
-
-# --------------------------------------------------------------------------- #
-# Cancellation
-# --------------------------------------------------------------------------- #
-def test_cancel_before_the_first_step_never_calls_the_provider(isolated_agent):
-    """Stop pressed before the loop starts must cost zero tokens."""
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([], strict=True)
-
-    result, events = _run(provider, [{"role": "user", "content": "hi"}], out_dir,
-                          cancel=lambda: True)
-
+    # Provider should not have executed turns if cancelled right away
     assert provider.call_count == 0
-    assert _types(events) == []
-    # The system prompt is still installed, so the conversation stays well-formed
-    # for a later retry on the same message list.
-    assert result[0]["role"] == "system"
 
 
-def test_cancel_between_steps_stops_before_the_next_provider_call(isolated_agent):
-    """After a tool call runs, a Stop must end the turn instead of paying for
-    another round trip."""
-    out_dir = isolated_agent / "out"
-    provider = FakeProvider([
-        ScriptedTurn(tool_calls=[("save_file", {"filename": "a.md", "content": "x"})]),
-    ])
-    calls = {"n": 0}
+def test_cleanup_turn_output_characterization(tmp_path: Path) -> None:
+    """Capture behavior of temporary .scratch folder cleanup and artifact preservation."""
+    output_dir = tmp_path / "output_cleanup"
+    output_dir.mkdir(parents=True, exist_ok=True)
+    scratch_dir = output_dir / ".scratch"
+    scratch_dir.mkdir(parents=True, exist_ok=True)
 
-    def cancel() -> bool:
-        # False on the first check (loop entry), True afterwards - i.e. the user
-        # pressed Stop while the first step was running.
-        calls["n"] += 1
-        return calls["n"] > 1
+    # Create a generator script and a deliverable inside scratch
+    generator_script = scratch_dir / "gen.py"
+    generator_script.write_text("print('generating')", encoding="utf-8")
+    deliverable = scratch_dir / "data.csv"
+    deliverable.write_text("a,b,c\n1,2,3", encoding="utf-8")
 
-    result, _ = _run(provider, [{"role": "user", "content": "hi"}], out_dir, cancel=cancel)
+    before_snapshot = chat_agent._snapshot(output_dir)
+    removed, moved = chat_agent._cleanup_cowork_intermediates(output_dir, before_snapshot, cancelled=False)
+
+    # .scratch directory should be removed
+    assert not scratch_dir.exists()
+    # deliverable should be moved to output root
+    root_csv = output_dir / "data.csv"
+    assert root_csv.exists()
+    # script should not be in output root
+    assert not (output_dir / "gen.py").exists()
 
-    assert provider.call_count == 1
-    assert result[-1]["role"] in {"assistant", "tool"}

tests/conftest.py

@@ -1,63 +1,10 @@
-"""Root pytest configuration: bind ``cowork_local`` to THIS checkout (R01-T02).
+"""Make the repository package importable when pytest runs from the repo root."""
 
-Why this file exists
---------------------
-The package directory is itself the distribution package (``__init__.py`` sits
-at the repo root), so ``import cowork_local`` only resolves when the checkout
-folder happens to be named exactly ``cowork_local``. It frequently is not — this
-one is checked out as ``cowork_local_gitea``, and developers keep several dated
-copies side by side (``cowork_local``, ``cowork_local_20260722``, ...).
-
-Left alone, ``sys.path``-based discovery would import whichever *sibling* folder
-is named ``cowork_local`` and the whole suite would silently test a DIFFERENT
-checkout: green here, broken in the branch under review. That is the worst kind
-of test failure, because it fails to fail.
-
-So instead of relying on the folder name, we load ``__init__.py`` by absolute
-path and register the result in ``sys.modules`` under the canonical name before
-any test imports it. Submodules (``cowork_local.providers.base``, ...) then
-resolve through this package's own ``__path__``, i.e. always this checkout.
-"""
 from __future__ import annotations
 
-import importlib.util
 import sys
 from pathlib import Path
 
-# .../<checkout>/tests/conftest.py -> .../<checkout>
-_PKG_DIR = Path(__file__).resolve().parents[1]
-_PKG_NAME = "cowork_local"
-
-
-def _bind_package_to_this_checkout() -> None:
-    """Make ``import cowork_local`` mean this directory, whatever it is named.
-
-    A no-op when the correct package object is already bound, so running the
-    suite from a folder that IS named ``cowork_local`` costs nothing and the
-    hook stays idempotent across repeated conftest collection.
-    """
-    existing = sys.modules.get(_PKG_NAME)
-    existing_file = getattr(existing, "__file__", None)
-    if existing_file and Path(existing_file).resolve().parent == _PKG_DIR:
-        return  # already the right one
-
-    spec = importlib.util.spec_from_file_location(
-        _PKG_NAME,
-        _PKG_DIR / "__init__.py",
-        # Setting the search locations is what makes dotted submodule imports
-        # (cowork_local.core.*, cowork_local.providers.*) resolve inside THIS
-        # directory rather than through sys.path.
-        submodule_search_locations=[str(_PKG_DIR)],
-    )
-    if spec is None or spec.loader is None:  # pragma: no cover - packaging error
-        raise RuntimeError(f"cannot load {_PKG_NAME} from {_PKG_DIR}")
-
-    module = importlib.util.module_from_spec(spec)
-    # Registered BEFORE exec_module so that a self-referential import inside
-    # __init__.py would find the partially-initialised module instead of
-    # recursing - the same protocol CPython's own import machinery follows.
-    sys.modules[_PKG_NAME] = module
-    spec.loader.exec_module(module)
-
-
-_bind_package_to_this_checkout()
+REPOSITORY_PARENT = Path(__file__).resolve().parents[2]
+if str(REPOSITORY_PARENT) not in sys.path:
+    sys.path.insert(0, str(REPOSITORY_PARENT))

tests/contracts/__init__.py

@@ -1,8 +0,0 @@
-"""Contract tests: one shared behaviour suite every implementation must satisfy.
-
-Unlike unit tests (which test one module in isolation) a contract test is
-parametrised over EVERY implementation of an interface, so a newly added
-provider either satisfies the same promises as the existing ones or the suite
-goes red on the day it is added - not months later, in production, on the one
-code path that assumed the promise held.
-"""

tests/contracts/test_providers.py

@@ -1,342 +0,0 @@
-"""Provider contract suite (R03-T01).
-
-Every provider - the two real adapters and the test double - must honour the
-same promises declared in ``providers/base.py``:
-
-1. ``chat()`` returns the canonical assistant message
-   ``{"role": "assistant", "content": str, "tool_calls": [...]}``.
-2. Answer text is streamed through ``on_text`` and equals the returned content.
-3. Private reasoning goes to ``on_reasoning`` ONLY - it must never leak into the
-   answer, or a reasoning model's chain of thought ends up persisted in history.
-4. Tool calls come back as ``{"id", "name", "arguments": dict}`` with arguments
-   already parsed - callers must never have to json.loads() them.
-5. A failure raises ``ProviderError`` and nothing else, so one except clause in
-   the agent loop covers every provider.
-
-The real adapters are exercised WITHOUT network access by replacing
-``Provider._request`` with a canned SSE response - which is exactly the seam
-``providers/base.py`` documents for its TLS retry, so no production code needed
-changing to make this testable.
-"""
-from __future__ import annotations
-
-import json
-from typing import Any, Dict, List, Optional
-
-import pytest
-
-from cowork_local.domain.models.provider_descriptor import ProviderCapability
-from cowork_local.infrastructure.providers.provider_registry import (
-    BUILT_IN_PROVIDERS,
-    ProviderRegistry,
-)
-from cowork_local.providers.anthropic import AnthropicProvider
-from cowork_local.providers.base import Provider, ProviderError, ToolSpec
-from cowork_local.providers.openai_compat import OpenAICompatProvider
-from tests.fakes import FakeProvider, ScriptedTurn
-
-
-class _StubResponse:
-    """Minimal stand-in for a streamed ``requests.Response``.
-
-    Only the members the provider code actually touches are implemented; adding
-    more would invite tests that pass against the stub but not against requests.
-    """
-
-    def __init__(self, lines: List[str], status_code: int = 200, text: str = "") -> None:
-        self._lines = lines
-        self.status_code = status_code
-        self.text = text
-        self.headers: Dict[str, str] = {}
-        self.encoding = "utf-8"
-        self.closed = False
-
-    def iter_lines(self, decode_unicode: bool = False):
-        yield from self._lines
-
-    def close(self) -> None:
-        self.closed = True
-
-    def json(self) -> Any:
-        return json.loads(self.text or "{}")
-
-
-def _sse(*payloads: Dict[str, Any]) -> List[str]:
-    """Render payloads as SSE ``data:`` lines, the wire shape both adapters parse."""
-    return [f"data: {json.dumps(p)}" for p in payloads]
-
-
-@pytest.fixture
-def canned(monkeypatch):
-    """Return a helper that makes every provider request answer with ``lines``."""
-
-    def _install(lines: List[str], status_code: int = 200, text: str = "") -> Dict[str, Any]:
-        seen: Dict[str, Any] = {}
-
-        def fake_request(self, method, url, **kwargs):
-            # Capture the outgoing payload so tests can assert on how the
-            # canonical message list was translated to the provider's wire format.
-            seen["method"] = method
-            seen["url"] = url
-            seen["json"] = kwargs.get("json")
-            return _StubResponse(lines, status_code=status_code, text=text)
-
-        monkeypatch.setattr(Provider, "_request", fake_request, raising=True)
-        return seen
-
-    return _install
-
-
-# --------------------------------------------------------------------------- #
-# Shared base-class behaviour every provider inherits
-# --------------------------------------------------------------------------- #
-def _providers_under_test() -> List[Provider]:
-    """One instance of each implementation, configured but never called."""
-    conf = {"base_url": "https://example.invalid/v1", "api_key": "k", "model": "m"}
-    return [
-        OpenAICompatProvider(dict(conf)),
-        AnthropicProvider(dict(conf)),
-        FakeProvider(),
-    ]
-
-
-@pytest.mark.parametrize("provider", _providers_under_test(), ids=lambda p: type(p).__name__)
-def test_every_provider_exposes_the_base_contract(provider):
-    assert isinstance(provider, Provider)
-    assert callable(provider.chat)
-    assert callable(provider.list_models)
-    assert callable(provider.test_connection)
-    # `name` identifies the provider in usage records and audit entries; an
-    # implementation that forgot to set it would silently report as "base".
-    assert provider.name and provider.name != "base"
-    assert isinstance(provider.supports_vision, bool)
-    assert provider.describe() == f"{provider.name}:{provider.model}"
-
-
-@pytest.mark.parametrize("provider", _providers_under_test(), ids=lambda p: type(p).__name__)
-def test_strip_think_removes_inline_reasoning_from_a_final_answer(provider):
-    """Safety net for gateways that fold reasoning into the content stream: the
-    answer stored in history must never contain a <think> block."""
-    assert provider.strip_think("<think>secret</think>Answer") == "Answer"
-    assert provider.strip_think("Plain answer") == "Plain answer"
-
-
-def test_tool_spec_translates_to_both_wire_formats():
-    """One ToolSpec must render for both protocols - this is what lets the agent
-    loop build its tool catalogue once and reuse it across providers."""
-    spec = ToolSpec(name="save_file", description="Write a file",
-                    parameters={"type": "object", "properties": {}})
-
-    openai_shape = spec.to_openai()
-    anthropic_shape = spec.to_anthropic()
-
-    assert openai_shape["type"] == "function"
-    assert openai_shape["function"]["name"] == "save_file"
-    assert openai_shape["function"]["parameters"] == spec.parameters
-    # Anthropic names the same field `input_schema`; the values must stay equal,
-    # otherwise the same tool would validate differently per provider.
-    assert anthropic_shape["name"] == "save_file"
-    assert anthropic_shape["input_schema"] == spec.parameters
-
-
-# --------------------------------------------------------------------------- #
-# Streaming contract - real adapters, canned transport
-# --------------------------------------------------------------------------- #
-def test_openai_compat_streams_text_and_returns_canonical_message(canned):
-    canned(_sse(
-        {"choices": [{"delta": {"content": "Hel"}}]},
-        {"choices": [{"delta": {"content": "lo"}}]},
-    ) + ["data: [DONE]"])
-    provider = OpenAICompatProvider({"base_url": "https://x.invalid/v1",
-                                     "api_key": "k", "model": "m"})
-    chunks: List[str] = []
-
-    result = provider.chat([{"role": "user", "content": "hi"}], on_text=chunks.append)
-
-    assert "".join(chunks) == "Hello"
-    assert result["role"] == "assistant"
-    assert result["content"] == "Hello"
-    assert result["tool_calls"] == []
-
-
-def test_openai_compat_keeps_reasoning_out_of_the_answer(canned):
-    canned(_sse(
-        {"choices": [{"delta": {"reasoning_content": "hmm..."}}]},
-        {"choices": [{"delta": {"content": "42"}}]},
-    ) + ["data: [DONE]"])
-    provider = OpenAICompatProvider({"base_url": "https://x.invalid/v1",
-                                     "api_key": "k", "model": "m"})
-    text: List[str] = []
-    reasoning: List[str] = []
-
-    result = provider.chat([{"role": "user", "content": "q"}],
-                           on_text=text.append, on_reasoning=reasoning.append)
-
-    assert reasoning == ["hmm..."]
-    assert result["content"] == "42"
-    assert "hmm" not in result["content"]
-
-
-def test_openai_compat_returns_tool_calls_with_parsed_arguments(canned):
-    """Arguments arrive as a JSON string split across chunks; the contract says
-    the caller receives a ready-to-use dict."""
-    canned(_sse(
-        {"choices": [{"delta": {"tool_calls": [
-            {"index": 0, "id": "call_a", "function": {"name": "save_file",
-                                                      "arguments": '{"filename":'}}]}}]},
-        {"choices": [{"delta": {"tool_calls": [
-            {"index": 0, "function": {"arguments": '"a.md"}'}}]}}]},
-    ) + ["data: [DONE]"])
-    provider = OpenAICompatProvider({"base_url": "https://x.invalid/v1",
-                                     "api_key": "k", "model": "m"})
-
-    result = provider.chat([{"role": "user", "content": "save it"}])
-
-    assert len(result["tool_calls"]) == 1
-    call = result["tool_calls"][0]
-    assert call["id"] == "call_a"
-    assert call["name"] == "save_file"
-    assert call["arguments"] == {"filename": "a.md"}
-
-
-def test_anthropic_streams_text_and_returns_canonical_message(canned):
-    canned(_sse(
-        {"type": "content_block_delta", "index": 0,
-         "delta": {"type": "text_delta", "text": "Hel"}},
-        {"type": "content_block_delta", "index": 0,
-         "delta": {"type": "text_delta", "text": "lo"}},
-        {"type": "message_stop"},
-    ))
-    provider = AnthropicProvider({"base_url": "https://x.invalid",
-                                  "api_key": "k", "model": "m"})
-    chunks: List[str] = []
-
-    result = provider.chat([{"role": "user", "content": "hi"}], on_text=chunks.append)
-
-    assert "".join(chunks) == "Hello"
-    assert result["content"] == "Hello"
-    assert result["role"] == "assistant"
-
-
-def test_anthropic_keeps_extended_thinking_out_of_the_answer(canned):
-    canned(_sse(
-        {"type": "content_block_delta", "index": 0,
-         "delta": {"type": "thinking_delta", "thinking": "reasoning..."}},
-        {"type": "content_block_delta", "index": 0,
-         "delta": {"type": "text_delta", "text": "42"}},
-        {"type": "message_stop"},
-    ))
-    provider = AnthropicProvider({"base_url": "https://x.invalid",
-                                 "api_key": "k", "model": "m"})
-    reasoning: List[str] = []
-
-    result = provider.chat([{"role": "user", "content": "q"}], on_reasoning=reasoning.append)
-
-    assert reasoning == ["reasoning..."]
-    assert result["content"] == "42"
-
-
-def test_anthropic_returns_tool_calls_with_parsed_arguments(canned):
-    canned(_sse(
-        {"type": "content_block_start", "index": 0,
-         "content_block": {"type": "tool_use", "id": "toolu_1", "name": "save_file"}},
-        {"type": "content_block_delta", "index": 0,
-         "delta": {"type": "input_json_delta", "partial_json": '{"filename":"a.md"}'}},
-        {"type": "message_stop"},
-    ))
-    provider = AnthropicProvider({"base_url": "https://x.invalid",
-                                 "api_key": "k", "model": "m"})
-
-    result = provider.chat([{"role": "user", "content": "save"}])
-
-    assert result["tool_calls"] == [
-        {"id": "toolu_1", "name": "save_file", "arguments": {"filename": "a.md"}}
-    ]
-
-
-@pytest.mark.parametrize("factory", [
-    lambda: OpenAICompatProvider({"base_url": "https://x.invalid/v1", "api_key": "k", "model": "m"}),
-    lambda: AnthropicProvider({"base_url": "https://x.invalid", "api_key": "k", "model": "m"}),
-], ids=["openai_compat", "anthropic"])
-def test_transport_failure_surfaces_as_provider_error(canned, factory):
-    """Every failure mode must arrive as ProviderError so the agent loop needs
-    exactly one except clause, whichever provider is active."""
-    canned([], status_code=500, text="boom")
-
-    with pytest.raises(ProviderError):
-        factory().chat([{"role": "user", "content": "hi"}])
-
-
-def test_fake_provider_satisfies_the_same_streaming_contract():
-    """The double is only useful as a stand-in if it keeps the same promises the
-    real adapters are held to above."""
-    provider = FakeProvider([ScriptedTurn(text="Hello", reasoning="hmm")])
-    text: List[str] = []
-    reasoning: List[str] = []
-
-    result = provider.chat([{"role": "user", "content": "hi"}],
-                           on_text=text.append, on_reasoning=reasoning.append)
-
-    assert "".join(text) == result["content"] == "Hello"
-    assert reasoning == ["hmm"]
-    assert result["role"] == "assistant"
-    assert result["tool_calls"] == []
-
-
-def test_fake_provider_raises_provider_error_like_the_real_ones():
-    provider = FakeProvider([ScriptedTurn(error="gateway exploded")])
-
-    with pytest.raises(ProviderError):
-        provider.chat([{"role": "user", "content": "hi"}])
-
-
-# --------------------------------------------------------------------------- #
-# Registry <-> implementation agreement
-# --------------------------------------------------------------------------- #
-@pytest.mark.parametrize("descriptor", BUILT_IN_PROVIDERS, ids=lambda d: d.id)
-def test_every_descriptor_builds_a_working_provider(descriptor):
-    """A descriptor that cannot be built is a catalogue lying to the UI: Settings
-    would list the provider and selecting it would fail at the first message."""
-    registry = ProviderRegistry()
-    conf = {"base_url": "https://x.invalid/v1", "api_key": "k"}
-
-    provider = registry.build(descriptor.id, conf)
-
-    assert isinstance(provider, Provider)
-    # The id, not the shared adapter class name: three descriptors map onto
-    # OpenAICompatProvider, and usage/audit records must still tell them apart.
-    assert provider.name == descriptor.id
-    assert provider.model == descriptor.default_model
-
-
-@pytest.mark.parametrize("descriptor", BUILT_IN_PROVIDERS, ids=lambda d: d.id)
-def test_declared_vision_capability_matches_the_implementation(descriptor):
-    """``supports_vision`` decides whether an image block may be sent. A
-    descriptor claiming vision for an adapter that cannot translate the block
-    would route image turns into a guaranteed failure."""
-    provider = ProviderRegistry().build(descriptor.id, {"base_url": "u", "api_key": "k"})
-
-    if descriptor.supports(ProviderCapability.VISION):
-        assert provider.supports_vision is True
-
-
-def test_registry_build_never_mutates_the_caller_config():
-    """The routing layer runs one turn on a different model; if build() wrote
-    that model back into the config dict it was handed, the override would
-    silently become the user's saved default."""
-    registry = ProviderRegistry()
-    conf = {"base_url": "u", "api_key": "k", "model": "configured-model"}
-
-    provider = registry.build("openai_compat", conf, model="routed-model")
-
-    assert provider.model == "routed-model"
-    assert conf["model"] == "configured-model"
-
-
-def test_registry_rejects_an_unknown_provider_with_provider_error():
-    with pytest.raises(ProviderError) as excinfo:
-        ProviderRegistry().build("does_not_exist", {})
-
-    # The message lists what IS known, so a typo in config is fixable from the
-    # error alone without opening the source.
-    assert "openai_compat" in str(excinfo.value)

tests/fakes/__init__.py

@@ -1,16 +1,5 @@
-"""Offline test doubles for the refactoring safety net (R01-T02).
+"""Test doubles and offline fakes package for Cowork Local test pyramid."""
+from .fake_provider import FakeProvider
+from .fake_tool_executor import FakeToolExecutor
 
-Every double here is deliberately Qt-free, network-free and disk-free so the
-unit/contract suites run in well under a second and give the same answer on a
-laptop, in CI and on a machine with no API keys configured.
-
-* :class:`~tests.fakes.fake_provider.FakeProvider` - a scripted
-  ``providers.base.Provider`` that streams canned text/tool calls.
-* :class:`~tests.fakes.fake_tool_executor.FakeToolExecutor` - a scripted stand-in
-  for the ``extra_executor`` callable that ``core.chat_agent.run_cowork`` routes
-  MCP/connector tool calls to.
-"""
-from .fake_provider import FakeProvider, ScriptedTurn
-from .fake_tool_executor import FakeToolExecutor, ToolInvocation
-
-__all__ = ["FakeProvider", "ScriptedTurn", "FakeToolExecutor", "ToolInvocation"]
+__all__ = ["FakeProvider", "FakeToolExecutor"]

tests/fakes/fake_provider.py

@@ -1,128 +1,58 @@
-"""FakeProvider - a scripted, offline stand-in for a real LLM provider (R01-T02).
+"""Fake LLM Provider for offline unit, contract, and characterization testing.
 
-The real providers (``providers/openai_compat.py``, ``providers/anthropic.py``)
-open HTTP connections, need API keys and stream at the mercy of the network, so
-nothing above them could be tested deterministically. This double implements the
-same :class:`providers.base.Provider` contract from a list of scripted turns:
-
-    provider = FakeProvider([
-        ScriptedTurn(tool_calls=[("save_file", {"filename": "a.md", "content": "hi"})]),
-        ScriptedTurn(text="Saved it."),
-    ])
-
-Turn 1 asks the agent loop to call a tool, turn 2 ends the loop with plain text -
-exactly the two-step shape ``run_cowork`` exercises, with zero I/O.
-
-It records every call it received (:attr:`FakeProvider.calls`) so a test can
-assert on what the layer above actually sent (message list, tool catalogue),
-which is how the characterization and contract suites pin current behaviour.
+Provides deterministic responses, stream simulation, tool-call dispatching,
+and fault injection without requiring any external network access or API keys.
 """
 from __future__ import annotations
 
-import itertools
-from dataclasses import dataclass
-from typing import Any, Dict, List, Optional, Sequence, Tuple
+from typing import Any, Callable, Dict, List, Optional
 
-from cowork_local.providers.base import (
-    CancelFn,
-    Provider,
-    ProviderError,
-    TextCallback,
-    ToolSpec,
-)
-
-# One scripted tool call: (name, arguments). Ids are generated by the provider so
-# a test never has to invent them, mirroring what a real gateway does.
-ToolCallScript = Tuple[str, Dict[str, Any]]
-
-
-@dataclass(frozen=True)
-class ScriptedTurn:
-    """What :class:`FakeProvider` should do for ONE ``chat()`` call.
-
-    ``text`` is streamed through ``on_text`` and returned as the assistant
-    message content. ``reasoning`` goes to ``on_reasoning`` only - it must never
-    leak into the answer, and asserting that is one of this double's jobs.
-
-    ``tool_calls`` makes the agent loop run tools and come back for another turn;
-    an empty tuple ends the loop.
-
-    ``error``, when set, raises :class:`ProviderError` instead of answering, so
-    error/recovery paths are testable without simulating a network fault.
-
-    ``chunk_size`` > 0 splits ``text`` into fixed-size pieces to exercise
-    chunk-boundary handling in stream consumers (the ``<think>`` splitter and the
-    UI's incremental markdown renderer both have boundary logic worth covering).
-    """
-
-    text: str = ""
-    reasoning: str = ""
-    tool_calls: Sequence[ToolCallScript] = ()
-    error: Optional[str] = None
-    chunk_size: int = 0
-
-
-@dataclass
-class RecordedCall:
-    """A snapshot of one ``chat()`` invocation, for assertions after the fact."""
-
-    messages: List[Dict[str, Any]]
-    tool_names: List[str]
-    cancelled: bool = False
+from providers.base import CancelFn, Provider, ProviderError, TextCallback, ToolSpec
 
 
 class FakeProvider(Provider):
-    """A ``Provider`` that replays :class:`ScriptedTurn` objects.
-
-    Args:
-        turns: the scripted turns, consumed in order.
-        model: the model id reported through ``describe()`` / usage records.
-        models: what :meth:`list_models` returns (Settings' "Load models").
-        strict: when True (default) running past the end of the script raises
-            ``AssertionError``. That is intentional noise: a silent extra turn
-            usually means the code under test looped more than the test author
-            expected, and hiding it behind an empty answer would turn a real
-            behaviour change into a passing test.
-    """
+    """Deterministic test double mimicking real LLM Providers (OpenAI, Anthropic, Ollama)."""
 
     name = "fake"
-    # The double can accept image content blocks, so vision code paths are
-    # reachable in tests without a real vision-capable gateway.
     supports_vision = True
 
-    def __init__(
+    def __init__(self, conf: Optional[Dict[str, Any]] = None) -> None:
+        # Initialize base provider with default configuration if none provided
+        super().__init__(conf or {"model": "fake-model-v1"})
+        # History of all message batches sent across all chat calls
+        self.call_history: List[List[Dict[str, Any]]] = []
+        # Queue of programmed assistant responses to return sequentially
+        self.response_queue: List[Dict[str, Any]] = []
+        # Queue of exceptions to raise on corresponding calls
+        self.error_queue: List[Exception] = []
+        # Default text returned when response queue is empty
+        self.default_text: str = "Fake model response."
+        # Total number of chat invocations
+        self.call_count: int = 0
+        # Recorded tool specs passed into each turn
+        self.last_tools: Optional[List[ToolSpec]] = None
+
+    def queue_response(
         self,
-        turns: Optional[Sequence[ScriptedTurn]] = None,
-        *,
-        model: str = "fake-model",
-        models: Optional[Sequence[str]] = None,
-        strict: bool = True,
-        conf: Optional[Dict[str, Any]] = None,
-    ) -> None:
-        super().__init__(dict(conf or {}, model=model))
-        self._turns: List[ScriptedTurn] = list(turns or [])
-        self._models = list(models or [model])
-        self._strict = strict
-        self._ids = itertools.count(1)  # deterministic tool-call ids: call_1, call_2, ...
-        self.calls: List[RecordedCall] = []
+        content: str = "",
+        tool_calls: Optional[List[Dict[str, Any]]] = None,
+        reasoning: Optional[str] = None,
+        chunks: Optional[List[str]] = None,
+    ) -> FakeProvider:
+        """Enqueue a pre-configured response structure for upcoming chat turns."""
+        self.response_queue.append({
+            "content": content,
+            "tool_calls": tool_calls or [],
+            "reasoning": reasoning,
+            "chunks": chunks or ([content] if content else []),
+        })
+        return self
 
-    # -- introspection helpers used by tests ---------------------------- #
-    @property
-    def call_count(self) -> int:
-        """How many times the layer above asked this provider to run a turn."""
-        return len(self.calls)
+    def queue_error(self, exc: Exception) -> FakeProvider:
+        """Enqueue an exception to simulate network/API errors on the next turn."""
+        self.error_queue.append(exc)
+        return self
 
-    @property
-    def remaining_turns(self) -> int:
-        """Scripted turns not consumed yet - assert 0 to prove the script was
-        fully used (an unused turn means the code stopped earlier than intended)."""
-        return len(self._turns)
-
-    def last_messages(self) -> List[Dict[str, Any]]:
-        """The message list sent on the most recent call (empty if never called)."""
-        return self.calls[-1].messages if self.calls else []
-
-    # -- Provider contract ---------------------------------------------- #
     def chat(
         self,
         messages: List[Dict[str, Any]],
@@ -131,83 +61,53 @@ class FakeProvider(Provider):
         cancel: Optional[CancelFn] = None,
         on_reasoning: Optional[TextCallback] = None,
     ) -> Dict[str, Any]:
-        """Replay the next scripted turn, honouring cancel and both callbacks.
+        """Simulate single LLM turn with full streaming and tool-call support."""
+        self.call_count += 1
+        self.call_history.append([dict(m) for m in messages])
+        self.last_tools = tools
 
-        The message list is deep-ish copied into the recording because the agent
-        loop keeps appending to the SAME list object; without the copy every
-        recorded call would show the final state and assertions on "what was
-        sent at step 1" would be meaningless.
-        """
-        record = RecordedCall(
-            messages=[dict(m) for m in messages],
-            tool_names=[t.name for t in (tools or [])],
-        )
-        self.calls.append(record)
+        # 1. Check for injected errors
+        if self.error_queue:
+            raise self.error_queue.pop(0)
 
-        turn = self._next_turn()
+        # 2. Check early cancellation before processing
+        if cancel and cancel():
+            raise ProviderError("Execution aborted by user cancel signal before response generation.")
 
-        # Checked before streaming anything: a provider that already knows the
-        # caller gave up must not spend callbacks on text nobody will render.
-        if self._is_cancelled(cancel):
-            record.cancelled = True
-            return {"role": "assistant", "content": "", "tool_calls": []}
+        # 3. Retrieve queued response or construct default response
+        if self.response_queue:
+            resp_spec = self.response_queue.pop(0)
+            content = resp_spec.get("content", "")
+            tool_calls = resp_spec.get("tool_calls", [])
+            reasoning = resp_spec.get("reasoning")
+            chunks = resp_spec.get("chunks", [content] if content else [])
+        else:
+            content = self.default_text
+            tool_calls = []
+            reasoning = None
+            chunks = [content]
 
-        if turn.error:
-            raise ProviderError(turn.error)
+        # 4. Stream reasoning chunks if provided
+        if reasoning and on_reasoning:
+            on_reasoning(reasoning)
 
-        if turn.reasoning and on_reasoning:
-            on_reasoning(turn.reasoning)
+        # 5. Stream text chunks, checking cancellation between fragments
+        for chunk in chunks:
+            if cancel and cancel():
+                raise ProviderError("Execution cancelled during text chunk streaming.")
+            if on_text and chunk:
+                on_text(chunk)
 
-        for piece in self._stream_pieces(turn):
-            # Re-checked between chunks so a mid-stream Stop truncates the answer
-            # the same way a real streamed response does.
-            if self._is_cancelled(cancel):
-                record.cancelled = True
-                break
-            if on_text:
-                on_text(piece)
-
-        return {
+        # 6. Return canonical assistant message payload
+        assistant_msg: Dict[str, Any] = {
             "role": "assistant",
-            "content": turn.text,
-            "tool_calls": [
-                {"id": f"call_{next(self._ids)}", "name": name, "arguments": dict(args)}
-                for name, args in turn.tool_calls
-            ],
+            "content": content,
         }
+        if tool_calls:
+            assistant_msg["tool_calls"] = tool_calls
+
+        return assistant_msg
 
     def list_models(self) -> List[str]:
-        """Configured model ids. Clears ``last_error`` so ``test_connection()``
-        reports success, matching how a healthy real provider behaves."""
-        self.last_error = ""
-        return list(self._models)
-
-    # -- internals ------------------------------------------------------- #
-    def _next_turn(self) -> ScriptedTurn:
-        """Pop the next scripted turn, or fail loudly when the script ran out."""
-        if self._turns:
-            return self._turns.pop(0)
-        if self._strict:
-            raise AssertionError(
-                f"FakeProvider script exhausted: chat() was called {len(self.calls)} "
-                "time(s) but fewer turns were scripted. Add a ScriptedTurn, or pass "
-                "strict=False if the extra call is genuinely expected."
-            )
-        return ScriptedTurn()
-
-    @staticmethod
-    def _stream_pieces(turn: ScriptedTurn) -> List[str]:
-        """Split a turn's answer into the fragments to stream.
-
-        ``chunk_size == 0`` streams the whole answer in one piece (the common
-        case); a positive size slices it so tests can drive chunk-boundary logic.
-        """
-        if not turn.text:
-            return []
-        if turn.chunk_size <= 0:
-            return [turn.text]
-        size = turn.chunk_size
-        return [turn.text[i:i + size] for i in range(0, len(turn.text), size)]
-
-
-__all__ = ["FakeProvider", "ScriptedTurn", "RecordedCall"]
+        """Return available mock models for settings and validation tests."""
+        return ["fake-model-v1", "fake-reasoner-pro", "fake-vision-plus"]