EPIC R05 (Team Hoa) - one security/approval path for every tool call.
R05-T01 domain/tools/{tool_descriptor,tool_registry}.py
ToolCapability (READ/WRITE/EXECUTE/NETWORK, composable) + ToolDescriptor +
ToolRegistry, replacing three independently-maintained gating lists
(core/tools.py::WRITE_TOOLS, code_agent.py's WRITE_TOOLS|MS365_WRITE_TOOLS,
chat_agent.py's literal ("run_command","install_package") tuple) with one
capability lookup.
R05-T02 infrastructure/filesystem/{file_tools,command_tools,fetch_tools,tool_context}.py
core/tools.py's execute_tool if/elif chain split into per-concern modules.
core/tools.py is now a strangler-fig shim: re-exports ToolContext/ToolError,
dispatches through a {name: handler} dict built from the split modules.
core/tools.py: 566 -> 291 lines.
R05-T03 application/conversations/tool_policy_gateway.py
ToolPolicyGateway.allow(name, gate, payload) - capability-driven ALLOW vs
ask-the-gate decision. Wired into both chat_agent.py::run_cowork and
code_agent.py::run_code, replacing their separate hand-rolled checks.
Verified equivalent to the old hardcoded sets by test.
R05-T04 (behavior change, not just refactor)
MCP/connector tools (core/mcp_client.py, core/ext_connectors.py) reached
chat_agent.py via extra_executor(name, args) with NO permission check at
all. They are now tagged with a conservative default capability
(WRITE|EXECUTE|NETWORK - no MCP tool self-declares risk) and routed through
the SAME ToolPolicyGateway as built-ins. When "confirm before running
commands" is on, MCP/connector calls now prompt like run_command already
did - a real gap closed, and a user-visible change worth calling out.
R05-T05 infrastructure/mcp/mcp_source_manager.py
McpToolSourceManager extracts the connection cache/lock/start-or-skip
lifecycle out of state.py::AppContext (_mcp_connections/_conn_lock) into a
standalone, directly-testable class. AppContext.build_mcp_tools and
_ms365_builtin_connection now call ensure()/stop(); _ext_connections
(unified Connectors) is out of scope for this task and keeps its own lock.
New tests: tests/unit/test_tool_registry_and_policy.py,
test_code_agent_tool_policy.py, test_cowork_extra_tool_policy.py,
test_mcp_source_manager.py (26 new tests).
Suite: 254 passed, 4 pre-existing failures unrelated to R05 (2 EPIC R02
config-security, 2 environment-dependent routing tests - see checklist).
check_imports: PASS. All new files < 400 LOC.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
108 lines
4.6 KiB
Python
108 lines
4.6 KiB
Python
"""Unit tests for EPIC R05: the tool descriptor/registry (R05-T01), the split
|
|
built-in handlers (R05-T02), and the policy gateway (R05-T03).
|
|
|
|
The gateway tests assert the SAME capability set each engine used to hard-code
|
|
as a name tuple still gets gated after the switch to capability lookup — that
|
|
equivalence is the whole point of R05-T03, not an incidental detail.
|
|
"""
|
|
from __future__ import annotations
|
|
|
|
from typing import Any, Dict
|
|
|
|
import pytest
|
|
|
|
from cowork_local.application.conversations import ToolPolicyGateway
|
|
from cowork_local.core.tools import TOOL_SPECS, ToolContext, execute_tool
|
|
from cowork_local.domain.tools import ToolCapability, ToolDescriptor, default_registry
|
|
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# R05-T01 - ToolDescriptor / ToolRegistry
|
|
# --------------------------------------------------------------------------- #
|
|
def test_capability_flags_compose():
|
|
install = ToolDescriptor("install_package", "", {}, ToolCapability.WRITE | ToolCapability.EXECUTE)
|
|
assert install.has(ToolCapability.WRITE)
|
|
assert install.has(ToolCapability.EXECUTE)
|
|
assert not install.has(ToolCapability.NETWORK)
|
|
|
|
|
|
def test_default_registry_matches_todays_hardcoded_gating_sets():
|
|
"""The two literal sets this EPIC replaces:
|
|
``core/tools.py::WRITE_TOOLS`` and ``core/chat_agent.py``'s
|
|
``("run_command", "install_package")`` tuple. The registry must agree
|
|
with both, or the capability switch silently changes who gets gated."""
|
|
registry = default_registry(TOOL_SPECS)
|
|
|
|
execute_gated = {d.name for d in registry.all() if d.has(ToolCapability.EXECUTE)}
|
|
assert execute_gated == {"run_command", "install_package"}
|
|
|
|
write_gated = {d.name for d in registry.all() if d.has(ToolCapability.WRITE)}
|
|
assert write_gated == {"write_file", "edit_file", "install_package"}
|
|
|
|
|
|
def test_unregistered_tool_has_no_capabilities():
|
|
registry = default_registry(TOOL_SPECS)
|
|
assert registry.capabilities_for("no_such_tool") is ToolCapability.NONE
|
|
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# R05-T02 - core/tools.py dispatch, now built from the split infra modules
|
|
# --------------------------------------------------------------------------- #
|
|
def test_execute_tool_still_dispatches_every_built_in(tmp_path):
|
|
ctx = ToolContext(tmp_path)
|
|
written = execute_tool(ctx, "write_file", {"path": "a.txt", "content": "hi"})
|
|
assert written["ok"] is True
|
|
read = execute_tool(ctx, "read_file", {"path": "a.txt"})
|
|
assert read == {"ok": True, "output": "hi"}
|
|
edited = execute_tool(ctx, "edit_file", {"path": "a.txt", "old_string": "hi", "new_string": "bye"})
|
|
assert edited["ok"] is True
|
|
assert execute_tool(ctx, "read_file", {"path": "a.txt"})["output"] == "bye"
|
|
listing = execute_tool(ctx, "list_dir", {})
|
|
assert listing["ok"] is True and "a.txt" in listing["output"]
|
|
|
|
|
|
def test_execute_tool_reports_unknown_name(tmp_path):
|
|
ctx = ToolContext(tmp_path)
|
|
result = execute_tool(ctx, "not_a_real_tool", {})
|
|
assert result == {"ok": False, "output": "Tool not found: not_a_real_tool"}
|
|
|
|
|
|
# --------------------------------------------------------------------------- #
|
|
# R05-T03 - ToolPolicyGateway
|
|
# --------------------------------------------------------------------------- #
|
|
class _RecordingGate:
|
|
def __init__(self, approve: bool):
|
|
self.approve = approve
|
|
self.calls: list = []
|
|
|
|
def request(self, payload: Dict[str, Any]) -> bool:
|
|
self.calls.append(payload)
|
|
return self.approve
|
|
|
|
|
|
@pytest.fixture
|
|
def cowork_policy() -> ToolPolicyGateway:
|
|
"""Same construction as ``core/chat_agent.py``'s module-level
|
|
``_COWORK_TOOL_POLICY`` - EXECUTE is exactly what Cowork used to gate via
|
|
the literal ``("run_command", "install_package")`` tuple."""
|
|
return ToolPolicyGateway(default_registry(TOOL_SPECS), ToolCapability.EXECUTE)
|
|
|
|
|
|
def test_no_gate_means_auto_run(cowork_policy):
|
|
assert cowork_policy.allow("run_command", None, {}) is True
|
|
|
|
|
|
def test_read_only_tool_never_asks_the_gate(cowork_policy):
|
|
gate = _RecordingGate(approve=False) # would reject if asked
|
|
assert cowork_policy.allow("write_file", gate, {}) is True
|
|
assert gate.calls == [] # never consulted - write_file isn't EXECUTE
|
|
|
|
|
|
def test_gated_capability_consults_the_gate_and_honors_its_answer(cowork_policy):
|
|
approving = _RecordingGate(approve=True)
|
|
assert cowork_policy.allow("run_command", approving, {"name": "run_command"}) is True
|
|
assert approving.calls == [{"name": "run_command"}]
|
|
|
|
rejecting = _RecordingGate(approve=False)
|
|
assert cowork_policy.allow("install_package", rejecting, {}) is False
|