create feature mcp-issue-context
CI / test (pull_request) Canceled after 0s

This commit is contained in:
2026-08-25 22:54:56 +09:00
parent 86c27e2e79
commit 98f4a1ed77
2 changed files with 1018 additions and 2 deletions
+372 -2
View File
@@ -1,11 +1,124 @@
"""Provider boundary owned with get_project_issue_context."""
"""Provider boundary owned with get_project_issue_context.
Read-only adapter that maps a Gitea issue/pull-request onto the neutral
``IssueContextOutput`` schema declared in ``tools/issue_context.py``. Nothing
here changes the shared runtime, the registry, or the tool schema — this
module only fills in ``build_provider`` so ``runtime.py``'s
``PROVIDER_FACTORIES["get_project_issue_context"]`` resolves to a real,
read-only Gitea client instead of :class:`UnconfiguredIssueProvider`.
Design notes (documented here so a reviewer does not have to guess):
* Credentials (``GITEA_TOKEN``) and the base URL are read INSIDE
``build_provider``/``_resolve_target`` only — never at import time — because
``build_provider`` is only ever invoked by
``ProjectProviderResolver.resolve`` (see ``runtime.py``), which
``server.py::dispatch`` calls strictly AFTER ``policy.decide`` returns
``True``. Reading env vars at module scope would read them before the
policy check ever runs.
* The provider is bound to exactly one project/repository at construction
time (``identity.project`` -> ``PROJECT_CONTEXT_REPO_MAP``). The
``project_id`` argument received per call is only used to echo it back in
the response and as a defensive equality check — it is never used to pick
which repository to query. This means a caller can never redirect the
provider to an arbitrary repository by tampering with ``project_id``.
* ``revision`` pins to the issue's own ``updated_at`` timestamp. Gitea issues
(unlike commits/PRs) have no natural commit SHA of their own, so the most
meaningful, verifiable "version marker" Gitea offers for an issue is its
last-modified time: given the same URL and the same ``revision`` string, a
reviewer can confirm whether the issue has changed since this was fetched.
* ``related`` items are derived from BARE ``#<number>`` cross-references
found in the issue body — never from a ``#<number>`` that is only the
label text of an explicit Markdown link (``[other-repo PR #4](http://.../
other-repo/pulls/4)``). A real issue body regularly links to a pull
request or an issue in a DIFFERENT repository this way; re-guessing that
as "issue #4 in THIS repo" would silently point at the wrong resource
entirely, which is worse than omitting it — the author already gave the
correct URL right there, so extraction strips whole ``[label](url)``
spans, then any remaining bare URLs, before scanning for mentions. Titles
for the mentions that DO survive are NOT fetched with an extra API call
(kept deliberately simple and bounded) — each related item's title is a
generic, honest placeholder ("Referenced item #<n>"), and its ``url``
always resolves to a real, openable page in this same repository. This
is a known, documented simplification, not a bug.
* ``acceptance_criteria`` looks for the issue's own "Acceptance Criteria"
Markdown heading (``#`` through ``######``, case-insensitive) and only
collects checklist items (``- [ ]``/``- [x]``) from inside that section,
up to the next heading of equal-or-shallower depth. This matters because
a real work-item body commonly has a SEPARATE "Definition of Done"
section that also uses checklist syntax — without scoping to the
matching heading, those unrelated items would be silently folded into
"acceptance criteria" and misrepresent the issue. When no such heading
exists at all (ad-hoc issues with no fixed template), this falls back to
scanning the whole body, so those issues still get a best-effort result
instead of an always-empty list.
* ``truncated``/``returned``/``remaining``/``next_cursor`` describe pagination
over the ``related`` list only. ``description`` has its own independent,
generous hard safety cap (so the response stays bounded even for a
pathological issue body) and is never silently cut without a visible
notice appended.
* The regexes that derive ``acceptance_criteria``/``related`` never scan more
than ``_MAX_SCAN_CHARS`` of the raw body, and always run on a copy with
``http(s)://`` links stripped first — otherwise a `#42` inside a URL
fragment (e.g. a doc anchor) would be mis-detected as a cross-reference to
issue #42, and a pathologically large body would cost unbounded regex work
on every call regardless of the display cap on ``description``.
"""
from __future__ import annotations
import json
import os
import re
from dataclasses import dataclass
from datetime import datetime, timezone
from typing import Any, Protocol
import requests
from ..foundation import IdentityContext, ProviderError
# ---- tunables (documented, not hardcoded secrets) -------------------------
_REQUEST_TIMEOUT_SECONDS = 10
_STANDARD_RELATED_PAGE_SIZE = 20
_FULL_RELATED_PAGE_SIZE = 100
_SUMMARY_DESCRIPTION_CHARS = 280
_MAX_DESCRIPTION_CHARS = 20_000
_MAX_SCAN_CHARS = 200_000 # hard cap on regex work, independent of the display cap above
_TRUNCATION_NOTICE = "\n\n[description truncated: exceeds the display size limit]"
_ISSUE_KEY_PATTERN = re.compile(r"^[1-9][0-9]*$")
_CHECKLIST_PATTERN = re.compile(r"^[-*]\s+\[[ xX]\]\s+(.+)$", re.MULTILINE)
_MENTION_PATTERN = re.compile(r"(?<!\w)#([1-9][0-9]*)\b")
_URL_PATTERN = re.compile(r"https?://\S+")
# A whole Markdown link span, label + target together — stripped as ONE unit
# so a `#<number>` that is only the link's label text (often a cross-repo or
# pull-request reference) is never re-guessed as a same-repo issue mention.
_MARKDOWN_LINK_PATTERN = re.compile(r"\[[^\]]*\]\([^)]*\)")
# ATX heading line, e.g. "# Acceptance Criteria" / "## Acceptance Criteria".
_HEADING_PATTERN = re.compile(r"^(#{1,6})[ \t]+(.+?)\s*$", re.MULTILINE)
_ACCEPTANCE_HEADING_NAMES = ("acceptance criteria",)
def _extract_heading_section(text: str, heading_names: tuple[str, ...]) -> str | None:
"""Return the body of the first ATX heading whose title case-insensitively
matches one of ``heading_names``, up to the next heading of equal or
shallower depth (or the end of ``text``). Returns ``None`` when no such
heading exists, so the caller can fall back to the whole body."""
wanted = {name.strip().lower() for name in heading_names}
headings = list(_HEADING_PATTERN.finditer(text))
for index, match in enumerate(headings):
if match.group(2).strip().lower() not in wanted:
continue
level = len(match.group(1))
end = len(text)
for later in headings[index + 1 :]:
if len(later.group(1)) <= level:
end = later.start()
break
return text[match.end() : end]
return None
class IssueProvider(Protocol):
def get_issue_context(self, **arguments: Any) -> dict[str, Any]: ...
@@ -20,6 +133,263 @@ class UnconfiguredIssueProvider:
)
@dataclass(frozen=True)
class _GiteaRepoTarget:
base_url: str
owner: str
repo: str
token: str
project_id: str
def _load_repo_map() -> dict[str, str]:
raw = os.environ.get("PROJECT_CONTEXT_REPO_MAP", "").strip()
if not raw:
return {}
try:
parsed = json.loads(raw)
except json.JSONDecodeError as exc:
raise ProviderError(
"UNAVAILABLE",
"PROJECT_CONTEXT_REPO_MAP is not valid JSON.",
retryable=False,
) from exc
if not isinstance(parsed, dict) or not all(
isinstance(k, str) and isinstance(v, str) for k, v in parsed.items()
):
raise ProviderError(
"UNAVAILABLE",
"PROJECT_CONTEXT_REPO_MAP must be a JSON object of project_id -> 'owner/repo'.",
retryable=False,
)
return parsed
def _resolve_target(identity: IdentityContext) -> _GiteaRepoTarget:
"""Read-only credential/target resolution. Only ever called AFTER the
runtime's policy has already granted ALLOW for this identity/tool."""
base_url = os.environ.get("GITEA_BASE_URL", "").strip().rstrip("/")
token = os.environ.get("GITEA_TOKEN", "").strip()
if not base_url or not token:
raise ProviderError(
"UNAVAILABLE",
"GITEA_BASE_URL/GITEA_TOKEN are not configured for this environment.",
retryable=False,
)
repo_map = _load_repo_map()
slug = repo_map.get(identity.project, "")
if not slug or "/" not in slug:
raise ProviderError(
"UNAVAILABLE",
"This project is not mapped to an approved Gitea repository.",
retryable=False,
)
owner, _, repo = slug.partition("/")
if not owner or not repo:
raise ProviderError(
"UNAVAILABLE",
"This project's Gitea repository mapping is malformed.",
retryable=False,
)
return _GiteaRepoTarget(
base_url=base_url, owner=owner, repo=repo, token=token, project_id=identity.project,
)
def build_provider(identity: IdentityContext) -> IssueProvider:
"""Replace only this factory when wiring the approved read-only issue adapter."""
return UnconfiguredIssueProvider()
return GiteaIssueProvider(_resolve_target(identity))
class GiteaIssueProvider:
"""Read-only adapter mapping one Gitea issue/PR onto the neutral schema."""
def __init__(self, target: _GiteaRepoTarget) -> None:
self._target = target
def get_issue_context(
self,
*,
project_id: str,
issue_key: str,
detail: str,
cursor: str | None,
**_: Any,
) -> dict[str, Any]:
if project_id != self._target.project_id:
# Defense in depth: the runtime's policy already guarantees this
# can never happen (DENIED would have fired first), but the
# provider never trusts caller-supplied routing regardless.
raise ProviderError(
"INTERNAL",
"Resolved provider does not match the requested project.",
retryable=False,
)
if not _ISSUE_KEY_PATTERN.match(issue_key):
raise ProviderError(
"INVALID_INPUT",
"issue_key must be a positive work item number.",
retryable=False,
)
offset = self._decode_cursor(cursor)
payload = self._fetch_issue(issue_key)
title = str(payload.get("title") or "")
raw_state = str(payload.get("state") or "")
status = raw_state if raw_state in {"open", "closed"} else "unknown"
body = str(payload.get("body") or "")
description = self._build_description(body, detail)
# Bounded regardless of the actual body size: caps worst-case regex
# cost, independently of `description`'s own display-only cap.
scan_text = body[:_MAX_SCAN_CHARS]
acceptance_section = _extract_heading_section(scan_text, _ACCEPTANCE_HEADING_NAMES)
acceptance_criteria = tuple(
_CHECKLIST_PATTERN.findall(acceptance_section if acceptance_section is not None else scan_text)
)
related_all = self._extract_related(scan_text, issue_key)
related_page, returned, remaining, truncated, next_cursor = self._paginate_related(
related_all, detail, offset,
)
html_url = str(
payload.get("html_url")
or f"{self._target.base_url}/{self._target.owner}/{self._target.repo}/issues/{issue_key}"
)
updated_at = str(payload.get("updated_at") or "")
retrieved_at = datetime.now(timezone.utc).isoformat()
return {
"project_id": project_id,
"issue_key": issue_key,
"title": title,
"status": status,
"description": description,
"acceptance_criteria": acceptance_criteria,
"related": related_page,
"source": {
"system": "gitea",
"url": html_url,
"revision": f"issue-updated:{updated_at or retrieved_at}",
"retrieved_at": retrieved_at,
},
"truncated": truncated,
"returned": returned,
"remaining": remaining,
"next_cursor": next_cursor,
}
# ---- internals ---------------------------------------------------
def _decode_cursor(self, cursor: str | None) -> int:
if cursor is None:
return 0
try:
offset = int(cursor)
except ValueError as exc:
raise ProviderError("INVALID_INPUT", "cursor is not valid.", retryable=False) from exc
if offset < 0:
raise ProviderError("INVALID_INPUT", "cursor is not valid.", retryable=False)
return offset
def _build_description(self, body: str, detail: str) -> str:
text = body.strip()
if detail == "summary":
return text.split("\n\n", 1)[0][:_SUMMARY_DESCRIPTION_CHARS]
if len(text) > _MAX_DESCRIPTION_CHARS:
return text[:_MAX_DESCRIPTION_CHARS] + _TRUNCATION_NOTICE
return text
def _extract_related(self, body: str, issue_key: str) -> tuple[dict[str, str], ...]:
# Strip whole `[label](url)` spans FIRST (as one unit) so a `#<number>`
# that only appears as a Markdown link's label — often a cross-repo or
# pull-request reference with its own, possibly different, URL right
# there — is never re-guessed as "issue #<number> in this repo".
text_without_links = _MARKDOWN_LINK_PATTERN.sub(" ", body)
# Then strip any remaining bare URLs so a doc-anchor link like
# ".../guide#42" is never mistaken for a cross-reference to issue #42.
text_without_urls = _URL_PATTERN.sub(" ", text_without_links)
numbers = sorted({int(n) for n in _MENTION_PATTERN.findall(text_without_urls) if n != issue_key})
return tuple(
{
"item_id": str(number),
"relation": "mentioned",
"title": f"Referenced item #{number}",
"url": f"{self._target.base_url}/{self._target.owner}/{self._target.repo}/issues/{number}",
}
for number in numbers
)
def _paginate_related(
self,
related_all: tuple[dict[str, str], ...],
detail: str,
offset: int,
) -> tuple[tuple[dict[str, str], ...], int, int, bool, str | None]:
if detail == "summary":
# Summary mode intentionally omits related items outright; it is
# not a size-limit truncation, so callers who need them must
# call again with detail="standard"/"full".
remaining = len(related_all)
return (), 0, remaining, remaining > 0, None
page_size = _FULL_RELATED_PAGE_SIZE if detail == "full" else _STANDARD_RELATED_PAGE_SIZE
page = related_all[offset : offset + page_size]
remaining = max(0, len(related_all) - (offset + page_size))
truncated = remaining > 0
next_cursor = str(offset + page_size) if truncated else None
return page, len(page), remaining, truncated, next_cursor
def _fetch_issue(self, issue_key: str) -> dict[str, Any]:
url = (
f"{self._target.base_url}/api/v1/repos/{self._target.owner}/"
f"{self._target.repo}/issues/{issue_key}"
)
headers = {"Authorization": f"token {self._target.token}"}
try:
response = requests.get(url, headers=headers, timeout=_REQUEST_TIMEOUT_SECONDS)
except requests.exceptions.Timeout as exc:
raise ProviderError(
"UPSTREAM_TIMEOUT", "The Gitea request timed out.", retryable=True,
) from exc
except requests.exceptions.RequestException as exc:
# Never surface str(exc) — it can embed the request URL/host and,
# in some transport errors, request headers.
raise ProviderError(
"UPSTREAM_ERROR", "The Gitea request failed.", retryable=True,
) from exc
if response.status_code == 404:
raise ProviderError(
"NOT_FOUND",
"The work item was not found or is not accessible.",
retryable=False,
)
if response.status_code == 429:
raise ProviderError("RATE_LIMITED", "Gitea rate-limited this request.", retryable=True)
if response.status_code in (401, 403):
raise ProviderError(
"UPSTREAM_ERROR",
"The read-only Gitea credential could not access the repository.",
retryable=False,
)
if response.status_code >= 500:
raise ProviderError("UPSTREAM_ERROR", "Gitea returned a server error.", retryable=True)
if response.status_code != 200:
raise ProviderError(
"UPSTREAM_ERROR", "Gitea returned an unexpected response.", retryable=False,
)
try:
data = response.json()
except ValueError as exc:
raise ProviderError(
"UPSTREAM_ERROR",
"Gitea returned a response that could not be parsed.",
retryable=False,
) from exc
if not isinstance(data, dict):
raise ProviderError(
"UPSTREAM_ERROR", "Gitea returned an unexpected response shape.", retryable=False,
)
return data
+646
View File
@@ -0,0 +1,646 @@
"""Member A's own test suite for get_project_issue_context.
Every test mocks the Gitea transport (``requests.get``) and never touches a
real network call or a real credential — per Issue #3 / MCP Contract v2:
unit tests must not call Gitea for real or use a real token.
"""
from __future__ import annotations
from dataclasses import dataclass
from typing import Any
import pytest
import requests
from cowork_local.mcp_servers.project_context.foundation import (
IdentityContext,
ProjectContextRuntime,
ProviderError,
)
from cowork_local.mcp_servers.project_context.providers.issue import (
GiteaIssueProvider,
UnconfiguredIssueProvider,
_GiteaRepoTarget,
)
from cowork_local.mcp_servers.project_context.runtime import ProjectProviderResolver
from cowork_local.mcp_servers.project_context.server import dispatch
FAKE_TOKEN = "super-secret-token-value" # noqa: S105 - test-only sentinel, never a real credential
# ---------------------------------------------------------------------------
# Shared fixtures / test doubles
# ---------------------------------------------------------------------------
@dataclass
class RecordingPolicy:
allowed: bool
calls: int = 0
def decide(self, identity: IdentityContext, tool_name: str, project_id: str) -> bool:
self.calls += 1
return self.allowed
@dataclass
class RecordingResolver:
provider: Any
calls: int = 0
def resolve(self, identity: IdentityContext, tool_name: str) -> Any:
self.calls += 1
return self.provider
class _FakeResponse:
def __init__(self, status_code: int, json_body: Any = "__missing__") -> None:
self.status_code = status_code
self._json_body = json_body
def json(self) -> Any:
if self._json_body == "__missing__":
raise ValueError("no json body")
return self._json_body
class _FakeTransport:
"""Drop-in replacement for ``requests.get`` that queues canned results
and records every call it received (url/headers/timeout)."""
def __init__(self, queue: list[Any]) -> None:
self._queue = list(queue)
self.calls: list[dict[str, Any]] = []
def __call__(self, url: str, headers: dict[str, str] | None = None, timeout: float | None = None):
self.calls.append({"url": url, "headers": headers, "timeout": timeout})
item = self._queue.pop(0)
if isinstance(item, BaseException):
raise item
return item
@pytest.fixture
def identity() -> IdentityContext:
return IdentityContext(
actor_id="member-a",
org_unit="fsg",
customer="internal",
project="cowork-local",
granted_scopes=frozenset({"read"}),
)
def _target(**overrides: Any) -> _GiteaRepoTarget:
base = dict(
base_url="http://example.test",
owner="gitea-admin",
repo="cowork-local",
token=FAKE_TOKEN,
project_id="cowork-local",
)
base.update(overrides)
return _GiteaRepoTarget(**base)
def _issue_payload(**overrides: Any) -> dict[str, Any]:
payload = {
"title": "MCP pilot",
"state": "open",
"body": "Build verifiable project context.\n\n- [ ] Every result has a source.",
"html_url": "http://example.test/gitea-admin/cowork-local/issues/1",
"updated_at": "2026-08-20T10:00:00Z",
}
payload.update(overrides)
return payload
def _runtime(
identity: IdentityContext, provider: Any, *, allowed: bool = True,
) -> tuple[ProjectContextRuntime, RecordingPolicy, RecordingResolver]:
policy = RecordingPolicy(allowed=allowed)
resolver = RecordingResolver(provider=provider)
return (
ProjectContextRuntime(identity=identity, policy=policy, credential_resolver=resolver),
policy,
resolver,
)
# ---------------------------------------------------------------------------
# Happy path
# ---------------------------------------------------------------------------
def test_happy_path_returns_full_schema_with_openable_source(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
transport = _FakeTransport([_FakeResponse(200, _issue_payload())])
monkeypatch.setattr(requests, "get", transport)
provider = GiteaIssueProvider(_target())
app, policy, resolver = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context",
{"project_id": "cowork-local", "issue_key": "1", "detail": "standard"},
app,
)
assert result.ok is True
assert policy.calls == 1
assert resolver.calls == 1
assert result.payload["project_id"] == "cowork-local"
assert result.payload["issue_key"] == "1"
assert result.payload["title"] == "MCP pilot"
assert result.payload["status"] == "open"
assert result.payload["acceptance_criteria"] == ["Every result has a source."]
assert result.payload["correlation_id"]
source = result.payload["source"]
assert source["system"] == "gitea"
assert source["url"].startswith("http://example.test/gitea-admin/cowork-local/issues/1")
assert source["revision"] == "issue-updated:2026-08-20T10:00:00Z"
assert source["retrieved_at"]
# exactly one Gitea call was made, to the expected REST path
assert len(transport.calls) == 1
assert transport.calls[0]["url"].endswith("/api/v1/repos/gitea-admin/cowork-local/issues/1")
assert transport.calls[0]["headers"] == {"Authorization": f"token {FAKE_TOKEN}"}
def test_source_fields_are_all_present_and_well_formed(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, _issue_payload())]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app,
)
source = result.payload["source"]
assert source["url"].startswith("http")
assert isinstance(source["revision"], str) and source["revision"]
assert "T" in source["retrieved_at"] # ISO-8601 timestamp, not a placeholder
# ---------------------------------------------------------------------------
# Invalid input (before any policy/provider call)
# ---------------------------------------------------------------------------
def test_invalid_input_is_rejected_before_policy_or_provider(identity: IdentityContext) -> None:
app, policy, resolver = _runtime(identity, UnconfiguredIssueProvider())
result = dispatch("get_project_issue_context", {"project_id": "cowork-local"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "INVALID_INPUT"
assert policy.calls == 0
assert resolver.calls == 0
# ---------------------------------------------------------------------------
# DENIED — zero upstream calls, security-critical
# ---------------------------------------------------------------------------
def test_denied_project_never_resolves_credentials_or_calls_gitea(
identity: IdentityContext,
) -> None:
# No transport is patched at all: if the provider were ever reached it
# would hit the real `requests.get` and fail loudly, so this test also
# proves "zero upstream calls" by construction, not just by call count.
app, policy, resolver = _runtime(identity, UnconfiguredIssueProvider(), allowed=False)
result = dispatch(
"get_project_issue_context",
{"project_id": "some-other-project", "issue_key": "1"},
app,
)
assert result.ok is False
assert result.payload["error"]["code"] == "DENIED"
assert policy.calls == 1
assert resolver.calls == 0
def test_permission_decision_lives_outside_the_tool(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Acceptance criterion: swapping ONLY the policy must change the
outcome, proving `tools/issue_context.py` contains no permission logic
of its own."""
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, _issue_payload())]))
provider = GiteaIssueProvider(_target())
arguments = {"project_id": "cowork-local", "issue_key": "1"}
allowed_app, _, _ = _runtime(identity, provider, allowed=True)
denied_app, _, _ = _runtime(identity, provider, allowed=False)
allowed_result = dispatch("get_project_issue_context", arguments, allowed_app)
denied_result = dispatch("get_project_issue_context", arguments, denied_app)
assert allowed_result.ok is True
assert denied_result.ok is False
assert denied_result.payload["error"]["code"] == "DENIED"
# ---------------------------------------------------------------------------
# Boundary / failure — distinct, non-leaking error codes
# ---------------------------------------------------------------------------
def test_not_found_issue_maps_to_not_found(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(404)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context", {"project_id": "cowork-local", "issue_key": "999999"}, app,
)
assert result.ok is False
assert result.payload["error"]["code"] == "NOT_FOUND"
assert result.payload["error"]["suggested_action"]
def test_provider_raises_provider_error_directly_for_not_found(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Unit-level check on the provider class itself (not only through
dispatch): the raised exception must carry the right `.code`/`.retryable`
for the runtime to map correctly."""
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(404)]))
provider = GiteaIssueProvider(_target())
with pytest.raises(ProviderError) as exc_info:
provider.get_issue_context(
project_id="cowork-local", issue_key="1", detail="standard", cursor=None,
)
assert exc_info.value.code == "NOT_FOUND"
assert exc_info.value.retryable is False
def test_upstream_timeout_maps_to_upstream_timeout(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setattr(requests, "get", _FakeTransport([requests.exceptions.Timeout("slow")]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UPSTREAM_TIMEOUT"
assert result.payload["error"]["retryable"] is True
@pytest.mark.parametrize(
("status_code", "expected_code"),
[(500, "UPSTREAM_ERROR"), (503, "UPSTREAM_ERROR"), (429, "RATE_LIMITED"),
(401, "UPSTREAM_ERROR"), (403, "UPSTREAM_ERROR")],
)
def test_upstream_status_codes_map_to_distinct_error_codes(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch, status_code: int, expected_code: str,
) -> None:
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(status_code)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == expected_code
def test_malformed_gitea_response_maps_to_upstream_error(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, json_body="__missing__")]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UPSTREAM_ERROR"
def test_provider_output_schema_mismatch_maps_to_upstream_error(identity: IdentityContext) -> None:
class BrokenProvider:
def get_issue_context(self, **_: Any) -> dict[str, Any]:
return {"project_id": "cowork-local"} # missing every other required field
app, _, _ = _runtime(identity, BrokenProvider())
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UPSTREAM_ERROR"
# ---------------------------------------------------------------------------
# Reject before any network call
# ---------------------------------------------------------------------------
def test_invalid_issue_key_format_rejected_before_network_call(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
transport = _FakeTransport([]) # empty queue: a real call would raise IndexError
monkeypatch.setattr(requests, "get", transport)
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context", {"project_id": "cowork-local", "issue_key": "not-a-number"}, app,
)
assert result.ok is False
assert result.payload["error"]["code"] == "INVALID_INPUT"
assert transport.calls == []
def test_invalid_cursor_rejected_before_network_call(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
transport = _FakeTransport([])
monkeypatch.setattr(requests, "get", transport)
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context",
{"project_id": "cowork-local", "issue_key": "1", "cursor": "not-a-number"},
app,
)
assert result.ok is False
assert result.payload["error"]["code"] == "INVALID_INPUT"
assert transport.calls == []
# ---------------------------------------------------------------------------
# Fail-closed configuration (build_provider itself, via the real resolver)
# ---------------------------------------------------------------------------
def test_missing_gitea_env_vars_returns_unavailable_with_no_network_call(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.delenv("GITEA_BASE_URL", raising=False)
monkeypatch.delenv("GITEA_TOKEN", raising=False)
monkeypatch.delenv("PROJECT_CONTEXT_REPO_MAP", raising=False)
def _fail_if_called(*_args: Any, **_kwargs: Any) -> Any:
raise AssertionError("Gitea must not be called when the provider is unconfigured")
monkeypatch.setattr(requests, "get", _fail_if_called)
policy = RecordingPolicy(allowed=True)
app = ProjectContextRuntime(identity=identity, policy=policy, credential_resolver=ProjectProviderResolver())
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UNAVAILABLE"
def test_project_without_repo_mapping_returns_unavailable(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
monkeypatch.setenv("GITEA_BASE_URL", "http://example.test")
monkeypatch.setenv("GITEA_TOKEN", FAKE_TOKEN)
monkeypatch.setenv("PROJECT_CONTEXT_REPO_MAP", '{"some-other-project": "gitea-admin/other"}')
policy = RecordingPolicy(allowed=True)
app = ProjectContextRuntime(identity=identity, policy=policy, credential_resolver=ProjectProviderResolver())
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UNAVAILABLE"
@pytest.mark.parametrize(
"raw_map",
[
"{not valid json", # malformed JSON
'["cowork-local", "gitea-admin/cowork-local"]', # valid JSON, wrong shape (array)
'{"cowork-local": 123}', # valid JSON object, non-string value
],
)
def test_malformed_repo_map_returns_unavailable_with_no_network_call(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch, raw_map: str,
) -> None:
monkeypatch.setenv("GITEA_BASE_URL", "http://example.test")
monkeypatch.setenv("GITEA_TOKEN", FAKE_TOKEN)
monkeypatch.setenv("PROJECT_CONTEXT_REPO_MAP", raw_map)
def _fail_if_called(*_args: Any, **_kwargs: Any) -> Any:
raise AssertionError("Gitea must not be called when the repo map is malformed")
monkeypatch.setattr(requests, "get", _fail_if_called)
policy = RecordingPolicy(allowed=True)
app = ProjectContextRuntime(identity=identity, policy=policy, credential_resolver=ProjectProviderResolver())
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UNAVAILABLE"
# ---------------------------------------------------------------------------
# Truncation + cursor pagination over `related`
# ---------------------------------------------------------------------------
def test_truncation_and_cursor_paginate_related_items(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
mentions = " ".join(f"#{n}" for n in range(2, 27)) # 25 distinct related items
payload = _issue_payload(body=f"See also {mentions}.")
transport = _FakeTransport([_FakeResponse(200, payload), _FakeResponse(200, payload)])
monkeypatch.setattr(requests, "get", transport)
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
first = dispatch(
"get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app,
)
assert first.ok is True
assert first.payload["returned"] == 20
assert first.payload["remaining"] == 5
assert first.payload["truncated"] is True
assert first.payload["next_cursor"] == "20"
assert len(first.payload["related"]) == 20
assert first.payload["related"][0]["url"].startswith("http://example.test/")
second = dispatch(
"get_project_issue_context",
{"project_id": "cowork-local", "issue_key": "1", "cursor": first.payload["next_cursor"]},
app,
)
assert second.ok is True
assert second.payload["returned"] == 5
assert second.payload["remaining"] == 0
assert second.payload["truncated"] is False
assert second.payload["next_cursor"] is None
def test_full_detail_uses_a_larger_related_page_than_standard(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Regression guard: `detail='full'` must genuinely page differently
from `detail='standard'` (100 vs 20) — this was previously unverified."""
mentions = " ".join(f"#{n}" for n in range(2, 32)) # 30 distinct related items
payload = _issue_payload(body=f"See also {mentions}.")
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context",
{"project_id": "cowork-local", "issue_key": "1", "detail": "full"},
app,
)
assert result.ok is True
assert result.payload["returned"] == 30
assert result.payload["remaining"] == 0
assert result.payload["truncated"] is False
assert result.payload["next_cursor"] is None
def test_url_fragment_is_not_mistaken_for_a_related_issue(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Regression guard: a doc-anchor link like '.../guide#42' must not be
reported as a related item pointing to issue #42, while a plain '#7'
text mention elsewhere in the same body still must be."""
body = "See http://example.test/gitea-admin/cowork-local/wiki/guide#42 and also #7 directly."
payload = _issue_payload(body=body)
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
related_ids = {item["item_id"] for item in result.payload["related"]}
assert related_ids == {"7"}
def test_related_excludes_number_that_is_only_a_markdown_link_label(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Regression guard (found via a real Gitea issue during manual smoke
testing): a Markdown link whose LABEL happens to contain '#<number>' —
e.g. a cross-repository pull-request reference — must not be re-guessed
as a same-repo issue mention, because that silently points at the wrong
resource. A plain '#9' mention elsewhere in the same body must still be
picked up."""
body = (
"See [other-repo PR #4](http://example.test/other-repo/pulls/4) "
"and also #9 directly."
)
payload = _issue_payload(body=body)
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
related_ids = {item["item_id"] for item in result.payload["related"]}
assert related_ids == {"9"}
def test_acceptance_criteria_is_scoped_to_its_own_heading_not_definition_of_done(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Regression guard (found via a real Gitea issue during manual smoke
testing): a body with a SEPARATE 'Definition of Done' checklist section
must not have those items folded into acceptance_criteria."""
body = (
"# Acceptance Criteria\n\n"
"- [ ] Real acceptance item one.\n"
"- [ ] Real acceptance item two.\n\n"
"# Definition of Done\n\n"
"- [ ] Unrelated DoD item one.\n"
"- [ ] Unrelated DoD item two.\n"
)
payload = _issue_payload(body=body)
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.payload["acceptance_criteria"] == [
"Real acceptance item one.",
"Real acceptance item two.",
]
def test_acceptance_criteria_falls_back_to_whole_body_without_a_heading(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""An issue with no 'Acceptance Criteria' heading at all (no fixed
template) must still get a best-effort result from the whole body,
rather than always coming back empty."""
body = "Ad-hoc issue, no headings.\n\n- [ ] Just do the thing.\n"
payload = _issue_payload(body=body)
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.payload["acceptance_criteria"] == ["Just do the thing."]
def test_summary_detail_omits_related_and_shortens_description(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
long_paragraph = "First paragraph. " * 40 # > 280 chars
payload = _issue_payload(body=f"{long_paragraph}\n\nSecond paragraph mentions #2.")
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(200, payload)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch(
"get_project_issue_context",
{"project_id": "cowork-local", "issue_key": "1", "detail": "summary"},
app,
)
assert result.ok is True
assert len(result.payload["description"]) <= 280
assert result.payload["related"] == []
assert result.payload["returned"] == 0
assert result.payload["remaining"] == 1
assert result.payload["truncated"] is True
assert result.payload["next_cursor"] is None
# ---------------------------------------------------------------------------
# No credential/exception leakage
# ---------------------------------------------------------------------------
def test_unexpected_transport_error_does_not_leak_credential_or_raw_exception(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
leaking_exception = requests.exceptions.ConnectionError(
f"connect failed for token={FAKE_TOKEN} at internal-host:5432"
)
monkeypatch.setattr(requests, "get", _FakeTransport([leaking_exception]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
assert result.ok is False
assert result.payload["error"]["code"] == "UPSTREAM_ERROR"
payload_text = str(result.payload)
assert FAKE_TOKEN not in payload_text
assert "internal-host" not in payload_text
def test_not_found_message_does_not_distinguish_missing_from_inaccessible(
identity: IdentityContext, monkeypatch: pytest.MonkeyPatch,
) -> None:
"""Security requirement: a denial/miss must not reveal whether the
underlying resource exists — the safe_message must stay generic."""
monkeypatch.setattr(requests, "get", _FakeTransport([_FakeResponse(404)]))
provider = GiteaIssueProvider(_target())
app, _, _ = _runtime(identity, provider)
result = dispatch("get_project_issue_context", {"project_id": "cowork-local", "issue_key": "1"}, app)
message = result.payload["error"]["message"].lower()
assert "not found or is not accessible" in message
assert "does not exist" not in message