fix(pr-review): fleet PRs are ours by branch ownership, not author identity (#668)

With a GitHub App bound, fleet PRs are authored by <app-slug>[bot] whose
author_association is NONE — the inbound classifier's author heuristics
read that as an outsider and ingested the org's own dev-stream PR as
external_pr for adversarial review (2026-07-23 live: PR #667). The
repo-owner author check only ever covered the PAT era.

_ingest_pr_if_reviewable now skips any same-repo PR whose head branch an
active task owns BEFORE the author-based classification, and
active_task_owns_branch widens from the single polled project to every
project sharing its git_url (the poll collapses a monorepo's
cell-projects to one canonical project, so a sibling cell's ownership
must count — the same sibling scope external_review_task_exists already
uses, now shared via _repo_sibling_project_ids). A deleted-fork head
(GitHub sends head.repo=null) now classifies as fork, failing closed to
review instead of risking a silent ownership skip on a branch-name
collision. Residual, documented: an org PR whose task went terminal with
the PR left open falls through to the author heuristics.

Co-authored-by: Renn F <rennf93@users.noreply.github.com>
This commit is contained in:
Renzo F
2026-07-23 21:54:45 +02:00
committed by GitHub
co-authored by Renn F
parent a036c97985
commit f8b4a6755c
6 changed files with 187 additions and 53 deletions
+19 -8
View File
@@ -9508,9 +9508,11 @@ Start by:
Repo-aware: collapses active projects to one canonical project per Repo-aware: collapses active projects to one canonical project per
distinct repo (so a monorepo product yields ONE review per PR, not one distinct repo (so a monorepo product yields ONE review per PR, not one
per cell-project), lists each repo's open PRs, and ingests a de-duped per cell-project), lists each repo's open PRs, and ingests a de-duped
review task for each reviewable one external/fork PRs, and (when review task for each reviewable one. A same-repo PR whose head branch
internal review is on) org-repo PRs not tied to an active task. Commits an active task owns (repo-wide, sibling cell-projects included) is the
once at the end. org's own and never ingested, regardless of author — the rest split
into external/fork PRs and (when internal review is on) org-repo PRs
opened outside the task flow. Commits once at the end.
""" """
from roboco.services.git import GitService from roboco.services.git import GitService
from roboco.services.project import get_project_service from roboco.services.project import get_project_service
@@ -9554,6 +9556,20 @@ Start by:
# 422), and re-reviewing the org's own in-flight PRs every poll is noise. # 422), and re-reviewing the org's own in-flight PRs every poll is noise.
if pr.get("author_is_owner"): if pr.get("author_is_owner"):
return False return False
# The org's own in-flight PRs are recognized by BRANCH OWNERSHIP, not
# author identity: with a GitHub App bound, fleet PRs are authored by
# <app-slug>[bot] whose author_association is NONE, which the external
# heuristic below reads as an outsider (2026-07-23 live incident: a
# same-repo dev-stream PR was ingested as external_pr and adversarially
# reviewed). A same-repo head branch owned by an active task is ours
# regardless of who authored the PR, so this check must run BEFORE the
# author-based classification. Residual: an org PR whose task went
# terminal with the PR left open falls through to the author heuristics.
if not pr.get("is_fork") and await task_service.active_task_owns_branch(
str(pr.get("head_ref") or ""),
cast("UUID", project.id),
):
return False
if self._is_external_pr(pr): if self._is_external_pr(pr):
if not settings.external_pr_enabled or not self._pr_author_allowed( if not settings.external_pr_enabled or not self._pr_author_allowed(
pr, allowlist pr, allowlist
@@ -9563,11 +9579,6 @@ Start by:
else: else:
if not settings.internal_pr_enabled: if not settings.internal_pr_enabled:
return False return False
if await task_service.active_task_owns_branch(
str(pr.get("head_ref") or ""),
cast("UUID", project.id),
):
return False
source = "internal_pr" source = "internal_pr"
created = await task_service.ingest_external_pr( created = await task_service.ingest_external_pr(
project_id=cast("UUID", project.id), project_id=cast("UUID", project.id),
+5 -1
View File
@@ -2316,7 +2316,11 @@ class GitService(BaseService):
"title": pr.get("title") or "", "title": pr.get("title") or "",
"head_ref": head.get("ref"), "head_ref": head.get("ref"),
"head_sha": head.get("sha"), "head_sha": head.get("sha"),
"is_fork": bool(head_full and head_full != base_full), # A null head repo (GitHub sends head.repo=null when the fork was
# deleted) is NOT ours — fail closed to fork/external so the
# branch-ownership skip can never silently swallow a genuine fork
# whose head_ref collides with an org branch name.
"is_fork": (head_full != base_full) if head_full else True,
"user_login": login, "user_login": login,
# The reviewer reviews PRs the org did NOT author. A PR opened by the # The reviewer reviews PRs the org did NOT author. A PR opened by the
# repo-owner account is a self-review (GitHub 422s REQUEST_CHANGES on # repo-owner account is a self-review (GitHub 422s REQUEST_CHANGES on
+39 -30
View File
@@ -1407,27 +1407,7 @@ class TaskService(BaseService):
- a legacy/markerless task exists, or ``head_sha`` is unknown -> True; - a legacy/markerless task exists, or ``head_sha`` is unknown -> True;
- tasks exist but all cover OTHER SHAs -> False (new commits re-review). - tasks exist but all cover OTHER SHAs -> False (new commits re-review).
""" """
# Resolve every project sharing this project's repo (git_url) so the scope_ids = await self._repo_sibling_project_ids(project_id)
# dedupe spans the whole monorepo, not just the one project. An unknown
# project_id falls back to itself (can't widen).
git_url = (
await self.session.execute(
select(ProjectTable.git_url).where(ProjectTable.id == project_id)
)
).scalar_one_or_none()
if git_url:
sibling_ids = (
(
await self.session.execute(
select(ProjectTable.id).where(ProjectTable.git_url == git_url)
)
)
.scalars()
.all()
)
scope_ids = list(sibling_ids) or [project_id]
else:
scope_ids = [project_id]
result = await self.session.execute( result = await self.session.execute(
select(TaskTable.orchestration_markers).where( select(TaskTable.orchestration_markers).where(
TaskTable.project_id.in_(scope_ids), TaskTable.project_id.in_(scope_ids),
@@ -1514,26 +1494,55 @@ class TaskService(BaseService):
await self.session.flush() await self.session.flush()
return task return task
async def active_task_owns_branch(self, branch_name: str, project_id: UUID) -> bool: async def _repo_sibling_project_ids(self, project_id: UUID) -> list[UUID]:
"""True if a non-terminal task on ``project_id`` already owns this branch. """Every project id sharing ``project_id``'s repo (``git_url``).
Lets the internal-PR reviewer skip the org's own in-flight integration The inbound-PR poll collapses a monorepo's cell-projects to one
canonical project per repo, so any repo-level lookup keyed by the
canonical id must widen back out to the siblings a branch/review
owned by a sibling cell-project is still "this repo's". Falls back to
``[project_id]`` when the project is unknown or has no git_url.
"""
git_url = (
await self.session.execute(
select(ProjectTable.git_url).where(ProjectTable.id == project_id)
)
).scalar_one_or_none()
if not git_url:
return [project_id]
sibling_ids = (
(
await self.session.execute(
select(ProjectTable.id).where(ProjectTable.git_url == git_url)
)
)
.scalars()
.all()
)
return list(sibling_ids) or [project_id]
async def active_task_owns_branch(self, branch_name: str, project_id: UUID) -> bool:
"""True if a non-terminal task on this REPO already owns this branch.
Lets the inbound-PR reviewer skip the org's own in-flight integration
PRs those whose head branch a live task created via the agent PRs those whose head branch a live task created via the agent
task-flow (and which therefore already pass QA + PM review) and review task-flow (and which therefore already pass QA + PM review) and review
only org-repo PRs opened outside that flow. only org-repo PRs opened outside that flow.
Scoped to the polled project: a branch in project A's repo can only be Scoped to the REPO (every project sharing ``project_id``'s git_url),
owned by a task whose ``project_id == A`` (each task branches in its not the single polled project: the poll collapses a monorepo's
own project's repo, including each root-subtask of a multi-repo cell-projects to one canonical project, so a fleet PR whose owning
MegaTask). An unscoped lookup would match the wrong project's task on a task lives on a sibling cell-project must still count as ours.
cross-project branch_name collision and false-skip project A's PR. Cross-REPO branch-name collisions stay excluded a different repo's
task can never own this repo's branch.
""" """
if not branch_name: if not branch_name:
return False return False
scope_ids = await self._repo_sibling_project_ids(project_id)
result = await self.session.execute( result = await self.session.execute(
select(TaskTable.id).where( select(TaskTable.id).where(
TaskTable.branch_name == branch_name, TaskTable.branch_name == branch_name,
TaskTable.project_id == project_id, TaskTable.project_id.in_(scope_ids),
TaskTable.status.notin_([TaskStatus.COMPLETED, TaskStatus.CANCELLED]), TaskTable.status.notin_([TaskStatus.COMPLETED, TaskStatus.CANCELLED]),
) )
) )
@@ -1,17 +1,16 @@
"""active_task_owns_branch must be scoped to the polled project. """active_task_owns_branch must be scoped to the polled project's REPO.
The internal-PR reviewer (orchestrator) calls this to skip the org's own The inbound-PR reviewer (orchestrator) calls this to skip the org's own
in-flight integration PRs a PR on project A's repo is "ours" only if a in-flight integration PRs. Scope is the repo (every project sharing the
non-terminal task ON PROJECT A owns its head branch. The query was unscoped polled project's ``git_url``), not one project: the poll collapses a
(``WHERE branch_name = ?``), so a cross-project branch_name collision (two monorepo's cell-projects to one canonical project, so a fleet PR whose
tasks sharing an 8-char-UUID-prefix branch on different projects) made it owning task lives on a sibling cell-project must still count as ours
match the WRONG project's task — project A's leftover PR was skipped because (2026-07-23: a GitHub-App-authored dev-stream PR was ingested as external_pr
project B happened to have an active task with the same branch_name. because the canonical project's id didn't match the owning cell's).
Scoping by ``project_id`` is correct for both single-project tasks and Cross-REPO branch_name collisions stay excluded the original bug this file
MegaTask multi-repo batches: each root-subtask carries its own ``project_id`` pinned: an unscoped query (``WHERE branch_name = ?``) matched the WRONG
matching its own repo, so a branch on project A's repo is owned only by a repo's task on an 8-char-UUID-prefix collision and false-skipped a real PR.
task whose ``project_id == A``.
""" """
from __future__ import annotations from __future__ import annotations
@@ -40,7 +39,9 @@ SYSTEM_UUID = _foundation.AGENTS["system"].uuid
_BRANCH = "feature/backend/collide001" _BRANCH = "feature/backend/collide001"
async def _seed_project(db: AsyncSession, slug: str) -> ProjectTable: async def _seed_project(
db: AsyncSession, slug: str, git_url: str | None = None
) -> ProjectTable:
if await db.get(AgentTable, SYSTEM_UUID) is None: if await db.get(AgentTable, SYSTEM_UUID) is None:
db.add( db.add(
AgentTable( AgentTable(
@@ -62,7 +63,7 @@ async def _seed_project(db: AsyncSession, slug: str) -> ProjectTable:
id=uuid4(), id=uuid4(),
name=slug, name=slug,
slug=slug, slug=slug,
git_url=f"https://github.com/rennf93/{slug}", git_url=git_url or f"https://github.com/rennf93/{slug}",
assigned_cell=Team.BACKEND, assigned_cell=Team.BACKEND,
created_by=SYSTEM_UUID, created_by=SYSTEM_UUID,
) )
@@ -118,6 +119,27 @@ async def test_branch_owned_only_by_its_own_project(db_session: AsyncSession) ->
assert await svc.active_task_owns_branch(_BRANCH, cast("UUID", proj_b.id)) is False assert await svc.active_task_owns_branch(_BRANCH, cast("UUID", proj_b.id)) is False
@pytest.mark.asyncio
async def test_monorepo_sibling_project_ownership_counts(
db_session: AsyncSession,
) -> None:
"""Two cell-projects share one git_url (monorepo). The active task lives
on the FRONTEND cell; the poll queries with the CANONICAL sibling's id.
Repo-scoping must still recognize the branch as ours the 2026-07-23
incident shape."""
repo = "https://github.com/rennf93/gca-mono"
canonical = await _seed_project(db_session, "gca-mono-api", git_url=repo)
frontend = await _seed_project(db_session, "gca-mono-panel", git_url=repo)
branch = "feature/frontend/mono0001--child001"
db_session.add(
_task(cast("UUID", frontend.id), branch=branch, status=TaskStatus.IN_PROGRESS)
)
await db_session.flush()
svc = get_task_service(db_session)
assert await svc.active_task_owns_branch(branch, cast("UUID", canonical.id)) is True
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_empty_branch_never_owned(db_session: AsyncSession) -> None: async def test_empty_branch_never_owned(db_session: AsyncSession) -> None:
proj = await _seed_project(db_session, "gca-collide-empty") proj = await _seed_project(db_session, "gca-collide-empty")
@@ -198,6 +198,78 @@ async def test_skip_internal_when_disabled(monkeypatch: pytest.MonkeyPatch) -> N
svc.ingest_external_pr.assert_not_awaited() svc.ingest_external_pr.assert_not_awaited()
@pytest.mark.asyncio
async def test_skip_app_bot_authored_fleet_pr(monkeypatch: pytest.MonkeyPatch) -> None:
"""The 2026-07-23 live incident: with a GitHub App bound, fleet PRs are
authored by <app-slug>[bot] whose author_association is NONE the external
heuristic reads that as an outsider. Branch ownership must win: a same-repo
PR whose head an active task owns is the org's own, whoever authored it."""
monkeypatch.setattr(orch_mod.settings, "external_pr_enabled", True)
monkeypatch.setattr(orch_mod.settings, "internal_pr_enabled", True)
svc = _svc(owns_branch=True)
pr = {
"number": 667,
"is_fork": False,
"author_is_owner": False,
"user_login": "roboco-app[bot]",
"author_association": "NONE",
"head_ref": "feature/frontend/170c9578--f1957610",
}
ok = await _orch()._ingest_pr_if_reviewable(
svc, SimpleNamespace(id=uuid4()), pr, uuid4(), set()
)
assert ok is False
svc.ingest_external_pr.assert_not_awaited()
@pytest.mark.asyncio
async def test_app_bot_pr_without_owning_task_still_reviews(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""The documented residual: a same-repo bot-authored PR with NO active
owning task falls through to the author heuristics and reviews as
external orphaned/unknown bot branches (dependabot included) keep
getting the read-only adversarial review."""
monkeypatch.setattr(orch_mod.settings, "external_pr_enabled", True)
monkeypatch.setattr(orch_mod.settings, "internal_pr_enabled", True)
svc = _svc(owns_branch=False)
pr = {
"number": 42,
"is_fork": False,
"author_is_owner": False,
"user_login": "dependabot[bot]",
"author_association": "NONE",
"head_ref": "dependabot/npm_and_yarn/foo-1.2.3",
}
ok = await _orch()._ingest_pr_if_reviewable(
svc, SimpleNamespace(id=uuid4()), pr, uuid4(), set()
)
assert ok is True
assert svc.ingest_external_pr.await_args.kwargs["source"] == "external_pr"
@pytest.mark.asyncio
async def test_fork_pr_never_consults_branch_ownership(
monkeypatch: pytest.MonkeyPatch,
) -> None:
"""A fork PR is external by definition — the ownership pre-check must not
run for it (a fork head ref can coincide with an org branch name)."""
monkeypatch.setattr(orch_mod.settings, "external_pr_enabled", True)
svc = _svc(owns_branch=True)
pr = {
"number": 7,
"is_fork": True,
"user_login": "outsider",
"head_ref": "feature/frontend/copycat",
}
ok = await _orch()._ingest_pr_if_reviewable(
svc, SimpleNamespace(id=uuid4()), pr, uuid4(), set()
)
assert ok is True
svc.active_task_owns_branch.assert_not_awaited()
assert svc.ingest_external_pr.await_args.kwargs["source"] == "external_pr"
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_skip_owner_authored_pr(monkeypatch: pytest.MonkeyPatch) -> None: async def test_skip_owner_authored_pr(monkeypatch: pytest.MonkeyPatch) -> None:
# The org's own account opened the PR → self-review, never ingest (even with # The org's own account opened the PR → self-review, never ingest (even with
@@ -104,6 +104,22 @@ async def test_list_open_prs_normalizes_and_flags_fork() -> None:
assert internal["author_is_owner"] is False assert internal["author_is_owner"] is False
def test_normalize_deleted_fork_head_is_fork() -> None:
"""GitHub sends head.repo=null when a fork was deleted — that head is NOT
ours, so it must classify as a fork (fail-closed to review) rather than
fall into the same-repo path where a branch-name collision could skip it."""
pr = {
"number": 11,
"html_url": "https://github.com/acme/repo/pull/11",
"title": "ghost fork",
"head": {"ref": "feature-x", "sha": "cafebabe", "repo": None},
"user": {"login": "ghost"},
"author_association": "NONE",
}
out = GitService._normalize_open_pr(pr, "acme/repo")
assert out["is_fork"] is True
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_list_open_prs_flags_owner_authored_pr() -> None: async def test_list_open_prs_flags_owner_authored_pr() -> None:
"""A PR opened by the repo-owner account is flagged author_is_owner.""" """A PR opened by the repo-owner account is flagged author_is_owner."""