[F061] audit status-transition rows now written in-session (F061/F073/F075)

_emit_status_transition_audit now writes AuditLogTable rows into
self.session synchronously (session.add) instead of dispatching
AuditService.log_task_event fire-and-forget on its own connection.

The audit row now commits/rolls back atomically with the status
transition in the caller's transaction, closing three facets at once:
- F061: audit commit no longer decoupled from the transition commit
- F073: a committed transition can no longer have NO audit row
  (the row rides the same transaction; a swallowed persist can't drop it)
- F075: a transition rolled back inside a verb savepoint no longer
  leaves a phantom audit row (the row is in the savepoint too)

log_task_event is now called only from this helper (narrow blast
radius verified); revision_count increment stays at this single
chokepoint. Cycle-time/bottleneck reconstruction from task.<status>
events is no longer silently corruptible.

Tests: test_emit_status_transition_audit_writes_in_session_atomically,
test_finalize_claim_rollback_emits_reversal_audit, escalation-audit
tests retargeted to in-session AuditLogTable rows.

Also: _canonical_bump_files grep-looseness follow-on (F058) -- filter
by subject, not body; git log --grep matches any message line, so a
non-release commit whose body references chore(release): shadowed the
real release commit. Test
test_canonical_bump_files_ignores_body_only_chore_release_match.

roboco/services/release_readiness.py

@@ -416,12 +416,28 @@ def _canonical_bump_files(root: Path, version: str) -> list[str]:
     # Subsequent releases derive the canonical bump set from the previous
     # ``chore(release):`` commit's touched files — the historical record of
     # what a release bumps.
-    sha = _run_git(
-        root, ["log", "--grep", "^chore(release):", "-n1", "--format=%H"]
-    ).strip()
-    if sha:
-        raw = _run_git(root, ["show", "--name-only", "--format=", sha])
-        return sorted(line.strip() for line in raw.splitlines() if line.strip())
+    #
+    # ``git log --grep "^chore(release):"`` matches ANY message line, including
+    # a body line of a non-release commit that merely REFERENCES the
+    # ``chore(release):`` type (e.g. a fix commit explaining the derivation). A
+    # newer such commit would shadow the real release commit and the bump plan
+    # would list the fix's files instead of the release's. Filter to the
+    # candidate whose SUBJECT (``%s``) starts with ``chore(release):`` — the
+    # real release-commit shape ``chore(release): X.Y.Z``. ``%x01`` separates
+    # sha from subject so a subject may contain spaces unambiguously.
+    raw = _run_git(
+        root,
+        ["log", "--grep", "^chore(release):", "--format=%H%x01%s"],
+    )
+    for line in raw.splitlines():
+        if "\x01" not in line:
+            continue
+        sha, subject = line.split("\x01", 1)
+        if subject.startswith("chore(release):"):
+            files_raw = _run_git(root, ["show", "--name-only", "--format=", sha])
+            return sorted(
+                line.strip() for line in files_raw.splitlines() if line.strip()
+            )
     # F058: the FIRST release has no prior ``chore(release):`` commit, so the
     # historical derivation returns ``[]`` and the executor would publish a tag
     # with no files bumped (a no-op release masquerading as X.Y.Z). Fall back to

roboco/services/task.py

@@ -662,25 +662,40 @@ class TaskService(BaseService):
         task without routing through the strict transition validator — record
         the same audit event. No status change may bypass the audit log.
 
-        Fire-and-forget, but we hold a strong reference to the background task
-        (via ``_background_tasks``): the event loop only weak-refs tasks, so
-        without it the audit write can be garbage-collected before it commits.
+        The row is written into the CALLER's session (``self.session.add``),
+        so it commits / rolls back ATOMICALLY with the status transition — the
+        audit journey can never diverge from real task state. This closes three
+        facets of the fire-and-forget decoupling gap:
+
+        * F061/F073 — the audit commit is no longer decoupled on a separate
+          connection whose failures were swallowed; a committed transition now
+          ALWAYS has its audit row (same transaction), and a persist failure
+          fails the transition (fail-closed for the metric source of truth)
+          instead of silently dropping the row.
+        * F075 — a transition rolled back inside a verb savepoint
+          (``_verb_runner`` wraps composed actions in ``begin_nested``) now
+          rolls its audit row back too — no phantom ``task.<status>`` row for a
+          transition that did not stick. (The claim-branch-failure rollback in
+          ``_finalize_claim`` is one such site; the savepoint rollback handles
+          the general case.)
+
+        The ``revision_count`` rework counter is incremented synchronously
+        here (the single chokepoint every transition funnels through).
 
         The explicit ``audit_agent_id`` (capture-before-mutate) wins: callers
         like ``submit_for_qa`` clear ``task.claimed_by`` before transitioning
         but still want the row attributed to the outgoing agent. Otherwise fall
-        back to ``task.claimed_by``.
+        back to ``task.claimed_by``. A structurally-invalid id coerces to
+        ``None`` (mirroring ``AuditService._coerce_uuid``) so the row still
+        lands unattributed rather than FK-violating; a valid-but-deleted agent
+        id is a real bug worth surfacing as a transition failure.
         """
-        import asyncio
-        import contextlib
-
-        from roboco.services.audit import get_audit_service
+        from roboco.db.tables import AuditLogTable
 
         # Rework counter: a bounce INTO needs_revision (not a re-entry) is one
         # rework cycle. Incremented at this single chokepoint — every transition
         # path funnels its audit through here exactly once — so the rework rate
-        # is an O(1) column read. Synchronous (part of this unit of work),
-        # unlike the fire-and-forget audit rows below.
+        # is an O(1) column read. Synchronous (part of this unit of work).
         if (
             to_status == TaskStatus.NEEDS_REVISION.value
             and from_status != TaskStatus.NEEDS_REVISION.value
@@ -694,6 +709,13 @@ class TaskService(BaseService):
         else:
             resolved_audit_agent_id = None
 
+        agent_uuid: UUID | None = None
+        if resolved_audit_agent_id:
+            try:
+                agent_uuid = UUID(resolved_audit_agent_id)
+            except (ValueError, AttributeError):
+                agent_uuid = None
+
         details = {
             "from_status": from_status,
             "to_status": to_status,
@@ -702,20 +724,17 @@ class TaskService(BaseService):
                 task.team.value if hasattr(task.team, "value") else str(task.team)
             ),
         }
-        audit = get_audit_service()
-        with contextlib.suppress(RuntimeError):
-            loop = asyncio.get_running_loop()
-            for event_type in self._audit_events_for(to_status, agent_role):
-                bg = loop.create_task(
-                    audit.log_task_event(
-                        event_type=event_type,
-                        task_id=str(task.id),
-                        agent_id=resolved_audit_agent_id,
-                        details=details,
-                    )
+        for event_type in self._audit_events_for(to_status, agent_role):
+            self.session.add(
+                AuditLogTable(
+                    event_type=event_type,
+                    agent_id=agent_uuid,
+                    target_type="task",
+                    target_id=task.id,
+                    severity="info",
+                    details=dict(details),
                 )
-                self._background_tasks.add(bg)
-                bg.add_done_callback(self._background_tasks.discard)
+            )
 
     @staticmethod
     def _audit_events_for(to_status: str, agent_role: str | None) -> list[str]:

tests/unit/services/test_escalation_board_guard.py

@@ -12,11 +12,11 @@ stranded on a board role.
 
 from __future__ import annotations
 
-import asyncio
-from unittest.mock import AsyncMock, MagicMock, patch
+from unittest.mock import AsyncMock, MagicMock
 from uuid import uuid4
 
 import pytest
+from roboco.db.tables import AuditLogTable
 from roboco.models.base import AgentRole, TaskStatus, TaskType, Team
 from roboco.services.task import (
     TaskService,
@@ -528,8 +528,14 @@ async def test_apply_escalation_blocks_non_terminal_task() -> None:
 async def test_apply_escalation_emits_blocked_audit_event() -> None:
     """A non-divert escalation sets BLOCKED and MUST record a task.blocked audit
     row. The escalate path sets status directly (bypassing the validated
-    transition), and used to skip the audit log entirely."""
+    transition), and used to skip the audit log entirely.
+
+    The audit row is written into the caller's session (F061/F073/F075: it
+    commits atomically with the transition, not fire-and-forget on a separate
+    connection), so we assert on the ``AuditLogTable`` added to the session."""
     svc = _service()
+    added: list[object] = []
+    svc.session.add = MagicMock(side_effect=added.append)  # type: ignore[assignment]
     task = MagicMock(
         id=uuid4(),
         parent_task_id=uuid4(),
@@ -542,56 +548,52 @@ async def test_apply_escalation_emits_blocked_audit_event() -> None:
         status=TaskStatus.IN_PROGRESS,
     )
     _bind(svc, "_is_board_advisory_agent", AsyncMock(return_value=False))
-    audit_mock = MagicMock(log_task_event=AsyncMock())
 
-    with patch("roboco.services.audit.get_audit_service", return_value=audit_mock):
-        await svc.apply_escalation(
-            task=task,
-            target_agent_id=uuid4(),
-            escalator_slug="be-pm",
-            target_slug="main-pm",
-            reason="needs a decision",
-        )
-        # Drain the fire-and-forget audit task so the assertion sees the call.
-        pending = list(svc._background_tasks)
-        if pending:
-            await asyncio.gather(*pending, return_exceptions=True)
+    await svc.apply_escalation(
+        task=task,
+        target_agent_id=uuid4(),
+        escalator_slug="be-pm",
+        target_slug="main-pm",
+        reason="needs a decision",
+    )
 
     assert task.status == TaskStatus.BLOCKED
-    audit_mock.log_task_event.assert_awaited_once()
-    kwargs = audit_mock.log_task_event.await_args.kwargs
-    assert kwargs["event_type"] == "task.blocked"
-    assert kwargs["details"]["from_status"] == "in_progress"
-    assert kwargs["details"]["to_status"] == "blocked"
+    rows = [r for r in added if isinstance(r, AuditLogTable)]
+    assert any(r.event_type == "task.blocked" for r in rows)
+    blocked = next(r for r in rows if r.event_type == "task.blocked")
+    assert blocked.details["from_status"] == "in_progress"
+    assert blocked.details["to_status"] == "blocked"
 
 
 @pytest.mark.asyncio
 async def test_unblock_with_restore_emits_audit_event() -> None:
     """The PM restore path sets status directly (bypassing the validated
-    transition) and used to skip the audit log; it must record the transition."""
+    transition) and used to skip the audit log; it must record the transition.
+
+    The audit row is written into the caller's session (F061/F073/F075: atomic
+    with the transition, not fire-and-forget), so we assert on the
+    ``AuditLogTable`` added to the session."""
     svc = _service()
+    added: list[object] = []
+    svc.session.add = MagicMock(side_effect=added.append)  # type: ignore[assignment]
     task = MagicMock(
         id=uuid4(),
         status=TaskStatus.BLOCKED,
         pre_block_state="in_progress",
         pre_block_assignee=None,
         claimed_by=uuid4(),
+        team=Team.BACKEND,
     )
     _bind(svc, "get", AsyncMock(return_value=task))
-    audit_mock = MagicMock(log_task_event=AsyncMock())
 
-    with patch("roboco.services.audit.get_audit_service", return_value=audit_mock):
-        await svc.unblock_with_restore(uuid4(), uuid4(), restore=True)
-        pending = list(svc._background_tasks)
-        if pending:
-            await asyncio.gather(*pending, return_exceptions=True)
+    await svc.unblock_with_restore(uuid4(), uuid4(), restore=True)
 
     assert task.status == TaskStatus.IN_PROGRESS
-    audit_mock.log_task_event.assert_awaited_once()
-    kwargs = audit_mock.log_task_event.await_args.kwargs
-    assert kwargs["event_type"] == "task.in_progress"
-    assert kwargs["details"]["from_status"] == "blocked"
-    assert kwargs["details"]["to_status"] == "in_progress"
+    rows = [r for r in added if isinstance(r, AuditLogTable)]
+    assert any(r.event_type == "task.in_progress" for r in rows)
+    restored = next(r for r in rows if r.event_type == "task.in_progress")
+    assert restored.details["from_status"] == "blocked"
+    assert restored.details["to_status"] == "in_progress"
 
 
 # ---------------------------------------------------------------------------

tests/unit/services/test_release_readiness_first_release.py

@@ -130,5 +130,48 @@ def test_first_release_emits_no_version_ref_gap_for_planned_files(
     assert not any(g.category == "version_ref" for g in report.gaps)
 
 
+def test_canonical_bump_files_ignores_body_only_chore_release_match(
+    tmp_path: Path,
+) -> None:
+    """A non-release commit whose message BODY references ``chore(release):``
+    (a body line starting with ``chore(release):``) must NOT be misidentified
+    as the last release commit.
+
+    ``git log --grep "^chore(release):"`` matches ANY message line, so a
+    fix/docs commit that explains the release process (body line
+    ``chore(release): the canonical set ...``) is matched and — being newer —
+    shadows the real release commit. The canonical set must come from the
+    commit whose SUBJECT starts with ``chore(release):`` (the real release
+    shape ``chore(release): X.Y.Z``), not a body-only match. (Regression: an
+    earlier fix commit's body referencing ``chore(release):`` was picked up,
+    so the bump plan listed that fix's files instead of the release's.)
+    """
+    root = _first_release_repo(tmp_path)
+    # Real release commit (older) touching the version-embedding file + marker.
+    (root / "RELEASE_MARKER.txt").write_text("released\n", encoding="utf-8")
+    (root / "pyproject.toml").write_text('version = "0.2.0"\n', encoding="utf-8")
+    _git(root, "add", "-A")
+    _git(root, "commit", "-m", "chore(release): 0.2.0")
+    # A NEWER non-release commit whose BODY has a line starting with the
+    # release-commit prefix.
+    (root / "other.py").write_text("x = 1\n", encoding="utf-8")
+    _git(root, "add", "-A")
+    _git(
+        root,
+        "commit",
+        "-m",
+        "fix: tighten release-readiness derivation",
+        "-m",
+        "chore(release): the canonical set derives from this commit type",
+    )
+
+    files = _canonical_bump_files(root, "0.2.0")
+    # The canonical set is the REAL release commit's files...
+    assert "RELEASE_MARKER.txt" in files
+    assert "pyproject.toml" in files
+    # ...NOT the body-only-matching fix commit's file.
+    assert "other.py" not in files
+
+
 if __name__ == "__main__":
     pytest.main([__file__, "-q"])

tests/unit/services/test_task.py

@@ -14,6 +14,7 @@ from unittest.mock import AsyncMock, MagicMock, patch
 from uuid import uuid4
 
 import pytest
+from roboco.db.tables import AuditLogTable
 from roboco.models.base import (
     AgentRole,
     AgentStatus,
@@ -1117,6 +1118,55 @@ async def test_finalize_claim_rollback_emits_reversal_audit() -> None:
     assert {"from": "claimed", "to": "pending"} in audit_calls
 
 
+@pytest.mark.asyncio
+async def test_emit_status_transition_audit_writes_in_session_atomically() -> None:
+    """F061/F073/F075: the status-transition audit row is written into the
+    CALLER's session (same transaction as the transition), not fire-and-forget
+    on a separate connection.
+
+    Fire-and-forget decouples the audit commit from the transition commit:
+    a transition that rolls back inside a verb savepoint leaves a PHANTOM audit
+    row (F075), and a swallowed persist failure means a committed transition
+    can have NO audit row (F073) — silently corrupting the cycle-time /
+    bottleneck metrics reconstructed from ``task.<status>`` events (F061).
+    Writing the row in-session makes it commit/roll back atomically with the
+    transition, closing all three. Asserted at the unit level: the row is
+    ``session.add``-ed (same txn) with the metric-reconstruction details, and
+    NO fire-and-forget background task is spawned.
+    """
+    svc = TaskService(MagicMock())
+    added: list[object] = []
+    svc.session.add = MagicMock(side_effect=added.append)  # type: ignore[assignment]
+    prior_bg = set(svc._background_tasks)
+
+    task = MagicMock(id=uuid4(), claimed_by=uuid4(), team=Team.BACKEND)
+
+    svc._emit_status_transition_audit(
+        task,
+        from_status="pending",
+        to_status="claimed",
+        agent_role="developer",
+        audit_agent_id=None,
+    )
+
+    # The audit row is added to the CALLER's session (same transaction) — not
+    # dispatched to a separate fire-and-forget connection.
+    rows = [r for r in added if isinstance(r, AuditLogTable)]
+    assert len(rows) == 1
+    row = rows[0]
+    assert row.event_type == "task.claimed"
+    assert row.target_type == "task"
+    assert row.target_id == task.id
+    assert row.details["from_status"] == "pending"
+    assert row.details["to_status"] == "claimed"
+    assert row.details["agent_role"] == "developer"
+    assert row.details["team"] == "backend"
+    # The claiming agent is attributed (resolved from claimed_by).
+    assert row.agent_id == task.claimed_by
+    # No fire-and-forget audit task was spawned (the old decoupled path).
+    assert svc._background_tasks == prior_bg
+
+
 # ---------------------------------------------------------------------------
 # _resolve_doc_abspath — normalize documenter-supplied paths under /app/docs
 # ---------------------------------------------------------------------------