[a1f52d13] Rebuild evidence-assembly timeout fix on current main (#796)

* [a1f52d13] feat(gateway): re-apply evidence-assembly timeout fix on current main Rebuilds the bounded-timeout fix for claim_review/evidence()/roboco_git_diff that PR #765 carried as contaminated pre-Board-Programs file copies. The rebase restored the current-main baseline; this commit re-applies ONLY the validated timeout-fix hunks on top of that baseline: - config.py: evidence_assembly_timeout_seconds (90.0s) and conventions_validator_timeout_seconds (45s) Settings fields - envelope.py: Envelope.gateway_timeout classmethod for structured timeout errors naming the stalled component - git.py: GitService.diff_and_files (single resolution, concurrent diff+--name-only subprocesses), settings-backed conventions-validator timeout, and changed_files passthrough to conventions_check_for_task - qa.py: claim_review bounded-timeout wrapper around _build_qa_claim_evidence, split into _qa_git_and_conventions and _qa_db_reads concurrent segments (pass_review signature and gate untouched) - content_actions.py: evidence() bounded wrapper with _evidence_git_and_fetch and _evidence_db_reads concurrent split (all other methods untouched) - git routes: GET /diff bounded asyncio.wait_for guard with 504 on trip The pre-existing methods (sync_env_branch, open_sync_pr, pass_review with criteria_verified gate, all propose_*/request_render methods) are preserved and untouched. The test files already exist on current main. * [a1f52d13] fix(git-routes): wrap workspace resolution in bounded timeout for GET /diff The bounded asyncio.wait_for must cover the workspace resolution step too, not just the diff+stat subprocesses — the integration test makes get_workspace slow (not _run_git), so the timeout guard needs to enclose the entire resolve-and-diff path to return a structured 504 on trip. * [a1f52d13] test: update mock setups from git.diff/list_changed_files to diff_and_files Tests that exercise claim_review and evidence() paths need their git mocks to return a tuple from diff_and_files instead of separate diff and list_changed_files return values, matching the new combined accessor introduced by the evidence-assembly timeout fix. * [a1f52d13] test: configure diff_and_files mock in lifecycle parity test _make_deps The spec-parity test test_claim_review_matches_spec[qa-awaiting_qa] failed with ValueError at qa.py:330 because _make_deps set git=AsyncMock() without configuring diff_and_files.return_value, so the await resolved to a MagicMock instead of a (diff_summary, files_changed) 2-tuple. Adds the same return_value = ("", []) configuration that commit 014a0d03 applied to the other 4 test files. * [a1f52d13] docs(evidence-assembly-timeout): correct GET /diff guard scope and _qa_git_and_conventions timing field --------- Co-authored-by: Backend Developer 2 <be-dev-2@roboco.tech> Co-authored-by: Backend Documenter <be-doc@roboco.tech>
2026-08-03 07:23:24 +02:00 · 2026-08-02 03:22:25 +00:00
parent 89254f796c
commit ce89a04d26
12 changed files with 6314 additions and 974 deletions
@@ -22,7 +22,7 @@ A new `settings.evidence_assembly_timeout_seconds` (default 90s, well under the

 - `ContentActions.evidence()` — wraps the `asyncio.gather` of the git-diff coroutine and `_evidence_db_reads`.
 - `QAMixin.claim_review()` — wraps the call to `_build_qa_claim_evidence`; the claim itself (`qa_claim` + `mark_evidence_inspected`) has already committed by this point, so a trip here means only evidence assembly is slow, not that the claim failed.
- `GET /api/git/diff` (`roboco/api/routes/git.py`, `roboco_git_diff`) — wraps the diff + `--stat` subprocess pair.
+- `GET /api/git/diff` (`roboco/api/routes/git.py`, `roboco_git_diff`) — wraps workspace resolution *plus* the diff + `--stat` subprocess pair in one bounded guard, so a slow workspace fetch trips the 504 too — not just a slow diff. (The workspace resolution is enclosed because an unbounded `get_workspace` is the step the integration test makes slow; leaving it outside the guard would let the whole request hang past the timeout anyway.)

 On a trip, each site returns a structured `Envelope.gateway_timeout(component, timeout_seconds, remediate)` (new classmethod on `roboco/services/gateway/envelope.py`'s `Envelope`) naming which segment stalled (e.g. `"git diff/fetch or a journal/ancestor/findings DB read"`) and a remediation hint, instead of the outer 120s rollback with no indication of which piece stuck. The route's HTTP path raises a `504 Gateway Timeout` with the equivalent detail message.

@@ -40,7 +40,7 @@ Each segment logs its own duration (structlog `.info()` calls) so a slow request
 - `GitService.diff_and_files` logs `resolve_ms` (workspace/token/head/base resolution) and `diff_ms` (the concurrent diff subprocesses) as "evidence diff_and_files timing".
 - `conventions_check_for_task` logs `total_ms` plus whether it reused the passed-in file list, as "conventions_check_for_task timing".
 - `evidence()` logs `git_diff_and_fetch_ms` ("evidence git diff/fetch timing") and the DB-reads batch logs `db_reads_ms` ("evidence db reads timing").
- `claim_review`'s two segment helpers (`_qa_git_and_conventions`, `_qa_db_reads`) log `git_diff_and_fetch_ms`/`conventions_ms` ("claim_review git+conventions timing") and `db_reads_ms` ("claim_review db reads timing") respectively.
+- `claim_review`'s two segment helpers (`_qa_git_and_conventions`, `_qa_db_reads`) log `git_diff_and_fetch_ms` ("claim_review git+conventions timing", covering the whole git+conventions block as one measurement) and `db_reads_ms` ("claim_review db reads timing") respectively.

 ## Related files

@@ -33,6 +33,8 @@ from roboco.api.deps import CurrentAgentContext, DbSession
 from roboco.api.schemas.git import (
    BranchInfo,
    CommitInfo,
+    GitBranchCleanupRequest,
+    GitBranchCleanupResponse,
    GitBranchListResponse,
    GitCheckoutRequest,
    GitCheckoutResponse,
@@ -45,6 +47,7 @@ from roboco.api.schemas.git import (
    GitDiffResponse,
    GitFetchRequest,
    GitFetchResponse,
+    GitFileContentResponse,
    GitLogResponse,
    GitMergePRRequest,
    GitMergePRResponse,
@@ -89,6 +92,43 @@ _LOG_FORMAT_PARTS = 5
 # bubbling as 500 Internal Server Errors with no `detail`.
 _TranslatableError = (ServiceError, GitError)

+# Cap an unbounded whole-file read so a huge file can't flood the panel.
+_FILE_MAX_LINES = 2000
+
+
+def _compute_file_range(
+    *,
+    total: int,
+    line: int | None,
+    context: int,
+    start: int | None,
+    end: int | None,
+) -> tuple[int, int, bool]:
+    """Resolve the (start, end, truncated) slice for a file-content read.
+
+    Explicit ``start``/``end`` win; else ``line`` centers a context window;
+    else the whole file. Whichever branch resolves the window, it is capped
+    at ``_FILE_MAX_LINES`` lines afterward. Returns 1-based inclusive
+    [start, end] and whether the slice is shorter than the file.
+    """
+    if start is not None and end is not None:
+        s, e_ = start, end
+    elif line is not None:
+        s = max(1, line - context)
+        e_ = min(total, line + context)
+    else:
+        s, e_ = 1, total
+
+    s = max(1, min(s, total))
+    e_ = max(s, min(e_, total))
+
+    truncated = e_ < total
+    if e_ - s + 1 > _FILE_MAX_LINES:
+        e_ = s + _FILE_MAX_LINES - 1
+        truncated = True
+    return s, e_, truncated
+
+
 # Roles permitted to rebase branches via the /rebase endpoint.
 # Rebase is a history-rewriting operation that should be authorised only by
 # PM-level or CEO-level callers. Developers are intentionally excluded:
@@ -206,6 +246,18 @@ async def get_git_log(
        if not branch:
            branch = await git_service.get_current_branch(workspace)

+        # This is the CALLER's own clone, which is never the branch's own
+        # author when inspecting another agent's task (QA/PM/documenter
+        # reading a dev's branch) — a local ref left over from an earlier
+        # inspection can be pinned stale (behind, or diverged after a
+        # routine rebase force-push) while origin has since moved. Resolve
+        # through _resolve_head_ref (fetch + prefer origin) instead of the
+        # bare branch name so this reads the same authoritative tip diff()/
+        # read_file_at_branch() do, not whatever this clone happened to
+        # have on disk from the last time it looked.
+        token = await git_service._token_for_branch(branch)
+        head_ref = await git_service._resolve_head_ref(workspace, branch, token=token)
+
        # Get log with format. Don't raise if the branch doesn't exist in
        # this workspace yet — that's a normal race (branch created in a
        # different agent's clone, not yet fetched here). Return empty.
@@ -216,7 +268,7 @@ async def get_git_log(
        log_format = "%H%x1f%h%x1f%s%x1f%an%x1f%aI"
        log_result = await git_service._run_git(
            workspace,
-            ["log", f"--format={log_format}", f"-n{limit}", branch],
+            ["log", f"--format={log_format}", f"-n{limit}", head_ref],
            check=False,
        )
        if log_result.returncode != 0:
@@ -253,6 +305,28 @@ async def get_git_log(
    )


+def _parse_branch_line(line: str) -> tuple[str, bool, str | None] | None:
+    """Classify one `%(refname)|%(objectname:short)` line as (name, is_remote,
+    last_commit), or None for skippable entries (blank, origin/HEAD, other ref
+    namespaces). Full refname, not `:short` — a remote-tracking ref shortens to
+    `origin/<branch>`, indistinguishable from a local branch literally named
+    that; classify on the `refs/heads/` vs `refs/remotes/` prefix instead.
+    """
+    if not line:
+        return None
+    parts = line.split("|")
+    ref = parts[0]
+    last_commit = parts[1] if len(parts) > 1 else None
+    if ref.startswith("refs/heads/"):
+        return ref.removeprefix("refs/heads/"), False, last_commit
+    if ref.startswith("refs/remotes/"):
+        _remote_name, _, name = ref.removeprefix("refs/remotes/").partition("/")
+        if not name or name == "HEAD":
+            return None  # origin/HEAD is a symbolic pointer, not a branch
+        return name, True, last_commit
+    return None
+
+
@router.get("/branches", response_model=GitBranchListResponse)
 async def list_branches(
    db: DbSession,
@@ -268,8 +342,12 @@ async def list_branches(
        workspace = await git_service.get_workspace(project_slug, agent.agent_id)
        current_branch = await git_service.get_current_branch(workspace)

-        # Get branches
-        args = ["branch", "--format=%(refname:short)|%(objectname:short)"]
+        if include_remote:
+            # Self-heal orphaned remote-tracking refs (branches deleted
+            # upstream via the forge API) before listing them.
+            await git_service.prune_remote_best_effort(workspace)
+
+        args = ["branch", "--format=%(refname)|%(objectname:short)"]
        if include_remote:
            args.append("-a")

@@ -279,16 +357,10 @@ async def list_branches(

    branches = []
    for line in branch_result.stdout.strip().split("\n"):
-        if not line:
+        parsed = _parse_branch_line(line)
+        if parsed is None:
            continue
-        parts = line.split("|")
-        name = parts[0]
-        last_commit = parts[1] if len(parts) > 1 else None
-
-        is_remote = name.startswith("remotes/")
-        if is_remote:
-            name = name.replace("remotes/origin/", "")
-
+        name, is_remote, last_commit = parsed
        branches.append(
            BranchInfo(
                name=name,
@@ -315,49 +387,50 @@ async def get_git_diff(
 ) -> GitDiffResponse:
    """Get git diff for a project.

-    The diff + stat subprocesses run inside a bounded
-    ``settings.evidence_assembly_timeout_seconds`` guard — well under the
-    outer gateway verb budget — so a genuinely slow diff computation (a
-    huge working-tree change, a hung git process) returns a structured 504
-    naming the slow component instead of hanging indefinitely.
+    The diff + stat subprocess pair is wrapped in a bounded
+    ``evidence_assembly_timeout_seconds`` guard so a genuinely slow diff
+    computation returns a structured 504 instead of hanging indefinitely.
    """
    project_slug = await _resolve_project_slug(project_slug, db)
    git_service = get_git_service(db)
-
-    async def _run_diff_and_stat() -> tuple[Any, Any]:
-        workspace = await git_service.get_workspace(project_slug, agent.agent_id)
-
-        args = ["diff"]
-        if staged:
-            args.append("--staged")
-        if file_path:
-            args.extend(["--", file_path])
-        diff_res = await git_service._run_git(workspace, args)
-
-        # Count files changed
-        stat_args = ["diff", "--stat"]
-        if staged:
-            stat_args.append("--staged")
-        stat_res = await git_service._run_git(workspace, stat_args)
-        return diff_res, stat_res
+    timeout = settings.evidence_assembly_timeout_seconds

    try:
+        # Wrap workspace resolution + diff + stat subprocesses in one bounded
+        # timeout so a slow workspace fetch or diff computation returns a
+        # structured 504 instead of hanging indefinitely.
+        async def _resolve_and_diff() -> tuple[Any, Any]:
+            workspace = await git_service.get_workspace(project_slug, agent.agent_id)
+
+            args = ["diff"]
+            if staged:
+                args.append("--staged")
+            if file_path:
+                args.extend(["--", file_path])
+
+            stat_args = ["diff", "--stat"]
+            if staged:
+                stat_args.append("--staged")
+
+            diff_result, stat_result = await asyncio.gather(
+                git_service._run_git(workspace, args),
+                git_service._run_git(workspace, stat_args),
+            )
+            return diff_result, stat_result
+
        diff_result, stat_result = await asyncio.wait_for(
-            _run_diff_and_stat(),
-            timeout=settings.evidence_assembly_timeout_seconds,
+            _resolve_and_diff(), timeout=timeout
        )
-    except TimeoutError as e:
+    except _TranslatableError as e:
+        raise _translate_error(e) from e
+    except TimeoutError:
        raise HTTPException(
            status_code=status.HTTP_504_GATEWAY_TIMEOUT,
            detail=(
-                "diff computation exceeded the "
-                f"{settings.evidence_assembly_timeout_seconds:.0f}s bounded "
-                "timeout — retry, or scope the request with file_path to a "
-                "single file"
+                f"git diff timed out after {timeout}s — the bounded "
+                "evidence_assembly_timeout_seconds guard tripped"
            ),
-        ) from e
-    except _TranslatableError as e:
-        raise _translate_error(e) from e
+        ) from None

    files_changed = stat_result.stdout.count("\n") - 1 if stat_result.stdout else 0

@@ -370,6 +443,60 @@ async def get_git_diff(
    )


+@router.get("/file", response_model=GitFileContentResponse)
+async def get_git_file(
+    *,
+    db: DbSession,
+    agent: CurrentAgentContext,
+    branch: str = Query(..., description="Task branch holding the file"),
+    path: str = Query(..., description="Repo-relative file path"),
+    line: int | None = Query(default=None, ge=1, description="Target line"),
+    context: int = Query(default=10, ge=0, le=100, description="Context around line"),
+    start: int | None = Query(default=None, ge=1, description="Explicit start line"),
+    end: int | None = Query(default=None, ge=1, description="Explicit end line"),
+) -> GitFileContentResponse:
+    """Return a file's content at a branch tip, optionally sliced to a range.
+
+    Reads straight out of the branch with ``git show`` (``read_file_at_branch``)
+    so a reviewer/CEO can read a finding's source lines without a workspace
+    mount. A missing file or bad ref yields 404. When `line` is given the
+    slice is centered on it (`line - context` .. `line + context`); explicit
+    `start`/`end` override. With none of the three, the whole file returns
+    (capped at 2000 lines to bound the payload — `truncated` flags the cut).
+    """
+    git_service = get_git_service(db)
+
+    try:
+        content = await git_service.read_file_at_branch(
+            branch_name=branch, path=path, actor_agent_id=agent.agent_id
+        )
+    except _TranslatableError as e:
+        raise _translate_error(e) from e
+
+    if content is None:
+        raise HTTPException(
+            status_code=status.HTTP_404_NOT_FOUND,
+            detail=f"File not found at {branch}:{path}",
+        )
+
+    all_lines = content.splitlines()
+    total = len(all_lines)
+
+    s, e_, truncated = _compute_file_range(
+        total=total, line=line, context=context, start=start, end=end
+    )
+
+    sliced = all_lines[s - 1 : e_]
+    return GitFileContentResponse(
+        branch=branch,
+        path=path,
+        content="\n".join(sliced),
+        start_line=s,
+        total_lines=total,
+        truncated=truncated,
+    )
+
+
 # =============================================================================
 # WRITE ENDPOINTS
 # =============================================================================
@@ -688,7 +815,7 @@ async def rebase_branch(
    try:
        workspace = await git_service.get_workspace(project_slug, agent.agent_id)
        conflict, conflicted_files = await git_service.rebase(
-            workspace, data.target_branch
+            workspace, data.target_branch, project_slug
        )
    except _TranslatableError as e:
        raise _translate_error(e) from e
@@ -698,3 +825,60 @@ async def rebase_branch(
        conflict=conflict,
        conflicted_files=conflicted_files,
    )
+
+
+@router.post("/branches/cleanup", response_model=GitBranchCleanupResponse)
+@guard_deco.rate_limit(requests=5, window=60)
+@guard_deco.max_request_size(size_bytes=65536)
+@guard_deco.block_clouds()
+@guard_deco.content_type_filter(["application/json"])
+async def cleanup_stale_branches(
+    data: GitBranchCleanupRequest,
+    db: DbSession,
+    agent: CurrentAgentContext,
+) -> GitBranchCleanupResponse:
+    """Sweep a project's terminal-task branches (PM/CEO only).
+
+    Deletes the remote + local branch of every completed/cancelled task in
+    the project (capped per call — see ``GitService.cleanup_stale_branches``),
+    skipping the default branch and any environment-ladder rung so a live
+    integration/prod branch is never touched. Role-gated identically to
+    ``/rebase`` — a history-affecting bulk operation shouldn't be open to
+    developers either. Purely a read + external-git-op endpoint: no task rows
+    are mutated, so there's nothing for this request to commit.
+    """
+    if agent.role not in _REBASE_ALLOWED_ROLES:
+        raise HTTPException(
+            status_code=status.HTTP_403_FORBIDDEN,
+            detail=(
+                f"BRANCH_CLEANUP_ROLE_RESTRICTED: Role '{agent.role}' is not "
+                "permitted to sweep branches. Only CEO and PM roles (cell_pm, "
+                "main_pm) may use this endpoint."
+            ),
+        )
+    project_slug = await _resolve_project_slug(data.project_slug, db)
+    git_service = get_git_service(db)
+
+    try:
+        (
+            remote_deleted,
+            local_deleted,
+            skipped,
+            errors,
+            truncated,
+            next_cursor,
+        ) = await git_service.cleanup_stale_branches(
+            project_slug, after_task_id=data.after_cursor
+        )
+    except _TranslatableError as e:
+        raise _translate_error(e) from e
+
+    return GitBranchCleanupResponse(
+        project_slug=project_slug,
+        remote_deleted=remote_deleted,
+        local_deleted=local_deleted,
+        skipped=skipped,
+        errors=errors,
+        truncated=truncated,
+        next_cursor=next_cursor,
+    )
@@ -38,7 +38,7 @@ class Settings(BaseSettings):
    # ==========================================================================
    # Application
    # ==========================================================================
-    app_version: str = "0.27.0"
+    app_version: str = "0.28.0"
    debug: bool = False
    environment: str = Field(
        default="development", pattern="^(development|staging|production)$"
@@ -1649,6 +1649,28 @@ class Settings(BaseSettings):
            "legitimate verbs are unaffected."
        ),
    )
+    evidence_assembly_timeout_seconds: float = Field(
+        default=90.0,
+        ge=1.0,
+        description=(
+            "Bounded inner-work timeout for the evidence-assembly segment of "
+            "claim_review / evidence() / roboco_git_diff — git diff/fetch, "
+            "conventions validation, and the DB reads that accompany them. "
+            "Kept comfortably below flow_verb_timeout_seconds so a slow "
+            "assembly returns a structured gateway_timeout envelope naming "
+            "the stalled component instead of hitting the outer 120s rollback."
+        ),
+    )
+    conventions_validator_timeout_seconds: int = Field(
+        default=45,
+        ge=1,
+        description=(
+            "Timeout for the conventions-validator subprocess "
+            "(python -m roboco.conventions check), kept comfortably below "
+            "evidence_assembly_timeout_seconds so a hung validator can't by "
+            "itself exhaust the outer verb's budget."
+        ),
+    )
    flow_verb_slow_timeout_seconds: int = Field(
        default=900,
        ge=1,
@@ -1706,30 +1728,6 @@ class Settings(BaseSettings):
            "repository; set this to sandbox smoke-test projects."
        ),
    )
-    evidence_assembly_timeout_seconds: float = Field(
-        default=90.0,
-        ge=1.0,
-        description=(
-            "Bounded inner-work timeout for the evidence-assembly segment of "
-            "claim_review / evidence() / roboco_git_diff (git fetch, diff "
-            "computation, conventions validation, DB reads). Well under "
-            "flow_verb_timeout_seconds (120s default) so a genuinely slow "
-            "sub-component returns a structured, named gateway_timeout "
-            "envelope instead of the whole verb hitting the outer server-side "
-            "rollback with no indication of which segment stalled."
-        ),
-    )
-    conventions_validator_timeout_seconds: int = Field(
-        default=45,
-        ge=5,
-        description=(
-            "Timeout in seconds for the conventions-validator subprocess "
-            "(``python -m roboco.conventions check``) run inside "
-            "claim_review / i_am_done / pr_pass. Kept comfortably below "
-            "evidence_assembly_timeout_seconds so a hung validator can't by "
-            "itself exhaust the outer verb's whole server-side budget."
-        ),
-    )

    # ==========================================================================
    # Agent Guardrails (per-session budgets, loop detection, SLAs)
@@ -49,16 +49,21 @@ from roboco.foundation.policy import tracing as _tr
 from roboco.foundation.policy.content import (
    ContentValidationError,
    markers,
-    validate_findings,
 )
 from roboco.services.content_notes import apply_structured_note
 from roboco.services.gateway.choreographer import findings as findings_lib
 from roboco.services.gateway.choreographer._protocol import actor_context_fields
+from roboco.services.gateway.choreographer.collision import build_collision_context
 from roboco.services.gateway.envelope import Envelope
 from roboco.services.gateway.evidence_builder import build_evidence_for_task

 logger = structlog.get_logger()

+# Cap on one criteria_verified entry's `evidence` — mirrors Finding.fix's cap
+# (roboco.foundation.policy.content.models._FINDING_FIX_CAP): a pointer
+# (file:line, screenshot ref, test name), not a transcript.
+_CRITERION_EVIDENCE_CAP = 500
+
 if TYPE_CHECKING:
    from uuid import UUID

@@ -177,37 +182,32 @@ class QAMixin(_Base):
        t = await self.task.qa_claim(qa_agent_id, task_id)
        await self.task.mark_evidence_inspected(task_id)

+        # Bounded-timeout wrapper around evidence assembly: the claim itself
+        # (qa_claim + mark_evidence_inspected) has already committed by this
+        # point, so a trip here means only evidence assembly is slow. The
+        # agent gets a structured gateway_timeout envelope pointing at
+        # evidence(task_id) (which re-runs the slow path but not the claim)
+        # instead of the outer 120s rollback with no indication of which
+        # piece stuck.
+        timeout = settings.evidence_assembly_timeout_seconds
        try:
            ev = await asyncio.wait_for(
                self._build_qa_claim_evidence(qa_agent_id, t, task_id),
-                timeout=settings.evidence_assembly_timeout_seconds,
+                timeout=timeout,
            )
        except TimeoutError:
-            # The claim itself already committed (qa_claim + mark_evidence_
-            # inspected above) — well under flow_verb_timeout_seconds's own
-            # ceiling. Only evidence assembly (git diff/fetch, conventions
-            # validation, or a DB read) is slow, so name it and point at the
-            # cheap fallback instead of a bare 504 that looks like the whole
-            # claim failed.
-            return await self._emit_rejection(
-                Envelope.gateway_timeout(
-                    component=(
-                        "evidence assembly (git diff/fetch, conventions "
-                        "validation, or a journal/findings DB read)"
-                    ),
-                    timeout_seconds=settings.evidence_assembly_timeout_seconds,
-                    remediate=(
-                        "the claim already succeeded (status is awaiting_qa, "
-                        "you are the claimant) — call evidence(task_id) to "
-                        "fetch the PR diff/files/findings directly instead "
-                        "of retrying claim_review"
-                    ),
-                    context_briefing=briefing,
-                ).with_introspection(task=t, role=role_str),
-                agent_id=qa_agent_id,
-                task_id=task_id,
-                verb="claim_review",
-            )
+            return Envelope.gateway_timeout(
+                component="git diff/fetch or a journal/ancestor/findings DB read",
+                timeout_seconds=timeout,
+                remediate=(
+                    "the claim itself already committed — call "
+                    "evidence(task_id) to re-fetch the PR diff, "
+                    "journal highlights, and findings without re-running "
+                    "the claim; or retry claim_review after a short wait"
+                ),
+                task_id=str(task_id),
+                context_briefing=briefing,
+            ).with_introspection(task=t, role=role_str)
        return Envelope.ok(
            status=str(t.status),
            task_id=str(task_id),
@@ -217,55 +217,155 @@ class QAMixin(_Base):
        ).with_introspection(task=t, role=role_str)

    async def _qa_convention_findings(
-        self,
-        qa_agent_id: UUID,
-        t: Any,
-        *,
-        changed_files: list[str] | None = None,
+        self, qa_agent_id: UUID, t: Any
    ) -> list[dict[str, Any]]:
        """Convention-validator findings on the task's changed files (flag-gated).

        Empty when the subsystem is off; a validator that could not run surfaces
        a single explicit ``could_not_run`` entry rather than being dropped, so
        QA never mistakes a silent failure for a clean diff.
-
-        ``changed_files``, when given, is passed through to
-        ``conventions_check_for_task`` so it skips its own redundant
-        ``list_changed_files`` call (a third re-derivation of the same list
-        on one claim_review request).
        """
        if not settings.conventions_enabled:
            return []
-        result = await self.git.conventions_check_for_task(
-            qa_agent_id, t, changed_files=changed_files
-        )
+        result = await self.git.conventions_check_for_task(qa_agent_id, t)
        if result.get("could_not_run"):
            reason = result.get("reason") or "validator could not run"
            return [{"could_not_run": True, "reason": reason}]
        return list(result.get("findings", []))

+    @staticmethod
+    def _qa_video_context(t: Any) -> dict[str, Any] | None:
+        """QA-facing artifact context for a video-authoring task.
+
+        None for every non-video task. Carries the composition id (from the
+        ``video_draft`` marker) + the latest ``request_render`` preview so QA
+        is pointed at the rendered artifact instead of the source alone.
+        """
+        if getattr(t, "source", None) != markers.VIDEO_TASK_SOURCE:
+            return None
+        draft = markers.get_video_draft(t) or {}
+        return {
+            "composition_id": draft.get("composition_id"),
+            "render_preview": markers.get_render_preview(t),
+            "note": (
+                "This task ships a rendered video. Call request_render to "
+                "render the PR branch state, then Read every returned frame "
+                "image — verify each acceptance criterion's scene appears "
+                "fully and legibly. Do not pass on source reading alone."
+            ),
+        }
+
+    async def _build_qa_claim_evidence(
+        self, qa_agent_id: UUID, t: Any, task_id: UUID
+    ) -> Any:
+        """Assemble the inline evidence payload returned by claim_review.
+
+        Bundles files_changed + pr_diff_summary (both from git, the
+        authoritative source) + journal_highlights so the QA agent has
+        the full PR context up-front and can't miss a piece.
+
+        files_changed comes from ``git.diff_and_files`` (a single call
+        that resolves the shared workspace/token/head/base state once and
+        runs the diff + --name-only subprocesses concurrently) instead of
+        the legacy ``work_session.files_modified`` path. The git+conventions
+        segment and the DB-reads segment run concurrently via
+        ``asyncio.gather`` so the git fetch overlaps the DB queries instead
+        of running serially.
+        """
+        git_coro = self._qa_git_and_conventions(qa_agent_id, t)
+        db_coro = self._qa_db_reads(task_id, t)
+        (
+            (diff_summary, files_changed, convention_findings),
+            (
+                journal_highlights,
+                parent_context,
+                open_findings,
+                prior_findings,
+            ),
+        ) = await asyncio.gather(git_coro, db_coro)
+
+        # The collision map: surfaced siblings (same parent) that would
+        # collide with this task, with the declared-vs-actual drift (QA has
+        # the real touched files in hand). Best-effort — a fetch failure omits
+        # the block rather than breaking claim_review.
+        collision_context: list[dict[str, Any]] | None = None
+        try:
+            if t.parent_task_id:
+                siblings = await self.task.get_subtasks(t.parent_task_id)
+                collision_context = build_collision_context(
+                    task=t, siblings=siblings, actual_files=files_changed
+                )
+        except Exception as exc:  # best-effort enrichment, never breaks the verb
+            logger.warning(
+                "qa_collision_context_skip", task_id=str(t.id), error=str(exc)
+            )
+        return build_evidence_for_task(
+            t,
+            journal_highlights=journal_highlights,
+            files_changed=files_changed,
+            pr_diff_summary=diff_summary,
+            convention_findings=convention_findings,
+            revision_findings=open_findings,
+            prior_findings=prior_findings,
+            parent_context=parent_context,
+            collision_context=collision_context,
+            video_context=self._qa_video_context(t),
+        )
+
+    async def _qa_git_and_conventions(
+        self, qa_agent_id: UUID, t: Any
+    ) -> tuple[str, list[str], list[dict[str, Any]]]:
+        """Git diff/fetch + conventions validation segment of claim_review.
+
+        Returns ``(diff_summary, files_changed, convention_findings)``.
+        Uses ``diff_and_files`` (single resolution, concurrent subprocesses)
+        and passes the file list straight into ``conventions_check_for_task``
+        so the changed-file list is computed once, not three times.
+        """
+        git_start = time.monotonic()
+        files_changed: list[str] = []
+        diff_summary = ""
+        convention_findings: list[dict[str, Any]] = []
+        if t.branch_name:
+            diff_summary, files_changed = await self.git.diff_and_files(
+                branch_name=t.branch_name, actor_agent_id=qa_agent_id
+            )
+            # Pass the file list through so conventions_check_for_task
+            # doesn't trigger a redundant list_changed_files call.
+            if not settings.conventions_enabled:
+                convention_findings = []
+            else:
+                result = await self.git.conventions_check_for_task(
+                    qa_agent_id,
+                    t,
+                    changed_files=files_changed,
+                )
+                if result.get("could_not_run"):
+                    reason = result.get("reason") or "validator could not run"
+                    convention_findings = [{"could_not_run": True, "reason": reason}]
+                else:
+                    convention_findings = list(result.get("findings", []))
+        git_ms = (time.monotonic() - git_start) * 1000
+        logger.info(
+            "claim_review git+conventions timing",
+            git_diff_and_fetch_ms=round(git_ms, 1),
+            task_id=str(t.id),
+        )
+        return diff_summary, files_changed, convention_findings
+
    async def _qa_db_reads(
        self, task_id: UUID, t: Any
-    ) -> tuple[list[dict[str, Any]], list[dict[str, Any]], list[Any], list[Any]]:
-        """The four DB reads behind claim_review's evidence, run sequentially
-        as ONE coroutine.
+    ) -> tuple[Any, Any, list[Any], list[Any]]:
+        """DB-reads segment of claim_review — runs concurrently with git work.

-        They stay sequential AMONG THEMSELVES on purpose: they all read
-        through the request-scoped ``self.task.session`` / evidence_repo, a
-        single ``AsyncSession`` bound to one DBAPI connection — asyncpg
-        cannot serve two in-flight queries on the same connection at once
-        (``asyncpg.exceptions`` / SQLAlchemy's own
-        ``IllegalStateChangeError: ... concurrent operations are not
-        permitted``, confirmed against a live Postgres while building this
-        fix; opening a second ad-hoc session per read would silently diverge
-        from the request's transaction and break every existing test that
-        injects a mocked ``evidence_repo``). Bundling them as one coroutine
-        lets ``_build_qa_claim_evidence`` still `asyncio.gather` this WHOLE
-        batch against the independent git + conventions-validator work
-        below — the actual "parallelize the independent reads" win, without
-        the same-connection hazard.
+        Returns ``(journal_highlights, parent_context, open_findings,
+        prior_findings)``. Reads stay sequential *within* this coroutine on
+        purpose: they all read through the same request-scoped AsyncSession,
+        and asyncpg cannot serve two in-flight queries on one connection at
+        once (IllegalStateChangeError) — the win is git-work-vs-DB-work
+        overlap via ``asyncio.gather``, not fanning out the DB reads.
        """
-        t0 = time.monotonic()
+        db_start = time.monotonic()
        journal_highlights = await self.evidence_repo.journal_highlights_for_task(
            task_id
        )
@@ -280,90 +380,14 @@ class QAMixin(_Base):
        prior_findings = await findings_lib.full_ledger_for_task(
            self.task.session, t.id
        )
+        db_ms = (time.monotonic() - db_start) * 1000
        logger.info(
            "claim_review db reads timing",
-            task_id=str(task_id),
-            db_reads_ms=round((time.monotonic() - t0) * 1000.0),
+            db_reads_ms=round(db_ms, 1),
+            task_id=str(t.id),
        )
        return journal_highlights, parent_context, open_findings, prior_findings

-    async def _qa_git_and_conventions(
-        self, qa_agent_id: UUID, t: Any
-    ) -> tuple[str, list[str], list[dict[str, Any]]]:
-        """The git diff/fetch + conventions-validator segment, as ONE
-        coroutine so it can run concurrently with the DB-reads batch.
-
-        The conventions check needs ``files_changed`` from the git segment,
-        so within this coroutine it still runs AFTER git — but that chain as
-        a whole now overlaps the independent DB reads (item d of the
-        claim_review timeout fix), instead of the old strictly-serial
-        diff -> list_changed_files -> conventions(list_changed_files again)
-        -> DB reads chain.
-        """
-        files_changed: list[str] = []
-        diff_summary = ""
-        git_t0 = time.monotonic()
-        if t.branch_name:
-            diff_summary, files_changed = await self.git.diff_and_files(
-                branch_name=t.branch_name
-            )
-        git_ms = (time.monotonic() - git_t0) * 1000.0
-        conv_t0 = time.monotonic()
-        convention_findings = await self._qa_convention_findings(
-            qa_agent_id, t, changed_files=files_changed
-        )
-        conv_ms = (time.monotonic() - conv_t0) * 1000.0
-        logger.info(
-            "claim_review git+conventions timing",
-            task_id=str(getattr(t, "id", None)),
-            git_diff_and_fetch_ms=round(git_ms),
-            conventions_ms=round(conv_ms),
-        )
-        return diff_summary, files_changed, convention_findings
-
-    async def _build_qa_claim_evidence(
-        self, qa_agent_id: UUID, t: Any, task_id: UUID
-    ) -> Any:
-        """Assemble the inline evidence payload returned by claim_review.
-
-        Bundles files_changed + pr_diff_summary (both from git, the
-        authoritative source) + journal_highlights so the QA agent has
-        the full PR context up-front and can't miss a piece.
-
-        files_changed comes from the combined ``diff_and_files``
-        accessor instead of ``work_session.files_modified``. The legacy
-        ``add_files_modified`` HTTP path that populated files_modified
-        is not called by the gateway ``commit()``, so the work_session
-        list was always empty — QA saw no files even on real PRs.
-
-        The git+conventions segment and the DB-reads segment run
-        concurrently via ``asyncio.gather`` (each internally sequential per
-        ``_qa_db_reads``'s docstring); per-segment timing is logged inside
-        each helper (item a of the claim_review timeout fix).
-        """
-        (
-            (diff_summary, files_changed, convention_findings),
-            (
-                journal_highlights,
-                parent_context,
-                open_findings,
-                prior_findings,
-            ),
-        ) = await asyncio.gather(
-            self._qa_git_and_conventions(qa_agent_id, t),
-            self._qa_db_reads(task_id, t),
-        )
-        return build_evidence_for_task(
-            t,
-            journal_highlights=journal_highlights,
-            files_changed=files_changed,
-            pr_diff_summary=diff_summary,
-            convention_findings=convention_findings,
-            revision_findings=open_findings,
-            prior_findings=prior_findings,
-            parent_context=parent_context,
-        )
-
    async def _verify_qa_owner(
        self, qa_agent_id: UUID, task_id: UUID, verb: str
    ) -> tuple[Envelope | None, Any]:
@@ -502,13 +526,13 @@ class QAMixin(_Base):
    def _qa_ac_coverage_check(
        cls, task: Any, ac_verdicts: list[str] | None
    ) -> Envelope | None:
-        """Per-acceptance-criterion verification gate for pass_review.
+        """Legacy count-only per-AC gate — superseded by ``criteria_verified``.

-        QA may not pass a task until it has recorded a verification for EVERY
-        acceptance criterion. A single gestalt "looks good" approval is how a
-        silently-unbuilt criterion slips through; requiring one verdict per
-        criterion forces QA to check each individually. If a criterion does not
-        hold, the QA fails the review instead of passing a partial.
+        No longer wired into ``pass_review`` (a count of arbitrary strings
+        never verified they actually named the right criterion — the live gap
+        ``_validate_criteria_verified`` closes). Kept for ``ac_verdicts``'
+        existing callers/tests; ``ac_verdicts`` itself still folds into the
+        persisted notes when supplied.
        """
        criteria = list(getattr(task, "acceptance_criteria", None) or [])
        if not criteria:
@@ -620,36 +644,181 @@ class QAMixin(_Base):
            verb=verb,
        )

+    @staticmethod
+    def _parse_criterion_entry(entry: Any, idx: int) -> tuple[str, str] | Envelope:
+        """Validate one ``criteria_verified`` entry into a (criterion, evidence)
+        pair, or an ``Envelope`` rejection on any structural problem."""
+        criterion = entry.get("criterion") if isinstance(entry, dict) else None
+        evidence = entry.get("evidence") if isinstance(entry, dict) else None
+        if not isinstance(criterion, str) or not criterion.strip():
+            return Envelope.invalid_state(
+                message=f"criteria_verified[{idx}] is missing a `criterion` string",
+                remediate=(
+                    "each entry needs {criterion, evidence} naming one "
+                    "acceptance criterion"
+                ),
+            )
+        if not isinstance(evidence, str) or not evidence.strip():
+            return Envelope.invalid_state(
+                message=(
+                    f"criteria_verified[{idx}] ({criterion!r}) is missing `evidence`"
+                ),
+                remediate=(
+                    "state concrete evidence: file:line, screenshot ref, "
+                    "rendered-frame path, test name"
+                ),
+            )
+        if len(evidence) > _CRITERION_EVIDENCE_CAP:
+            return Envelope.invalid_state(
+                message=(
+                    f"criteria_verified[{idx}] evidence exceeds "
+                    f"{_CRITERION_EVIDENCE_CAP} chars"
+                ),
+                remediate="keep evidence concise — a pointer, not a transcript",
+            )
+        return criterion.strip(), evidence.strip()
+
+    @classmethod
+    def _parse_criteria_verified_entries(
+        cls, criteria_verified: list[dict[str, Any]]
+    ) -> tuple[list[tuple[str, str]], Envelope | None]:
+        """Shape + soup validation for every ``criteria_verified`` entry.
+
+        Pure parsing — AC matching/coverage is the caller's job. Split out
+        of ``_validate_criteria_verified`` to keep its return count under
+        the complexity bound.
+        """
+        pairs: list[tuple[str, str]] = []
+        for idx, entry in enumerate(criteria_verified):
+            parsed = cls._parse_criterion_entry(entry, idx)
+            if isinstance(parsed, Envelope):
+                return [], parsed
+            pairs.append(parsed)
+        soup = cls._free_text_soup(
+            checks=(("criteria_verified.evidence", [e for _, e in pairs], 8),)
+        )
+        if soup is not None:
+            return [], soup
+        return pairs, None
+
+    @classmethod
+    def _validate_criteria_verified(
+        cls, t: Any, criteria_verified: list[dict[str, Any]] | None
+    ) -> tuple[list[tuple[str, str]], Envelope | None]:
+        """Mandatory per-AC verification gate for pass_review.
+
+        Returns ``(pairs, rejection)`` — ``pairs`` is ``[]`` and ``rejection``
+        non-None on any failure: none supplied (lists every AC verbatim),
+        a malformed or soupy entry, an entry naming a criterion absent from
+        the task (names the valid criteria), or a task AC left uncovered
+        (names the gap). No task ACs imposes no requirement (mirrors the
+        legacy ``_qa_ac_coverage_check``). A gestalt "looks good" is no
+        longer enough — every criterion needs its own matched, evidenced
+        entry, or QA must call ``fail_review`` instead of passing a partial.
+        """
+        criteria = list(getattr(t, "acceptance_criteria", None) or [])
+        if not criteria:
+            return [], None
+        if not criteria_verified:
+            return [], Envelope.invalid_state(
+                message=(
+                    "pass_review needs criteria_verified naming every "
+                    f"acceptance criterion; none supplied. Unverified: {criteria!r}"
+                ),
+                remediate=(
+                    "re-run the review and call pass_review with "
+                    "criteria_verified=[{criterion, evidence}, ...] — stamp "
+                    "EACH criterion with concrete evidence (file:line, "
+                    "screenshot ref, rendered-frame path, test name)"
+                ),
+            )
+        pairs, bad = cls._parse_criteria_verified_entries(criteria_verified)
+        if bad is not None:
+            return [], bad
+        provided = [c for c, _ in pairs]
+        if unknown := findings_lib.unmatched_criteria(t, provided):
+            return [], Envelope.invalid_state(
+                message=(
+                    f"criteria_verified names criteria not on this task: {unknown!r}"
+                ),
+                remediate=(
+                    "each entry's criterion must match one of the task's "
+                    f"acceptance criteria (by id or exact text): {criteria!r}"
+                ),
+            )
+        if uncovered := findings_lib.uncovered_acceptance_criteria(t, provided):
+            return [], Envelope.invalid_state(
+                message=(
+                    "criteria_verified is missing these acceptance criteria: "
+                    f"{uncovered!r}"
+                ),
+                remediate=(
+                    "stamp every criterion with concrete evidence, or call "
+                    "fail_review with the specific gap if one does not hold"
+                ),
+            )
+        return pairs, None
+
+    @staticmethod
+    def _render_criteria_verified(pairs: list[tuple[str, str]]) -> list[str]:
+        """One '[AC] <criterion> — verified: <evidence>' line per entry.
+
+        Style-matched to the findings ledger's '[F-<id8>] ...' bracket-tag
+        rendering (``findings_lib.render_finding_line``).
+        """
+        return [
+            f"[AC] {criterion} — verified: {evidence}" for criterion, evidence in pairs
+        ]
+
+    @classmethod
+    def _merge_criteria_verified_into_notes(
+        cls, notes: str, pairs: list[tuple[str, str]]
+    ) -> str:
+        """Fold the per-AC verification lines into the persisted QA notes.
+
+        Mirrors ``_merge_ac_verdicts_into_notes`` — keeps the per-criterion
+        verification in the audit trail (qa_notes) so PM/CEO see exactly how
+        QA verified each acceptance criterion.
+        """
+        lines = cls._render_criteria_verified(pairs)
+        if not lines:
+            return notes
+        return f"{notes}\n\n" + "\n".join(lines)
+
    async def _qa_pass_final_gates(
        self,
        qa_agent_id: UUID,
        task_id: UUID,
        t: Any,
        role_str: str,
-        ac_verdicts: list[str] | None,
-    ) -> Envelope | None:
-        """AC-coverage + toolchain-runnability gates for pass_review.
+        criteria_verified: list[dict[str, Any]] | None,
+    ) -> tuple[Envelope | None, list[tuple[str, str]]]:
+        """Per-AC verification + toolchain-runnability gates for pass_review.

-        Returns the first rejection (already emitted), else None. QA must not
-        PASS on a workspace that cannot run the suite — that is a source-read
+        Returns ``(rejection, pairs)`` — the first emitted rejection (else
+        None) and the validated ``criteria_verified`` (criterion, evidence)
+        pairs for the caller to render into notes. QA must not PASS on a
+        workspace that cannot run the suite — that is a source-read
        "verification"; fail_review is unaffected.
        """
-        ac_rejection = self._qa_ac_coverage_check(t, ac_verdicts)
-        if ac_rejection is not None:
-            return await self._emit_rejection(
-                ac_rejection.with_introspection(task=t, role=role_str),
+        pairs, bad = self._validate_criteria_verified(t, criteria_verified)
+        if bad is not None:
+            rejection = await self._emit_rejection(
+                bad.with_introspection(task=t, role=role_str),
                agent_id=qa_agent_id,
                task_id=task_id,
                verb="pass_review",
            )
+            return rejection, []
        if toolchain := await self._toolchain_broken_guard(qa_agent_id, t):
-            return await self._emit_rejection(
+            rejection = await self._emit_rejection(
                toolchain.with_introspection(task=t, role=role_str),
                agent_id=qa_agent_id,
                task_id=task_id,
                verb="pass_review",
            )
-        return None
+            return rejection, []
+        return None, pairs

    async def pass_review(
        self,
@@ -657,6 +826,7 @@ class QAMixin(_Base):
        task_id: UUID,
        notes: str,
        ac_verdicts: list[str] | None = None,
+        criteria_verified: list[dict[str, Any]] | None = None,
    ) -> Envelope:
        """QA passes the task; transitions awaiting_qa → awaiting_documentation.

@@ -671,6 +841,16 @@ class QAMixin(_Base):
        The composed atomic ``qa_pass`` is then dispatched through
        ``VerbRunner.run_intent``, after which the verb body reassigns
        the documenter for handoff.
+
+        ``criteria_verified`` ({criterion, evidence} entries) is the
+        mandatory per-AC verification gate (``_validate_criteria_verified``):
+        every one of the task's acceptance criteria must be named by exactly
+        one entry — matched by AC id or exact text, the same match
+        ``fail_review``'s findings ledger uses for its own ``criterion``
+        field — carrying substantive evidence, or the pass is refused. A
+        gestalt "looks good" is no longer enough; QA must walk each
+        criterion. ``ac_verdicts`` (legacy, count-only) still folds into the
+        persisted notes when supplied but no longer gates the pass.
        """
        rejection, t = await self._verify_qa_owner(qa_agent_id, task_id, "pass_review")
        if rejection is not None:
@@ -696,18 +876,22 @@ class QAMixin(_Base):
            soup_checks=(("notes", notes, 8),),
        ):
            return gate_rejection
-        if rej := await self._qa_pass_final_gates(
-            qa_agent_id, task_id, t, role_str, ac_verdicts
-        ):
-            return rej
+        final_rejection, criteria_pairs = await self._qa_pass_final_gates(
+            qa_agent_id, task_id, t, role_str, criteria_verified
+        )
+        if final_rejection is not None:
+            return final_rejection

        briefing = await self._briefing_for(qa_agent_id, task_id)
+        merged_notes = self._merge_criteria_verified_into_notes(
+            self._merge_ac_verdicts_into_notes(notes, ac_verdicts), criteria_pairs
+        )
        spec_ctx = spec_module.Context(
            actor_id=qa_agent_id,
            actor_slug=getattr(agent, "slug", None) if agent is not None else None,
            agent_team=str(agent.team) if agent is not None and agent.team else None,
            original_developer_slug=_extract_original_developer(t),
-            notes=self._merge_ac_verdicts_into_notes(notes, ac_verdicts),
+            notes=merged_notes,
        )
        self._store_qa_note(t, notes, ac_verdicts, passed=True)
        runner = self._verb_runner()
@@ -803,16 +987,9 @@ class QAMixin(_Base):
            )
        if cap := findings_lib.findings_count_guard(raw):
            return [], cap
-        try:
-            validated = validate_findings(raw)
-        except ContentValidationError as exc:
-            return [], Envelope.invalid_state(
-                message=f"malformed finding: {exc.field} — {exc.reason}",
-                remediate=(
-                    "each finding needs expected + actual (file/line/severity/"
-                    "criterion/fix/evidence optional)"
-                ),
-            )
+        validated, bad = findings_lib.validate_or_reject(raw)
+        if bad is not None:
+            return [], bad
        if unknown := findings_lib.unknown_finding_criteria(t, validated):
            return [], findings_lib.criterion_mismatch_rejection(t, unknown)
        return validated, None
@@ -245,19 +245,27 @@ class Envelope:
        component: str,
        timeout_seconds: float,
        remediate: str,
+        task_id: str | None = None,
        context_briefing: dict[str, Any] | None = None,
    ) -> Envelope:
-        """A bounded inner-work timeout tripped (e.g. evidence assembly),
-        well under the outer server-side ``flow_verb_timeout_seconds``
-        rollback (``api.middleware``'s own ``gateway_timeout``). Names the
-        slow ``component`` so the agent isn't left guessing which segment
-        (git fetch, diff computation, conventions validation, a DB read)
-        stalled, instead of a bare 504 that reads as the whole verb failing.
+        """A bounded inner-work timeout tripped — the evidence-assembly segment
+        (git diff/fetch, conventions validation, or a DB read) exceeded
+        ``evidence_assembly_timeout_seconds`` without completing.
+
+        Distinct from the outer ``flow_verb_timeout_seconds`` rollback: that
+        one cancels the whole request transaction with no indication of which
+        piece stuck. This returns a structured, named error so the agent (and
+        the audit log) sees exactly which component stalled and can retry or
+        escalate accordingly. On a trip the verb's own state change (e.g.
+        ``qa_claim``) has already committed, so the remediate points at
+        ``evidence(task_id)`` rather than re-running the whole verb.
        """
        return cls(
            error="gateway_timeout",
+            task_id=task_id,
            message=(
-                f"{component} exceeded the {timeout_seconds:.0f}s bounded timeout"
+                f"evidence assembly timed out after {timeout_seconds}s — "
+                f"the {component} segment did not complete in time"
            ),
            remediate=remediate,
            context_briefing=context_briefing or {},
@@ -32,6 +32,9 @@ def _make_deps(task_svc: Any = None) -> ChoreographerDeps:
        "audit": AsyncMock(),
        "evidence_repo": AsyncMock(),
    }
+    # diff_and_files returns a (diff_summary, files_changed) 2-tuple; without
+    # this the await resolves to a MagicMock and qa.py:330 unpacking fails.
+    base["git"].diff_and_files.return_value = ("", [])
    repo = base["evidence_repo"]
    for method in (
        "list_unread_a2a",
@@ -41,10 +41,12 @@ def _over_cap_project() -> MagicMock:


 def _make_deps(task_svc: AsyncMock, **overrides: Any) -> ChoreographerDeps:
+    git = AsyncMock()
+    git.diff_and_files.return_value = ("", [])
    base: dict[str, Any] = {
        "task": task_svc,
        "work_session": AsyncMock(),
-        "git": AsyncMock(),
+        "git": git,
        "a2a": AsyncMock(),
        "journal": AsyncMock(),
        "audit": AsyncMock(),
@@ -37,7 +37,7 @@ def _make_deps(**overrides: AsyncMock) -> ContentActionsDeps:
    else:
        git = AsyncMock()
        git.commit.return_value = {"sha": "abc12345"}
-        git.diff.return_value = ""
+        git.diff_and_files.return_value = ("", [])

    a2a = overrides.get("a2a", AsyncMock())
    journal = overrides.get("journal", AsyncMock())
@@ -270,7 +270,7 @@ async def test_evidence_blocks_when_not_assignee() -> None:
    task_svc = AsyncMock()
    task_svc.get.return_value = task_obj
    git_svc = AsyncMock()
-    git_svc.diff.return_value = ""
+    git_svc.diff_and_files.return_value = ("", [])
    workspace_svc = AsyncMock()
    deps = _make_deps(task=task_svc, git=git_svc, workspace=workspace_svc)
    ca = ContentActions(deps)
@@ -279,7 +279,7 @@ async def test_evidence_blocks_when_not_assignee() -> None:
    body = env.as_dict()
    assert body["error"] == "not_authorized"
    workspace_svc.fetch_branch_for_inspection.assert_not_awaited()
-    git_svc.diff.assert_not_awaited()
+    git_svc.diff_and_files.assert_not_awaited()


 # ---------------------------------------------------------------------------
@@ -410,7 +410,7 @@ async def test_evidence_unassigned_task_allows_inspection() -> None:
    task_svc = AsyncMock()
    task_svc.get.return_value = task_obj
    git_svc = AsyncMock()
-    git_svc.diff.return_value = "diff content"
+    git_svc.diff_and_files.return_value = ("diff content", [])
    workspace_svc = AsyncMock()
    deps = _make_deps(task=task_svc, git=git_svc, workspace=workspace_svc)
    ca = ContentActions(deps)
@@ -446,8 +446,7 @@ async def test_evidence_allows_dependency_inspection() -> None:
    task_svc.get.return_value = target
    task_svc.list_assigned_for_agent.return_value = [callers_task]
    git_svc = AsyncMock()
-    git_svc.diff.return_value = ""
-    git_svc.list_changed_files.return_value = []
+    git_svc.diff_and_files.return_value = ("", [])
    workspace_svc = AsyncMock()
    deps = _make_deps(task=task_svc, git=git_svc, workspace=workspace_svc)
    ca = ContentActions(deps)
@@ -7,7 +7,7 @@ Bug:
    commit's delta, even when GitHub showed a multi-commit change set.

 Fix:
-    Pull files via ``git.list_changed_files(branch_name=...)`` (no base
+    Pull files via ``git.diff_and_files(branch_name=...)`` (no base
    → full diff vs parent branch). Pull diff with ``base=None`` so the
    full PR diff comes through. Both use git as the authoritative source
    instead of the legacy ``work_session.files_modified`` field, which
@@ -65,8 +65,10 @@ async def test_evidence_populates_files_changed_from_git() -> None:
    task_svc = AsyncMock()
    task_svc.get.return_value = _task_with_pr(task_id, commits=["abc", "def"])
    git_svc = AsyncMock()
-    git_svc.diff.return_value = "diff --git a/README.md b/README.md\n+added line\n"
-    git_svc.list_changed_files.return_value = ["README.md", "docs/guide.md"]
+    git_svc.diff_and_files.return_value = (
+        "diff --git a/README.md b/README.md\n+added line\n",
+        ["README.md", "docs/guide.md"],
+    )
    workspace_svc = AsyncMock()
    evidence_repo = AsyncMock()
    evidence_repo.journal_highlights_for_task.return_value = []
@@ -80,12 +82,12 @@ async def test_evidence_populates_files_changed_from_git() -> None:
    assert body["error"] is None
    assert body["evidence"]["files_changed"] == ["README.md", "docs/guide.md"]
    assert "diff --git" in body["evidence"]["pr_diff_summary"]
-    git_svc.list_changed_files.assert_awaited_once()
+    git_svc.diff_and_files.assert_awaited_once()


@pytest.mark.asyncio
 async def test_evidence_uses_full_pr_diff_not_head_minus_one() -> None:
-    """git.diff must be called with base=None (full PR diff vs parent),
+    """diff_and_files must be called with base=None (full PR diff vs parent),
    not base='HEAD~1' (only the last commit)."""
    agent_id = uuid4()
    task_id = uuid4()
@@ -94,8 +96,7 @@ async def test_evidence_uses_full_pr_diff_not_head_minus_one() -> None:
    # task.commits was non-empty, masking earlier commits' changes.
    task_svc.get.return_value = _task_with_pr(task_id, commits=["sha1", "sha2", "sha3"])
    git_svc = AsyncMock()
-    git_svc.diff.return_value = "full diff"
-    git_svc.list_changed_files.return_value = []
+    git_svc.diff_and_files.return_value = ("full diff", [])
    workspace_svc = AsyncMock()
    evidence_repo = AsyncMock()
    evidence_repo.journal_highlights_for_task.return_value = []
@@ -105,13 +106,13 @@ async def test_evidence_uses_full_pr_diff_not_head_minus_one() -> None:
    )
    await ca.evidence(agent_id=agent_id, task_id=task_id)

-    git_svc.diff.assert_awaited_once()
-    call_kwargs = git_svc.diff.await_args.kwargs
+    git_svc.diff_and_files.assert_awaited_once()
+    call_kwargs = git_svc.diff_and_files.await_args.kwargs
    # Pre-fix bug: kwargs['base'] would be 'HEAD~1' for any multi-commit
    # branch. Post-fix: base is omitted (or explicitly None).
    base = call_kwargs.get("base")
    assert base in (None, ""), (
-        f"git.diff must use full-PR diff (base=None), got base={base!r}"
+        f"diff_and_files must use full-PR diff (base=None), got base={base!r}"
    )
    assert call_kwargs.get("branch_name") == "feature/backend/abc12345--def67890"

@@ -125,8 +126,7 @@ async def test_evidence_populates_journal_highlights() -> None:
    task_svc = AsyncMock()
    task_svc.get.return_value = _task_with_pr(task_id, commits=["abc"])
    git_svc = AsyncMock()
-    git_svc.diff.return_value = ""
-    git_svc.list_changed_files.return_value = []
+    git_svc.diff_and_files.return_value = ("", [])
    workspace_svc = AsyncMock()
    evidence_repo = AsyncMock()
    highlights = [
@@ -179,5 +179,4 @@ async def test_evidence_no_branch_skips_git_calls() -> None:
    assert body["error"] is None
    assert body["evidence"]["files_changed"] == []
    assert body["evidence"]["pr_diff_summary"] == ""
-    git_svc.diff.assert_not_awaited()
-    git_svc.list_changed_files.assert_not_awaited()
+    git_svc.diff_and_files.assert_not_awaited()
@@ -107,8 +107,7 @@ async def test_claim_review_evidence_carries_prior_findings(
    task_svc.list_paused_for_agent.return_value = []
    task_svc.qa_claim.return_value = t_claimed
    git_svc = AsyncMock()
-    git_svc.diff.return_value = ""
-    git_svc.list_changed_files.return_value = []
+    git_svc.diff_and_files.return_value = ("", [])
    deps = _make_deps(task=task_svc, git=git_svc)
    c = Choreographer(deps)