From 25622aa748987ff12adce10ebb38c1eadc59c7c9 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:58:40 +0800
Subject: [PATCH 01/17] make-pr: preflight runs the PR-body validator on
 --body-file

description_check only read the description's prose for claims about the
repo's past, so preflight printed "ok preflight passed" for a body the
required PR Body check then rejected after publication (#780-#789, #793,
#795). describe() now also runs
engine/skills/draft-pr/scripts/validate-pr-body.mjs on the same file and
prints its errors under the gate line.

Exit 1 from the validator fails preflight. No node, a missing validator,
a timeout, or any other exit code prints "description unchecked: ..." and
fails too -- a check that could not run is not a pass.

The #795 description lands as a fixture: it exits 1 with Test Plan and
Revert Plan outside <details> and code names in the Summary, and the new
tests show preflight failing on it and still passing a valid body.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---
 engine/skills/make-pr/scripts/preflight.py    |  39 ++++-
 .../tests/fixtures/pr795-failing-body.md      |  54 ++++++
 .../make-pr/tests/test_description_check.py   |   6 +
 engine/skills/make-pr/tests/test_preflight.py | 165 ++++++++++++++++++
 4 files changed, 263 insertions(+), 1 deletion(-)
 create mode 100644 engine/skills/make-pr/tests/fixtures/pr795-failing-body.md

diff --git a/engine/skills/make-pr/scripts/preflight.py b/engine/skills/make-pr/scripts/preflight.py
index 36024a5e3..e3c237048 100644
--- a/engine/skills/make-pr/scripts/preflight.py
+++ b/engine/skills/make-pr/scripts/preflight.py
@@ -39,6 +39,9 @@
 
 DEFAULT_CONFIG = os.path.join(REPO_ROOT, "drafter.config.json")
 UNCHECKED_EXIT = 3
+VALIDATOR = "engine/skills/draft-pr/scripts/validate-pr-body.mjs"
+VALIDATOR_NAME = os.path.basename(VALIDATOR)
+VALIDATOR_TIMEOUT = 120
 
 
 class UnitRulesUnreadable(Exception):
@@ -222,6 +225,34 @@ def changed_paths(base: str, repo: str = REPO_ROOT) -> list[str]:
     return sorted({p for p in (out + untracked).splitlines() if p.strip()})
 
 
+def validate_body(body_file: str, run=None, which=None) -> tuple[int, list[str]]:
+    """Run the PR-body schema validator that the required PR Body check runs.
+
+    description_check only reads the prose for history claims, so a body whose
+    Test Plan is not inside a <details> block passed preflight and then failed
+    the required check after publication (PRs #780-#789, #793, #795). Exit 1
+    from the validator is a real rejection; anything else -- no node, a
+    timeout, an unexpected exit code -- is unchecked, which also fails.
+    """
+    run = run if run is not None else subprocess.run
+    which = which if which is not None else shutil.which
+    if not which("node"):
+        return 1, [f"description unchecked: node is not on PATH, so {VALIDATOR_NAME} did not run"]
+    if not os.path.isfile(os.path.join(REPO_ROOT, VALIDATOR)):
+        return 1, [f"description unchecked: no {VALIDATOR} under {REPO_ROOT}"]
+    cmd = ["node", VALIDATOR, "--body-file", os.path.abspath(body_file)]
+    try:
+        res = run(cmd, cwd=REPO_ROOT, capture_output=True, text=True, timeout=VALIDATOR_TIMEOUT)
+    except subprocess.TimeoutExpired:
+        return 1, [f"description unchecked: {VALIDATOR_NAME} did not finish in {VALIDATOR_TIMEOUT}s"]
+    except OSError as exc:
+        return 1, [f"description unchecked: cannot run {VALIDATOR_NAME}: {exc}"]
+    lines = (res.stdout + res.stderr).strip().splitlines()
+    if res.returncode in (0, 1):
+        return res.returncode, lines
+    return 1, lines + [f"description unchecked: {VALIDATOR_NAME} exited {res.returncode}"]
+
+
 def describe(body_file: str | None) -> int:
     if not body_file:
         print("fail    description unchecked: pass --body-file with the PR description")
@@ -239,7 +270,13 @@ def describe(body_file: str | None) -> int:
     for line in lines:
         print("        " + line)
     print(f"        description {outcome}")
-    return 0 if outcome == "clean" else 1
+    status = 0 if outcome == "clean" else 1
+
+    print(f"gate    node {VALIDATOR} --body-file " + body_file)
+    schema_status, schema_lines = validate_body(body_file)
+    for line in schema_lines:
+        print("        " + line)
+    return status or schema_status
 
 
 def main(argv: list[str] | None = None) -> int:
diff --git a/engine/skills/make-pr/tests/fixtures/pr795-failing-body.md b/engine/skills/make-pr/tests/fixtures/pr795-failing-body.md
new file mode 100644
index 000000000..f97a5d9ec
--- /dev/null
+++ b/engine/skills/make-pr/tests/fixtures/pr795-failing-body.md
@@ -0,0 +1,54 @@
+## Summary
+
+We run the same tools on seven machines. Keeping them all on the same version used to mean updating each one by hand.
+
+Now `scripts/update_fleet.sh` does it. It reads `remoteTargets` from `~/.invoker/config.json`, updates every machine, then prints one line per machine saying whether it worked.
+
+A machine it could not reach shows as failed, never as fine. A `--dry-run` changes nothing.
+
+## Review Claim
+
+The update script reports every machine it was asked about, and a machine it could not check shows as failed, not as fine.
+
+## Review Lane
+
+behavior
+
+## Review Unit
+
+corpus-lesson
+
+## Safety Invariant
+
+The script changes nothing on a `--dry-run`, and replacing the live app on that machine happens only behind an explicit flag. Release assets are checksum-verified before anything is installed.
+
+## Slice Rationale
+
+One claim, one review unit: the cat-mode skill package plus its colocated tests. The two mined rules are prose in the same skill; they ship no code and cannot be reviewed apart from the file they live in.
+
+## Non-goals
+
+- No Invoker-side change: the script consumes published release assets as they are.
+- No scheduler or worker. Running it is still a person's decision.
+
+## Test Plan
+
+```
+$ python3 -m unittest tests.test_cat_mode
+Ran 92 tests in 0.113s
+OK
+```
+
+Gates:
+
+```
+$ python3 scripts/ci/check_skill_file_refs.py    -> exit=0  ok      skill file refs
+$ python3 scripts/ci/check_codify_has_code.py    -> exit=0  ok      codify-has-code
+```
+
+## Revert Plan
+
+- Safe to revert? Yes
+- Revert command: `git revert <sha>`
+- Post-revert steps: None. No installed hook, skill, or machine reads the script.
+- Data migration? No
diff --git a/engine/skills/make-pr/tests/test_description_check.py b/engine/skills/make-pr/tests/test_description_check.py
index c0bbc8aec..60286071e 100644
--- a/engine/skills/make-pr/tests/test_description_check.py
+++ b/engine/skills/make-pr/tests/test_description_check.py
@@ -95,8 +95,13 @@ def test_unreadable_body_file_fails_as_unchecked(self):
         self.assertIn("cannot read", out.getvalue())
 
     def test_clean_body_file_passes(self):
+        """BODY is a stub, not a schema-valid description, so describe()'s other
+        step -- the PR-body schema validator -- stands in as passing here. Its
+        own cases live in test_preflight.TestDescriptionSchemaValidator."""
         original = description_check.check
+        original_validate = preflight.validate_body
         description_check.check = lambda body: ("clean", [])
+        preflight.validate_body = lambda body_file: (0, ["PR body validation passed."])
         try:
             with tempfile.NamedTemporaryFile("w", suffix=".md", delete=False) as handle:
                 handle.write(BODY)
@@ -105,6 +110,7 @@ def test_clean_body_file_passes(self):
                 status = preflight.describe(handle.name)
         finally:
             description_check.check = original
+            preflight.validate_body = original_validate
             os.unlink(handle.name)
         self.assertEqual(status, 0)
         self.assertIn("description clean", out.getvalue())
diff --git a/engine/skills/make-pr/tests/test_preflight.py b/engine/skills/make-pr/tests/test_preflight.py
index 22fa2d144..7744a9c0f 100644
--- a/engine/skills/make-pr/tests/test_preflight.py
+++ b/engine/skills/make-pr/tests/test_preflight.py
@@ -3,12 +3,15 @@
 sets of PRs in this repo (e.g. #89 visual-proof, the 2026-09-01 hook slices)."""
 from __future__ import annotations
 
+import contextlib
+import io
 import json
 import os
 import shutil
 import subprocess
 import sys
 import tempfile
+import types
 import unittest
 
 HERE = os.path.dirname(os.path.abspath(__file__))
@@ -16,6 +19,7 @@
 import preflight as pf  # noqa: E402
 
 SCRIPT = os.path.join(os.path.dirname(HERE), "scripts", "preflight.py")
+PR795_BODY = os.path.join(HERE, "fixtures", "pr795-failing-body.md")
 
 # real: PR #89 "Require actual captures for visual proof"
 PR89 = ["product/skills/visual-proof/SKILL.md", "product/skills/visual-proof/tests/fires_example.md"]
@@ -295,5 +299,166 @@ def test_no_paths_exits_2(self):
         self.assertEqual(res.returncode, 2)
 
 
+VALID_BODY = """## Summary
+
+We run the same tools on seven machines. Keeping them all on the same version used to mean updating each one by hand.
+
+Now one script does it. It updates every machine, then prints one line per machine saying whether it worked.
+
+A machine it could not reach shows as failed, never as fine. A practice run changes nothing.
+
+## Review Claim
+
+The update script reports every machine it was asked about, and a machine it could not check shows as failed, not as fine.
+
+## Review Lane
+
+behavior
+
+## Review Unit
+
+corpus-lesson
+
+## Safety Invariant
+
+The script changes nothing on a practice run, and replacing the live app on a machine happens only behind an explicit flag.
+
+## Slice Rationale
+
+One claim, one review unit: the cat-mode skill package plus its colocated tests.
+
+## Non-goals
+
+- No Invoker-side change: the script consumes published release assets as they are.
+
+## Test Plan
+
+<details>
+<summary>Test Plan</summary>
+
+```
+$ python3 -m unittest tests.test_cat_mode
+Ran 92 tests in 0.113s
+OK
+```
+
+</details>
+
+## Revert Plan
+
+<details>
+<summary>Revert Plan</summary>
+
+- Safe to revert? Yes
+- Revert command: `git revert <sha>`
+- Post-revert steps: None.
+- Data migration? No
+
+</details>
+"""
+
+
+@contextlib.contextmanager
+def clean_history_judge():
+    """Stand in for description_check so these tests exercise the schema
+    validator alone. The real checker asks a background LLM judge."""
+    stub = types.ModuleType("description_check")
+    stub.check = lambda body: ("clean", [])
+    real = sys.modules.get("description_check")
+    sys.modules["description_check"] = stub
+    try:
+        yield
+    finally:
+        if real is None:
+            sys.modules.pop("description_check", None)
+        else:
+            sys.modules["description_check"] = real
+
+
+def run_preflight(body_file):
+    """main() on a neutral path, so the exit status is the description's."""
+    real = pf.changed_paths
+    pf.changed_paths = lambda base, repo=pf.REPO_ROOT: ["docs/ecosystem.md"]
+    buf = io.StringIO()
+    try:
+        with clean_history_judge(), contextlib.redirect_stdout(buf):
+            status = pf.main(["--body-file", body_file])
+    finally:
+        pf.changed_paths = real
+    return status, buf.getvalue()
+
+
+class TestDescriptionSchemaValidator(unittest.TestCase):
+    """preflight's description step must run the same validator as the
+    required PR Body check. description_check only reads the prose for claims
+    about the repo's past, so a body whose Test Plan sits outside a <details>
+    block got `ok preflight passed` and was then rejected after publication."""
+
+    def test_the_pr795_body_is_rejected_with_the_validators_own_errors(self):
+        status, lines = pf.validate_body(PR795_BODY)
+        report = "\n".join(lines)
+        self.assertEqual(status, 1, report)
+        self.assertIn("## Test Plan must wrap its content in a collapsed <details> block", report)
+        self.assertIn("## Revert Plan must wrap its content in a collapsed <details> block", report)
+        self.assertIn("must not use code names", report)
+
+    def test_preflight_fails_on_the_pr795_body_and_prints_the_errors(self):
+        status, out = run_preflight(PR795_BODY)
+        self.assertEqual(status, 1, out)
+        self.assertIn("## Test Plan must wrap its content in a collapsed <details> block", out)
+        self.assertNotIn("ok      preflight passed", out)
+        self.assertIn("fail    preflight", out)
+
+    def test_a_body_the_validator_accepts_still_passes(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            body_file = os.path.join(tmp, "body.md")
+            with open(body_file, "w", encoding="utf-8") as handle:
+                handle.write(VALID_BODY)
+            status, out = run_preflight(body_file)
+        self.assertEqual(status, 0, out)
+        self.assertIn("PR body validation passed.", out)
+        self.assertIn("ok      preflight passed", out)
+
+    def test_a_missing_node_is_unchecked_not_a_pass(self):
+        status, lines = pf.validate_body(PR795_BODY, which=lambda name: None)
+        self.assertEqual(status, 1, lines)
+        self.assertIn("description unchecked", "\n".join(lines))
+        self.assertIn("node is not on PATH", "\n".join(lines))
+
+    def test_a_validator_that_is_not_in_the_checkout_is_unchecked_not_a_pass(self):
+        real = pf.VALIDATOR
+        pf.VALIDATOR = "engine/skills/draft-pr/scripts/no-such-validator.mjs"
+        try:
+            status, lines = pf.validate_body(PR795_BODY)
+        finally:
+            pf.VALIDATOR = real
+        self.assertEqual(status, 1, lines)
+        self.assertIn("description unchecked", "\n".join(lines))
+
+    def test_a_timeout_is_unchecked_not_a_pass(self):
+        def times_out(*args, **kwargs):
+            raise subprocess.TimeoutExpired(cmd="node", timeout=pf.VALIDATOR_TIMEOUT)
+
+        status, lines = pf.validate_body(PR795_BODY, run=times_out, which=lambda name: "/usr/bin/node")
+        self.assertEqual(status, 1, lines)
+        self.assertIn("description unchecked", "\n".join(lines))
+
+    def test_an_unexpected_exit_code_is_unchecked_not_a_pass(self):
+        def crashes(*args, **kwargs):
+            return subprocess.CompletedProcess(args, 7, stdout="", stderr="node: bad option")
+
+        status, lines = pf.validate_body(PR795_BODY, run=crashes, which=lambda name: "/usr/bin/node")
+        self.assertEqual(status, 1, lines)
+        self.assertIn("exited 7", "\n".join(lines))
+
+    def test_a_node_that_cannot_be_spawned_is_unchecked_not_a_pass(self):
+        def refuses(*args, **kwargs):
+            raise OSError("Exec format error")
+
+        status, lines = pf.validate_body(PR795_BODY, run=refuses, which=lambda name: "/usr/bin/node")
+        self.assertEqual(status, 1, lines)
+        self.assertIn("description unchecked", "\n".join(lines))
+
+
 if __name__ == "__main__":
     unittest.main()

From aeb2d1c931330fb5ec1b2bf1579e69e6a6d964af Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:58:53 +0800
Subject: [PATCH 02/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/implement-p?=
 =?UTF-8?q?reflight-runs-validator=20=E2=80=94=20Review=20claim:=20make-pr?=
 =?UTF-8?q?=20preflight=20runs=20the=20PR-description=20validator=20on=20-?=
 =?UTF-8?q?-body-file=20and=20fails=20when=20it=20fails,=20so=20it=20no=20?=
 =?UTF-8?q?longer=20prints=20'preflight=20passed'=20for=20a=20description?=
 =?UTF-8?q?=20the=20required=20PR=20Body=20check=20rejects.=20Review=20lan?=
 =?UTF-8?q?e:=20policy=20Safety=20invariant:=20The=20change=20can=20only?=
 =?UTF-8?q?=20add=20a=20failure=20or=20an=20UNCHECKED=20notice;=20a=20PR?=
 =?UTF-8?q?=20description=20that=20passes=20engine/skills/draft-pr/scripts?=
 =?UTF-8?q?/validate-pr-body.mjs=20today=20is=20never=20newly=20blocked.?=
 =?UTF-8?q?=20Effectiveness=20measurement:=20A=20preflight=20test=20feeds?=
 =?UTF-8?q?=20the=20original=20#795=20description=20(Test=20Plan=20and=20R?=
 =?UTF-8?q?evert=20Plan=20not=20inside=20<details>,=20code=20names=20in=20?=
 =?UTF-8?q?Summary),=20which=20the=20validator=20rejects=20with=20exit=201?=
 =?UTF-8?q?,=20and=20asserts=20preflight=20exits=20non-zero=20and=20prints?=
 =?UTF-8?q?=20the=20validator's=20errors;=20a=20passing=20description=20st?=
 =?UTF-8?q?ill=20gives=20'ok=20preflight=20passed';=20a=20validator=20that?=
 =?UTF-8?q?=20cannot=20run=20gives=20an=20unchecked=20failure,=20not=20a?=
 =?UTF-8?q?=20pass.=20Slice=20rationale:=20One=20script,=20one=20claim:=20?=
 =?UTF-8?q?preflight's=20description=20step=20also=20runs=20the=20schema?=
 =?UTF-8?q?=20validator.=20Architectural=20effect:=20preflight's=20describ?=
 =?UTF-8?q?e()=20step=20covers=20the=20same=20rules=20as=20the=20required?=
 =?UTF-8?q?=20PR=20Body=20check.=20Goal:=20Stop=20preflight=20from=20appro?=
 =?UTF-8?q?ving=20a=20PR=20description=20the=20required=20check=20will=20r?=
 =?UTF-8?q?eject.=20Motivation:=20A=20reflect=20pass=20found=20PR=20descri?=
 =?UTF-8?q?ptions=20on=20catstack=20PRs=20#780-#789,=20#793=20and=20#795?=
 =?UTF-8?q?=20failing=20the=20required=20PR=20Body=20check=20after=20publi?=
 =?UTF-8?q?cation,=20and=20the=20user=20had=20to=20ask=20for=20a=20manual?=
 =?UTF-8?q?=20fix=20of=20every=20PR.=20In=20catstack=20the=20guard=20staye?=
 =?UTF-8?q?d=20silent:=20fed=20`mergify=20stack=20push`=20and=20`gh=20pr?=
 =?UTF-8?q?=20create=20--body-file=20<failing=20body>`=20it=20exited=200?=
 =?UTF-8?q?=20with=20no=20output,=20while=20the=20same=20stack=20push=20in?=
 =?UTF-8?q?=20the=20Invoker=20checkout=20fired.=20Read=20at=20origin/main:?=
 =?UTF-8?q?=20engine/skills/make-pr/scripts/preflight.py=20describe()=20on?=
 =?UTF-8?q?ly=20runs=20description=5Fcheck,=20then=20main()=20prints=20'ok?=
 =?UTF-8?q?=20preflight=20passed';=20the=20session=20that=20opened=20#795?=
 =?UTF-8?q?=20got=20that=20line=20and=20then=20ran=20gh=20pr=20create.=20A?=
 =?UTF-8?q?lternative=20considerations:=20Leaving=20the=20validator=20to?=
 =?UTF-8?q?=20CI=20was=20set=20aside=20because=20the=20failure=20then=20sh?=
 =?UTF-8?q?ows=20up=20only=20after=20publication.=20Reimplementing=20the?=
 =?UTF-8?q?=20rules=20in=20Python=20was=20set=20aside=20because=20it=20wou?=
 =?UTF-8?q?ld=20drift=20from=20the=20validator.=20Implementation=20details?=
 =?UTF-8?q?:=20In=20engine/skills/make-pr/scripts/preflight.py=20describe(?=
 =?UTF-8?q?),=20after=20description=5Fcheck,=20run=20`node=20engine/skills?=
 =?UTF-8?q?/draft-pr/scripts/validate-pr-body.mjs=20--body-file=20<file>`?=
 =?UTF-8?q?=20from=20the=20repo=20root=20with=20a=20timeout.=20Print=20its?=
 =?UTF-8?q?=20output=20lines=20indented=20like=20the=20other=20gates.=20Ex?=
 =?UTF-8?q?it=201=20from=20the=20validator=20fails=20preflight;=20a=20miss?=
 =?UTF-8?q?ing=20node=20binary,=20a=20timeout,=20or=20any=20other=20exit?=
 =?UTF-8?q?=20code=20prints=20'description=20unchecked:=20<reason>'=20and?=
 =?UTF-8?q?=20fails=20preflight.=20Add=20the=20tests=20to=20engine/skills/?=
 =?UTF-8?q?make-pr/tests/test=5Fpreflight.py=20with=20the=20#795=20body=20?=
 =?UTF-8?q?as=20a=20fixture=20file=20under=20engine/skills/make-pr/tests/f?=
 =?UTF-8?q?ixtures/.=20Non-goals:=20No=20change=20to=20the=20validator,=20?=
 =?UTF-8?q?the=20hook,=20or=20the=20make-pr=20SKILL.md=20rules.=20Do=20not?=
 =?UTF-8?q?=20edit=20engine/skills/draft-pr/scripts/validate-pr-body.mjs,?=
 =?UTF-8?q?=20scripts/pr/validate-pr-body-local.mjs,=20or=20any=20other=20?=
 =?UTF-8?q?file=20open=20PR=20#742=20changes;=20call=20the=20validator=20a?=
 =?UTF-8?q?s=20it=20is.=20If=20a=20change=20would=20overlap=20an=20open=20?=
 =?UTF-8?q?PR,=20stop=20and=20report=20instead.=20Layer:=20domain=20Featur?=
 =?UTF-8?q?e=20state:=20active=20Files:=20-=20engine/skills/make-pr/script?=
 =?UTF-8?q?s/preflight.py=20-=20engine/skills/make-pr/tests/test=5Fpreflig?=
 =?UTF-8?q?ht.py=20-=20engine/skills/make-pr/tests/fixtures/pr795-failing-?=
 =?UTF-8?q?body.md=20Change=20types:=20-=20engine/skills/make-pr/scripts/p?=
 =?UTF-8?q?reflight.py:=20modify=20-=20engine/skills/make-pr/tests/test=5F?=
 =?UTF-8?q?preflight.py:=20modify=20-=20engine/skills/make-pr/tests/fixtur?=
 =?UTF-8?q?es/pr795-failing-body.md:=20create=20Acceptance=20criteria:=20-?=
 =?UTF-8?q?=20`python3=20-m=20unittest=20discover=20-s=20engine/skills/mak?=
 =?UTF-8?q?e-pr/tests=20-v`=20exits=200.=20-=20`python3=20scripts/check=5F?=
 =?UTF-8?q?skill=5Ftest=5Fcoverage.py=20--base=20origin/main=20--head=20HE?=
 =?UTF-8?q?AD`=20exits=200.=20-=20`python3=20scripts/check=5Fskill=5Ffile?=
 =?UTF-8?q?=5Frefs.py`=20exits=200.=20-=20`python3=20scripts/check=5Fno=5F?=
 =?UTF-8?q?new=5Fcomments.py`=20exits=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From e89618223402b2c1497635bfe6cd884299d7a7cc Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:59:12 +0800
Subject: [PATCH 03/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-1=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20-m=20unittest=20discover=20-s=20engine/skills/make-pr/tests?=
 =?UTF-8?q?=20-v`=20passes=20on=20the=20finished=20branch.=20Review=20lane?=
 =?UTF-8?q?:=20proof=20Safety=20invariant:=20Verification=20is=20read-only?=
 =?UTF-8?q?=20and=20alters=20no=20repository=20file.=20Effectiveness=20mea?=
 =?UTF-8?q?surement:=20The=20command's=20exit=20code=20is=20the=20direct?=
 =?UTF-8?q?=20measurement.=20Slice=20rationale:=20One=20check=20per=20proo?=
 =?UTF-8?q?f=20task.=20Architectural=20effect:=20None;=20verification=20on?=
 =?UTF-8?q?ly.=20Goal:=20Prove=20the=20slice.=20Motivation:=20Running=20th?=
 =?UTF-8?q?e=20check=20is=20the=20proof.=20Alternative=20considerations:?=
 =?UTF-8?q?=20The=20full=20suite=20was=20not=20required=20because=20the=20?=
 =?UTF-8?q?slice=20touches=20one=20component=20with=20its=20own=20tests.?=
 =?UTF-8?q?=20Implementation=20details:=20Run=20`python3=20-m=20unittest?=
 =?UTF-8?q?=20discover=20-s=20engine/skills/make-pr/tests=20-v`.=20Non-goa?=
 =?UTF-8?q?ls:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature=20s?=
 =?UTF-8?q?tate:=20active=20Layer=20exception:=20allowed.=20Verification?=
 =?UTF-8?q?=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20co?=
 =?UTF-8?q?mmit=20so=20they=20check=20the=20final=20branch=20the=20PR=20wi?=
 =?UTF-8?q?ll=20carry.=20Acceptance=20criteria:=20-=20The=20command=20exit?=
 =?UTF-8?q?s=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From fe1887b3effe7d00d47acbd00ae378a62893680f Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:59:13 +0800
Subject: [PATCH 04/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-4=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fno=5Fnew=5Fcomments.py`=20passes=20on=20the?=
 =?UTF-8?q?=20finished=20branch.=20Review=20lane:=20proof=20Safety=20invar?=
 =?UTF-8?q?iant:=20Verification=20is=20read-only=20and=20alters=20no=20rep?=
 =?UTF-8?q?ository=20file.=20Effectiveness=20measurement:=20The=20command'?=
 =?UTF-8?q?s=20exit=20code=20is=20the=20direct=20measurement.=20Slice=20ra?=
 =?UTF-8?q?tionale:=20One=20check=20per=20proof=20task.=20Architectural=20?=
 =?UTF-8?q?effect:=20None;=20verification=20only.=20Goal:=20Prove=20the=20?=
 =?UTF-8?q?slice.=20Motivation:=20Running=20the=20check=20is=20the=20proof?=
 =?UTF-8?q?.=20Alternative=20considerations:=20The=20full=20suite=20was=20?=
 =?UTF-8?q?not=20required=20because=20the=20slice=20touches=20one=20compon?=
 =?UTF-8?q?ent=20with=20its=20own=20tests.=20Implementation=20details:=20R?=
 =?UTF-8?q?un=20`python3=20scripts/check=5Fno=5Fnew=5Fcomments.py`.=20Non-?=
 =?UTF-8?q?goals:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature?=
 =?UTF-8?q?=20state:=20active=20Layer=20exception:=20allowed.=20Verificati?=
 =?UTF-8?q?on=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20?=
 =?UTF-8?q?commit=20so=20they=20check=20the=20final=20branch=20the=20PR=20?=
 =?UTF-8?q?will=20carry.=20Acceptance=20criteria:=20-=20The=20command=20ex?=
 =?UTF-8?q?its=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 1

From 394d32186d7f4efea35f3c7aae3abc98b7a4aef6 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:59:28 +0800
Subject: [PATCH 05/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-2=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fskill=5Ftest=5Fcoverage.py=20--base=20origin?=
 =?UTF-8?q?/main=20--head=20HEAD`=20passes=20on=20the=20finished=20branch.?=
 =?UTF-8?q?=20Review=20lane:=20proof=20Safety=20invariant:=20Verification?=
 =?UTF-8?q?=20is=20read-only=20and=20alters=20no=20repository=20file.=20Ef?=
 =?UTF-8?q?fectiveness=20measurement:=20The=20command's=20exit=20code=20is?=
 =?UTF-8?q?=20the=20direct=20measurement.=20Slice=20rationale:=20One=20che?=
 =?UTF-8?q?ck=20per=20proof=20task.=20Architectural=20effect:=20None;=20ve?=
 =?UTF-8?q?rification=20only.=20Goal:=20Prove=20the=20slice.=20Motivation:?=
 =?UTF-8?q?=20Running=20the=20check=20is=20the=20proof.=20Alternative=20co?=
 =?UTF-8?q?nsiderations:=20The=20full=20suite=20was=20not=20required=20bec?=
 =?UTF-8?q?ause=20the=20slice=20touches=20one=20component=20with=20its=20o?=
 =?UTF-8?q?wn=20tests.=20Implementation=20details:=20Run=20`python3=20scri?=
 =?UTF-8?q?pts/check=5Fskill=5Ftest=5Fcoverage.py=20--base=20origin/main?=
 =?UTF-8?q?=20--head=20HEAD`.=20Non-goals:=20No=20mutations.=20Layer:=20ap?=
 =?UTF-8?q?p=5Fregression=20Feature=20state:=20active=20Layer=20exception:?=
 =?UTF-8?q?=20allowed.=20Verification=20and=20the=20terminal=20scrub=20run?=
 =?UTF-8?q?=20after=20the=20docs=20commit=20so=20they=20check=20the=20fina?=
 =?UTF-8?q?l=20branch=20the=20PR=20will=20carry.=20Acceptance=20criteria:?=
 =?UTF-8?q?=20-=20The=20command=20exits=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From 5950111b5bcc17cdad2ddff472ed8ec0f8f5f494 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:59:31 +0800
Subject: [PATCH 06/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-3=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fskill=5Ffile=5Frefs.py`=20passes=20on=20the?=
 =?UTF-8?q?=20finished=20branch.=20Review=20lane:=20proof=20Safety=20invar?=
 =?UTF-8?q?iant:=20Verification=20is=20read-only=20and=20alters=20no=20rep?=
 =?UTF-8?q?ository=20file.=20Effectiveness=20measurement:=20The=20command'?=
 =?UTF-8?q?s=20exit=20code=20is=20the=20direct=20measurement.=20Slice=20ra?=
 =?UTF-8?q?tionale:=20One=20check=20per=20proof=20task.=20Architectural=20?=
 =?UTF-8?q?effect:=20None;=20verification=20only.=20Goal:=20Prove=20the=20?=
 =?UTF-8?q?slice.=20Motivation:=20Running=20the=20check=20is=20the=20proof?=
 =?UTF-8?q?.=20Alternative=20considerations:=20The=20full=20suite=20was=20?=
 =?UTF-8?q?not=20required=20because=20the=20slice=20touches=20one=20compon?=
 =?UTF-8?q?ent=20with=20its=20own=20tests.=20Implementation=20details:=20R?=
 =?UTF-8?q?un=20`python3=20scripts/check=5Fskill=5Ffile=5Frefs.py`.=20Non-?=
 =?UTF-8?q?goals:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature?=
 =?UTF-8?q?=20state:=20active=20Layer=20exception:=20allowed.=20Verificati?=
 =?UTF-8?q?on=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20?=
 =?UTF-8?q?commit=20so=20they=20check=20the=20final=20branch=20the=20PR=20?=
 =?UTF-8?q?will=20carry.=20Acceptance=20criteria:=20-=20The=20command=20ex?=
 =?UTF-8?q?its=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From e59f975f1c96f06b16439853b22beee3a5c974f0 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 00:59:40 +0800
Subject: [PATCH 07/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-4=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fno=5Fnew=5Fcomments.py`=20passes=20on=20the?=
 =?UTF-8?q?=20finished=20branch.=20Review=20lane:=20proof=20Safety=20invar?=
 =?UTF-8?q?iant:=20Verification=20is=20read-only=20and=20alters=20no=20rep?=
 =?UTF-8?q?ository=20file.=20Effectiveness=20measurement:=20The=20command'?=
 =?UTF-8?q?s=20exit=20code=20is=20the=20direct=20measurement.=20Slice=20ra?=
 =?UTF-8?q?tionale:=20One=20check=20per=20proof=20task.=20Architectural=20?=
 =?UTF-8?q?effect:=20None;=20verification=20only.=20Goal:=20Prove=20the=20?=
 =?UTF-8?q?slice.=20Motivation:=20Running=20the=20check=20is=20the=20proof?=
 =?UTF-8?q?.=20Alternative=20considerations:=20The=20full=20suite=20was=20?=
 =?UTF-8?q?not=20required=20because=20the=20slice=20touches=20one=20compon?=
 =?UTF-8?q?ent=20with=20its=20own=20tests.=20Implementation=20details:=20R?=
 =?UTF-8?q?un=20`python3=20scripts/check=5Fno=5Fnew=5Fcomments.py`.=20Non-?=
 =?UTF-8?q?goals:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature?=
 =?UTF-8?q?=20state:=20active=20Layer=20exception:=20allowed.=20Verificati?=
 =?UTF-8?q?on=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20?=
 =?UTF-8?q?commit=20so=20they=20check=20the=20final=20branch=20the=20PR=20?=
 =?UTF-8?q?will=20carry.=20Acceptance=20criteria:=20-=20The=20command=20ex?=
 =?UTF-8?q?its=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 1

From 0c52061d7f2ecd8a37fbac3e3baf0e393472e6a5 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 01:01:07 +0800
Subject: [PATCH 08/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-4=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fno=5Fnew=5Fcomments.py`=20passes=20on=20the?=
 =?UTF-8?q?=20finished=20branch.=20Review=20lane:=20proof=20Safety=20invar?=
 =?UTF-8?q?iant:=20Verification=20is=20read-only=20and=20alters=20no=20rep?=
 =?UTF-8?q?ository=20file.=20Effectiveness=20measurement:=20The=20command'?=
 =?UTF-8?q?s=20exit=20code=20is=20the=20direct=20measurement.=20Slice=20ra?=
 =?UTF-8?q?tionale:=20One=20check=20per=20proof=20task.=20Architectural=20?=
 =?UTF-8?q?effect:=20None;=20verification=20only.=20Goal:=20Prove=20the=20?=
 =?UTF-8?q?slice.=20Motivation:=20Running=20the=20check=20is=20the=20proof?=
 =?UTF-8?q?.=20Alternative=20considerations:=20The=20full=20suite=20was=20?=
 =?UTF-8?q?not=20required=20because=20the=20slice=20touches=20one=20compon?=
 =?UTF-8?q?ent=20with=20its=20own=20tests.=20Implementation=20details:=20R?=
 =?UTF-8?q?un=20`python3=20scripts/check=5Fno=5Fnew=5Fcomments.py`.=20Non-?=
 =?UTF-8?q?goals:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature?=
 =?UTF-8?q?=20state:=20active=20Layer=20exception:=20allowed.=20Verificati?=
 =?UTF-8?q?on=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20?=
 =?UTF-8?q?commit=20so=20they=20check=20the=20final=20branch=20the=20PR=20?=
 =?UTF-8?q?will=20carry.=20Acceptance=20criteria:=20-=20The=20command=20ex?=
 =?UTF-8?q?its=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Solution:
  Review claim: `python3 scripts/check_no_new_comments.py` passes on the finished branch.
Review lane: proof
Safety invariant: Verification is read-only and alters no repository file.
Effectiveness measurement: The command's exit code is the direct measurement.
Slice rationale: One check per proof task.
Architectural effect: None; verification only.
Goal: Prove the slice.
Motivation: Running the check is the proof.
Alternative considerations: The full suite was not required because the slice touches one component with its own tests.
Implementation details: Run `python3 scripts/check_no_new_comments.py`.
Non-goals: No mutations.
Layer: app_regression
Feature state: active
Layer exception: allowed. Verification and the terminal scrub run after the docs commit so they check the final branch the PR will carry.
Acceptance criteria:
- The command exits 0.
---
 scripts/ci/check_no_new_comments.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/scripts/ci/check_no_new_comments.py b/scripts/ci/check_no_new_comments.py
index 97210b61c..010231f9d 100644
--- a/scripts/ci/check_no_new_comments.py
+++ b/scripts/ci/check_no_new_comments.py
@@ -16,7 +16,7 @@
 import subprocess
 import sys
 
-SCRIPTS_DIR = os.path.dirname(os.path.abspath(__file__))
+SCRIPTS_DIR = os.path.dirname(os.path.realpath(__file__))
 REPO_ROOT = os.path.dirname(os.path.dirname(SCRIPTS_DIR))
 sys.path.insert(0, os.path.join(REPO_ROOT, "engine", "hooks", "no-comments"))
 sys.path.insert(0, SCRIPTS_DIR)

From 54e76b3fe572e414b2a511e31d2e5c2daec5222f Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 01:01:13 +0800
Subject: [PATCH 09/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/verify-pref?=
 =?UTF-8?q?light-runs-validator-4=20=E2=80=94=20Review=20claim:=20`python3?=
 =?UTF-8?q?=20scripts/check=5Fno=5Fnew=5Fcomments.py`=20passes=20on=20the?=
 =?UTF-8?q?=20finished=20branch.=20Review=20lane:=20proof=20Safety=20invar?=
 =?UTF-8?q?iant:=20Verification=20is=20read-only=20and=20alters=20no=20rep?=
 =?UTF-8?q?ository=20file.=20Effectiveness=20measurement:=20The=20command'?=
 =?UTF-8?q?s=20exit=20code=20is=20the=20direct=20measurement.=20Slice=20ra?=
 =?UTF-8?q?tionale:=20One=20check=20per=20proof=20task.=20Architectural=20?=
 =?UTF-8?q?effect:=20None;=20verification=20only.=20Goal:=20Prove=20the=20?=
 =?UTF-8?q?slice.=20Motivation:=20Running=20the=20check=20is=20the=20proof?=
 =?UTF-8?q?.=20Alternative=20considerations:=20The=20full=20suite=20was=20?=
 =?UTF-8?q?not=20required=20because=20the=20slice=20touches=20one=20compon?=
 =?UTF-8?q?ent=20with=20its=20own=20tests.=20Implementation=20details:=20R?=
 =?UTF-8?q?un=20`python3=20scripts/check=5Fno=5Fnew=5Fcomments.py`.=20Non-?=
 =?UTF-8?q?goals:=20No=20mutations.=20Layer:=20app=5Fregression=20Feature?=
 =?UTF-8?q?=20state:=20active=20Layer=20exception:=20allowed.=20Verificati?=
 =?UTF-8?q?on=20and=20the=20terminal=20scrub=20run=20after=20the=20docs=20?=
 =?UTF-8?q?commit=20so=20they=20check=20the=20final=20branch=20the=20PR=20?=
 =?UTF-8?q?will=20carry.=20Acceptance=20criteria:=20-=20The=20command=20ex?=
 =?UTF-8?q?its=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From cc4dd390202022a891ca379285c8ffff8cda7f5e Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 01:02:02 +0800
Subject: [PATCH 10/17] =?UTF-8?q?invoker:=20wf-1790182452439-6/scrub-hando?=
 =?UTF-8?q?ff-artifacts=20=E2=80=94=20Review=20claim:=20No=20ephemeral=20i?=
 =?UTF-8?q?nter-task=20handoff=20files=20remain=20in=20the=20worktree=20be?=
 =?UTF-8?q?fore=20the=20merge=20gate.=20Review=20lane:=20cleanup=20Safety?=
 =?UTF-8?q?=20invariant:=20The=20scrub=20script=20only=20checks=20for=20kn?=
 =?UTF-8?q?own=20handoff=20artifact=20names=20and=20never=20touches=20sour?=
 =?UTF-8?q?ce,=20tests,=20or=20other=20repository=20files.=20Effectiveness?=
 =?UTF-8?q?=20measurement:=20The=20script=20exits=20non-zero=20if=20any=20?=
 =?UTF-8?q?handoff=20artifact=20remains.=20Slice=20rationale:=20Required?=
 =?UTF-8?q?=20terminal=20scrub=20for=20every=20implementation=20workflow.?=
 =?UTF-8?q?=20Architectural=20effect:=20None;=20hygiene=20only.=20Goal:=20?=
 =?UTF-8?q?Leave=20the=20branch=20free=20of=20handoff=20artifacts.=20Motiv?=
 =?UTF-8?q?ation:=20Handoff=20files=20must=20not=20reach=20the=20PR.=20Alt?=
 =?UTF-8?q?ernative=20considerations:=20Manual=20cleanup=20was=20set=20asi?=
 =?UTF-8?q?de=20as=20non-deterministic.=20Implementation=20details:=20Run?=
 =?UTF-8?q?=20scripts/scrub-handoff-artifacts.sh.=20Non-goals:=20No=20prod?=
 =?UTF-8?q?uct=20edits.=20Layer:=20app=5Fregression=20Feature=20state:=20a?=
 =?UTF-8?q?ctive=20Layer=20exception:=20allowed.=20Verification=20and=20th?=
 =?UTF-8?q?e=20terminal=20scrub=20run=20after=20the=20docs=20commit=20so?=
 =?UTF-8?q?=20they=20check=20the=20final=20branch=20the=20PR=20will=20carr?=
 =?UTF-8?q?y.=20Acceptance=20criteria:=20-=20The=20command=20exits=200.?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0

From 13ab621032e05a184793ddf0f3012ebac5084eea Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Tue, 22 Sep 2026 00:59:32 -0700
Subject: [PATCH 11/17] reflect: catch an evidence-order correction from the
 ledger transition, not from wording

Both retraction detectors need a wrongness word. Two real corrections carry
none, because the claim was true and only its order was wrong:

  "Correcting one claim and arming the check I implied:"
  "I was right - but I said it a turn before I checked it"

Primary fix, no regex and no model: a ledger row going outstanding ->
discharged IS the event "a claim was asserted before its check ran, and the
check has now run". The Stop that discharges a row now reports it as a reflect
trigger. It reads state, so the reply's wording is not consulted at all.

Secondary fix: the judge dictionary's meaning now covers "right, but said
before it was checked", and both real strings are eval cases. Run against the
real judge, all six cases land where the dictionary says they should.

The regex scan keeps its hole on purpose, pinned by three tests so the next
author widens it knowingly rather than by accident.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Change-Id: I99190104b68d88f78aa9c8a706c9a6d23055aecc
---
 .../phrases/wrong-check-reflect.json          |  6 ++--
 engine/hooks/unverified-tag-ledger/README.md  |  6 ++++
 engine/hooks/unverified-tag-ledger/detect.py  | 35 ++++++++++++++-----
 .../unverified-tag-ledger/tests/test_hooks.py | 30 ++++++++++++++++
 .../wrong-check-reflect/eval_dictionary.py    |  3 ++
 .../tests/test_self_retraction_scan.py        | 24 +++++++++++++
 6 files changed, 93 insertions(+), 11 deletions(-)

diff --git a/engine/hooks/llm-judge/phrases/wrong-check-reflect.json b/engine/hooks/llm-judge/phrases/wrong-check-reflect.json
index 173dbcbef..cf121c913 100644
--- a/engine/hooks/llm-judge/phrases/wrong-check-reflect.json
+++ b/engine/hooks/llm-judge/phrases/wrong-check-reflect.json
@@ -1,7 +1,7 @@
 {
   "checker": "wrong-check-reflect",
   "reads": "reply",
-  "meaning": "The latest assistant reply admits that something it told the user earlier was wrong, misread, or answered the wrong question.",
+  "meaning": "The latest assistant reply admits that something it told the user earlier was wrong, misread, or answered the wrong question, OR that an earlier claim of its own went out before the check backing it had run. A reply saying the claim turned out correct but was stated before it was checked IS a match: the failure is the order, not the answer, so no word in the reply has to say anything was wrong.",
   "match": [
     "my earlier check was wrong",
     "You're right, I misread that",
@@ -10,7 +10,9 @@
     "I was wrong about the path",
     "Correction: the file I pointed you to earlier is not the one in use; the real one is src/b.py.",
     "Good catch. The earlier number was off; the real count is 12.",
-  "My earlier read of the config was wrong \u2014 the default is 4, not 8."
+    "My earlier read of the config was wrong \u2014 the default is 4, not 8.",
+    "Correcting one claim and arming the check I implied:",
+    "I was right - but I said it a turn before I checked it"
   ],
   "not_match": [
     "You're right. Let's go with option B.",
diff --git a/engine/hooks/unverified-tag-ledger/README.md b/engine/hooks/unverified-tag-ledger/README.md
index ccfa1d8cf..121cd94ad 100644
--- a/engine/hooks/unverified-tag-ledger/README.md
+++ b/engine/hooks/unverified-tag-ledger/README.md
@@ -53,6 +53,12 @@ fixtures in `tests/test_hooks.py`.
   reason is written to stderr.
 - **Escalation** — a claim outstanding `ESCALATE_AFTER_TURNS` (3) turns or more
   is reported as a reflect trigger rather than accumulating quietly.
+- **Discharge is itself a reflect trigger** — a row going outstanding ->
+  discharged is the record of a claim that went out first and was checked
+  after. That is an evidence-order miss, and it carries no wrongness word, so
+  the phrase scanners (`engine/skills/reflect/scripts/self_retraction_scan.py`,
+  and the `wrong-check-reflect` dictionary) cannot see it from the text. This
+  hook sees it from state instead, and says so on the Stop that discharges.
 
 Malformed tags are deliberately ignored here; `diu-stop` already rejects those.
 
diff --git a/engine/hooks/unverified-tag-ledger/detect.py b/engine/hooks/unverified-tag-ledger/detect.py
index 8a47c2eff..1f54902da 100644
--- a/engine/hooks/unverified-tag-ledger/detect.py
+++ b/engine/hooks/unverified-tag-ledger/detect.py
@@ -34,6 +34,15 @@
 ESCALATE_AFTER_TURNS = 3
 MAX_LISTED = 5
 
+DISCHARGE_REFLECT = (
+    "unverified-tag-ledger: {count} claim(s) went from unverified to checked this turn: "
+    "{claims}. That transition is the whole event: the claim went out first and the check "
+    "ran after. No wording has to admit anything for this to be true, which is why the "
+    "phrase scanners miss it -- an evidence-order miss carries no wrongness word. "
+    "Treat it as a reflect trigger, not a milestone: run reflect on this transcript, or "
+    "say plainly why this one does not need it."
+)
+
 CLAIM_RE = re.compile(
     r"\{\{\s*CAT-UNVERIFIED\s*:?\s*(?P<claim>.*?)(?:--|—)\s*cannot\s+verify\s*:\s*(?P<reason>[^}]*)\}\}",
     re.IGNORECASE | re.DOTALL,
@@ -177,18 +186,26 @@ def evaluate(payload: dict) -> dict:
     session_id = str(payload.get("session_id") or "")
     message = _last_assistant_text(payload)
     tools = tools_used_this_turn(payload)
+    was_open = {row["claim"] for row in outstanding(read_ledger(session_id))}
     rows = record_turn(session_id, message, tools)
+    notes = []
+    discharged = sorted(
+        row["claim"] for row in rows if row.get("resolved") and row["claim"] in was_open)
+    if discharged:
+        notes.append(DISCHARGE_REFLECT.format(
+            count=len(discharged), claims="; ".join(discharged[:MAX_LISTED])))
 
     new_claims = {tag["claim"] for tag in parse_tags(message)}
     if not new_claims:
-        return {"note": "", "block": ""}
+        return {"note": "\n".join(notes), "block": ""}
 
     if tools is None:
-        return {"note": (
+        notes.append(
             f"unverified-tag-ledger: logged {len(new_claims)} CAT-UNVERIFIED claim(s), but this "
             "turn's tool calls could not be read from transcript_path (see the line above), so "
             "whether a check was attempted is UNCHECKED, not clean. Nothing was discharged and "
-            "the turn was not refused."), "block": ""}
+            "the turn was not refused.")
+        return {"note": "\n".join(notes), "block": ""}
 
     if not tools & VERIFY_TOOLS and not payload.get("stop_hook_active"):
         claims = "; ".join(sorted(new_claims)[:MAX_LISTED])
@@ -201,12 +218,12 @@ def evaluate(payload: dict) -> dict:
 
     fresh = [row for row in rows
              if row["claim"] in new_claims and not row.get("resolved") and row.get("turns", 0) == 0]
-    if not fresh:
-        return {"note": "", "block": ""}
-    return {"note": (
-        f"unverified-tag-ledger: logged {len(fresh)} CAT-UNVERIFIED claim(s) against this session. "
-        "They are deferred, not discharged, and will be raised again next turn "
-        "(cat-mode/SKILL.md:269)."), "block": ""}
+    if fresh:
+        notes.append(
+            f"unverified-tag-ledger: logged {len(fresh)} CAT-UNVERIFIED claim(s) against this "
+            "session. They are deferred, not discharged, and will be raised again next turn "
+            "(cat-mode/SKILL.md:269).")
+    return {"note": "\n".join(notes), "block": ""}
 
 
 def decide_stop(payload: dict) -> str:
diff --git a/engine/hooks/unverified-tag-ledger/tests/test_hooks.py b/engine/hooks/unverified-tag-ledger/tests/test_hooks.py
index 12ab99310..251cede62 100644
--- a/engine/hooks/unverified-tag-ledger/tests/test_hooks.py
+++ b/engine/hooks/unverified-tag-ledger/tests/test_hooks.py
@@ -38,6 +38,9 @@
     "-- cannot verify: my own reasoning isn't observable by any command}}")
 MALFORMED = "{{CAT-UNVERIFIED: something I did not check}}"
 
+EVIDENCE_ORDER_1 = "Correcting one claim and arming the check I implied:"
+EVIDENCE_ORDER_2 = "I was right - but I said it a turn before I checked it"
+
 
 def _real_transcript_lines() -> list[str]:
     with open(REAL_TRANSCRIPT, encoding="utf-8") as handle:
@@ -209,6 +212,33 @@ def test_verified_and_dropped_tag_is_discharged_through_the_real_payload(self) -
         self.assertEqual(self.detect.outstanding(self.detect.read_ledger("s1")), [])
         self.assertEqual(self.detect.reminder("s1"), "")
 
+    def test_a_discharged_claim_fires_the_reflect_trigger(self) -> None:
+        self.detect.evaluate(self.payload(REAL_TAG_1, tools=True))
+        verdict = self.detect.evaluate(
+            self.payload("Here is the pasted output proving it.", tools=True))
+        self.assertIn("reflect trigger", verdict["note"])
+        self.assertIn("widened scope", verdict["note"])
+
+    def test_an_evidence_order_correction_triggers_with_no_wrongness_word(self) -> None:
+        """The transition fires; the reply's wording is not consulted at all."""
+        for index, reply in enumerate((EVIDENCE_ORDER_1, EVIDENCE_ORDER_2)):
+            session = f"evidence-order-{index}"
+            self.detect.evaluate(self.payload(REAL_TAG_1, tools=True, session_id=session))
+            verdict = self.detect.evaluate(
+                self.payload(reply, tools=True, session_id=session))
+            self.assertIn("reflect trigger", verdict["note"])
+            self.assertIn("carries no wrongness word", verdict["note"])
+
+    def test_a_turn_that_discharges_nothing_stays_silent_about_reflect(self) -> None:
+        verdict = self.detect.evaluate(
+            self.payload("Ran the tests, all green.", tools=True))
+        self.assertEqual(verdict["note"], "")
+
+    def test_a_reemitted_tag_is_not_reported_as_discharged(self) -> None:
+        self.detect.evaluate(self.payload(REAL_TAG_1, tools=True))
+        verdict = self.detect.evaluate(self.payload(REAL_TAG_1, tools=True))
+        self.assertNotIn("reflect trigger", verdict["note"])
+
     def test_unchecked_turn_does_not_discharge_a_row(self) -> None:
         self.detect.evaluate(self.payload(REAL_TAG_1, tools=True))
         self.detect.evaluate({
diff --git a/engine/hooks/wrong-check-reflect/eval_dictionary.py b/engine/hooks/wrong-check-reflect/eval_dictionary.py
index 258dabb2f..9a9e89273 100644
--- a/engine/hooks/wrong-check-reflect/eval_dictionary.py
+++ b/engine/hooks/wrong-check-reflect/eval_dictionary.py
@@ -17,6 +17,9 @@
     (HIT_TEXT, True),
     ("You're right. Let's go with option B.", False),
     ("I double-checked my earlier count and it holds; nothing in it was wrong.", False),
+    ("Correcting one claim and arming the check I implied:", True),
+    ("I was right - but I said it a turn before I checked it", True),
+    ("I ran the check first and then said it, so the order was right.", False),
 )
 
 
diff --git a/engine/skills/reflect/scripts/tests/test_self_retraction_scan.py b/engine/skills/reflect/scripts/tests/test_self_retraction_scan.py
index 72266f872..852582772 100644
--- a/engine/skills/reflect/scripts/tests/test_self_retraction_scan.py
+++ b/engine/skills/reflect/scripts/tests/test_self_retraction_scan.py
@@ -74,6 +74,30 @@ def test_third_party_blame_inside_a_reported_clause_stays_clean(self):
         self.assertIsNone(self_retraction_scan.find_admission(text))
 
 
+class TestEvidenceOrderIsOutOfReach(unittest.TestCase):
+    """Two real corrections this scan cannot see, and the reason it cannot.
+
+    Both are corrections about evidence ORDER: the claim was true, and it was
+    asserted before the check ran. Nothing in either sentence says anything was
+    wrong, so every pattern here misses them by construction. Pinned so the
+    next author widens the regex knowingly rather than by accident: the catch
+    for this class is the unverified-tag-ledger discharge transition, which
+    reads state rather than wording.
+    """
+
+    def test_arming_the_implied_check_is_not_reachable_by_wording(self):
+        text = "Correcting one claim and arming the check I implied:"
+        self.assertIsNone(self_retraction_scan.find_admission(text))
+
+    def test_right_but_asserted_early_is_not_reachable_by_wording(self):
+        text = "I was right - but I said it a turn before I checked it"
+        self.assertIsNone(self_retraction_scan.find_admission(text))
+
+    def test_the_same_sentence_with_a_wrongness_word_does_fire(self):
+        text = "I was wrong about the path; I said it a turn before I checked it."
+        self.assertIsNotNone(self_retraction_scan.find_admission(text))
+
+
 class TestScanAssistantTexts(unittest.TestCase):
     def test_collects_one_hit_per_admission(self):
         hits = self_retraction_scan.scan_assistant_texts(

From 0ddfd2984d6512cfe70e70c0e37180997bf1f744 Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 11:44:36 +0800
Subject: [PATCH 12/17] cat-mode: a park beside a killed copy wins over the
 partial Invoker.app

A SIGKILL during cp -R cannot be trapped and leaves a partial Invoker.app
next to the parked original. parked_app ignored the park whenever
Invoker.app existed, so the next run parked the partial as its backup and a
failed retry put the partial back. The park is now always adopted first, and
--dry-run warns about it instead of reporting ok.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Change-Id: Ib73a4c7561d453162d30c5806e5fe063d5021e2c
---
 .../skills/cat-mode/scripts/update_fleet.sh   |  3 +-
 tests/test_cat_mode.py                        | 34 +++++++++++++++++--
 2 files changed, 33 insertions(+), 4 deletions(-)

diff --git a/corpus/skills/cat-mode/scripts/update_fleet.sh b/corpus/skills/cat-mode/scripts/update_fleet.sh
index 732359f43..9dbb1b0f2 100755
--- a/corpus/skills/cat-mode/scripts/update_fleet.sh
+++ b/corpus/skills/cat-mode/scripts/update_fleet.sh
@@ -169,7 +169,6 @@ app_version() {
 
 parked_app() {
   local candidate
-  [ -d "$APP_DIR/Invoker.app" ] && return 0
   for candidate in "$APP_DIR"/Invoker.app.replacing.*; do
     [ -d "$candidate" ] || continue
     printf '%s' "$candidate"
@@ -192,7 +191,7 @@ local_app() {
   parked="$(parked_app)"
   if [ "$DRY_RUN" = 1 ]; then
     if [ -n "$parked" ]; then
-      row warn local "app: an interrupted run left $parked and no $APP_DIR/Invoker.app; a real run puts it back first (dry-run)" "$parked"
+      row warn local "app: an interrupted run left $parked; a real run puts it back over any $APP_DIR/Invoker.app first (dry-run)" "$parked"
       return 0
     fi
     row ok local "app $(app_version) -> $RELEASE_VERSION (dry-run)" ""; return 0
diff --git a/tests/test_cat_mode.py b/tests/test_cat_mode.py
index 8eca54c7b..4ca53c1ab 100644
--- a/tests/test_cat_mode.py
+++ b/tests/test_cat_mode.py
@@ -725,7 +725,7 @@ def test_script_carries_no_comments(self):
 APP_DIR="$TEST_APP_DIR"
 WORK_DIR="$TEST_WORK_DIR"
 RELEASE_VERSION="9.9.9"
-DRY_RUN=0
+DRY_RUN="${{TEST_DRY_RUN:-0}}"
 row() {{ printf '%s\\t%s\\t%s\\n' "$1" "$2" "$3" >> "$TEST_ROWS"; return 0; }}
 fetch_asset() {{ printf '%s' "$WORK_DIR/Invoker.dmg"; }}
 {functions}
@@ -754,7 +754,7 @@ def test_script_carries_no_comments(self):
 FAILING_CP = "exit 1\n"
 
 
-def run_local_app(script_path, tmp, dmg_version="9.9.9", cp_fails=False):
+def run_local_app(script_path, tmp, dmg_version="9.9.9", cp_fails=False, dry_run=False):
     """Run update_fleet.sh's app-replace step alone, against a fake /Applications.
 
     The local-Mac section is sliced out of the real script and sourced into a
@@ -800,6 +800,7 @@ def run_local_app(script_path, tmp, dmg_version="9.9.9", cp_fails=False):
         TEST_WORK_DIR=work_dir,
         TEST_ROWS=rows,
         TEST_DMG_VERSION=dmg_version,
+        TEST_DRY_RUN="1" if dry_run else "0",
     )
     out = subprocess.run(["bash", harness], capture_output=True, text=True, env=env)
     with open(rows, encoding="utf-8") as handle:
@@ -887,6 +888,35 @@ def test_a_good_replace_reports_ok_and_keeps_the_previous_bundle(self):
         leftovers = [n for n in os.listdir(app_dir) if ".replacing." in n]
         self.assertEqual(leftovers, [], f"parked bundle left behind: {leftovers}")
 
+    def test_a_park_beside_a_killed_copy_is_put_back_before_the_retry(self):
+        """A SIGKILL mid `cp -R` cannot be trapped. It leaves a partial
+        Invoker.app beside the parked original. The next run has to trust the
+        park, not the partial, or a failed retry restores the partial and the
+        good bundle is lost."""
+        app_dir = os.path.join(self.tmp, "Applications")
+        os.makedirs(app_dir, exist_ok=True)
+        write_bundle(app_dir, "Invoker.app", "partial")
+        write_bundle(app_dir, "Invoker.app.replacing.4242", "1.0.0")
+
+        app_dir, row, out = run_local_app(self.SCRIPT, self.tmp, cp_fails=True)
+
+        self.assertTrue(row.startswith("fail\t"), f"row was {row!r}\n{out.stderr}")
+        self.assertEqual(bundle_version(app_dir, "Invoker.app"), "1.0.0", row)
+        leftovers = [n for n in os.listdir(app_dir) if ".replacing." in n]
+        self.assertEqual(leftovers, [], f"parked bundle left behind: {leftovers}")
+
+    def test_dry_run_warns_on_a_park_beside_a_killed_copy(self):
+        app_dir = os.path.join(self.tmp, "Applications")
+        os.makedirs(app_dir, exist_ok=True)
+        write_bundle(app_dir, "Invoker.app", "partial")
+        write_bundle(app_dir, "Invoker.app.replacing.4242", "1.0.0")
+
+        app_dir, row, out = run_local_app(self.SCRIPT, self.tmp, dry_run=True)
+
+        self.assertTrue(row.startswith("warn\t"), f"row was {row!r}\n{out.stderr}")
+        self.assertIn("Invoker.app.replacing.4242", row)
+        self.assertEqual(bundle_version(app_dir, "Invoker.app.replacing.4242"), "1.0.0", row)
+
 
 CATSTACK_FUNCTIONS = re.compile(r"^write_payloads\(\) \{.*?(?=^write_payloads$)", re.S | re.M)
 

From 1c598ed94e91a09e71c6d260d3d25b56d5be232d Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbertchantech@gmail.com>
Date: Thu, 24 Sep 2026 12:07:55 +0800
Subject: [PATCH 13/17] make-pr: preflight hands the validator the changed
 files

The required PR Body check passes --changed-files-file, which is how the
validator rejects changed file and folder names in the Summary and a
Review Unit that does not match the diff. Preflight passed only
--body-file, so those descriptions passed locally and failed after
publish. describe() now takes the changed paths preflight already
computed and writes them to a temp file for the validator.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Change-Id: Id36a64869b7cfb2b76b912fd3cf81a88a39dd344
---
 engine/skills/make-pr/scripts/preflight.py    | 29 +++++++++++--------
 .../make-pr/tests/test_description_check.py   |  8 ++---
 engine/skills/make-pr/tests/test_preflight.py | 26 +++++++++++++++++
 3 files changed, 47 insertions(+), 16 deletions(-)

diff --git a/engine/skills/make-pr/scripts/preflight.py b/engine/skills/make-pr/scripts/preflight.py
index e3c237048..91b0fd0f6 100644
--- a/engine/skills/make-pr/scripts/preflight.py
+++ b/engine/skills/make-pr/scripts/preflight.py
@@ -28,6 +28,7 @@
 import shutil
 import subprocess
 import sys
+import tempfile
 
 sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
 
@@ -225,7 +226,7 @@ def changed_paths(base: str, repo: str = REPO_ROOT) -> list[str]:
     return sorted({p for p in (out + untracked).splitlines() if p.strip()})
 
 
-def validate_body(body_file: str, run=None, which=None) -> tuple[int, list[str]]:
+def validate_body(body_file: str, run=None, which=None, changed_paths: list[str] | None = None) -> tuple[int, list[str]]:
     """Run the PR-body schema validator that the required PR Body check runs.
 
     description_check only reads the prose for history claims, so a body whose
@@ -240,20 +241,24 @@ def validate_body(body_file: str, run=None, which=None) -> tuple[int, list[str]]
         return 1, [f"description unchecked: node is not on PATH, so {VALIDATOR_NAME} did not run"]
     if not os.path.isfile(os.path.join(REPO_ROOT, VALIDATOR)):
         return 1, [f"description unchecked: no {VALIDATOR} under {REPO_ROOT}"]
-    cmd = ["node", VALIDATOR, "--body-file", os.path.abspath(body_file)]
-    try:
-        res = run(cmd, cwd=REPO_ROOT, capture_output=True, text=True, timeout=VALIDATOR_TIMEOUT)
-    except subprocess.TimeoutExpired:
-        return 1, [f"description unchecked: {VALIDATOR_NAME} did not finish in {VALIDATOR_TIMEOUT}s"]
-    except OSError as exc:
-        return 1, [f"description unchecked: cannot run {VALIDATOR_NAME}: {exc}"]
+    with tempfile.TemporaryDirectory(prefix="preflight-") as tmp:
+        files_file = os.path.join(tmp, "changed-files.txt")
+        with open(files_file, "w", encoding="utf-8") as handle:
+            handle.writelines(p + "\n" for p in changed_paths or [])
+        cmd = ["node", VALIDATOR, "--body-file", os.path.abspath(body_file), "--changed-files-file", files_file]
+        try:
+            res = run(cmd, cwd=REPO_ROOT, capture_output=True, text=True, timeout=VALIDATOR_TIMEOUT)
+        except subprocess.TimeoutExpired:
+            return 1, [f"description unchecked: {VALIDATOR_NAME} did not finish in {VALIDATOR_TIMEOUT}s"]
+        except OSError as exc:
+            return 1, [f"description unchecked: cannot run {VALIDATOR_NAME}: {exc}"]
     lines = (res.stdout + res.stderr).strip().splitlines()
     if res.returncode in (0, 1):
         return res.returncode, lines
     return 1, lines + [f"description unchecked: {VALIDATOR_NAME} exited {res.returncode}"]
 
 
-def describe(body_file: str | None) -> int:
+def describe(body_file: str | None, changed_paths: list[str]) -> int:
     if not body_file:
         print("fail    description unchecked: pass --body-file with the PR description")
         return 1
@@ -272,8 +277,8 @@ def describe(body_file: str | None) -> int:
     print(f"        description {outcome}")
     status = 0 if outcome == "clean" else 1
 
-    print(f"gate    node {VALIDATOR} --body-file " + body_file)
-    schema_status, schema_lines = validate_body(body_file)
+    print(f"gate    node {VALIDATOR} --body-file {body_file} --changed-files-file <{len(changed_paths)} changed path(s)>")
+    schema_status, schema_lines = validate_body(body_file, changed_paths=changed_paths)
     for line in schema_lines:
         print("        " + line)
     return status or schema_status
@@ -338,7 +343,7 @@ def main(argv: list[str] | None = None) -> int:
         if res.returncode != 0:
             status = 1
     if args.paths is None and not args.dry_run:
-        if describe(args.body_file) != 0:
+        if describe(args.body_file, paths) != 0:
             status = 1
     print("ok      preflight passed" if status == 0 else "fail    preflight: fix the above before gh pr create")
     return status
diff --git a/engine/skills/make-pr/tests/test_description_check.py b/engine/skills/make-pr/tests/test_description_check.py
index 60286071e..1058f206b 100644
--- a/engine/skills/make-pr/tests/test_description_check.py
+++ b/engine/skills/make-pr/tests/test_description_check.py
@@ -83,14 +83,14 @@ class TestPreflightDescription(unittest.TestCase):
     def test_missing_body_file_fails_as_unchecked(self):
         out = io.StringIO()
         with redirect_stdout(out):
-            status = preflight.describe(None)
+            status = preflight.describe(None, [])
         self.assertEqual(status, 1)
         self.assertIn("description unchecked", out.getvalue())
 
     def test_unreadable_body_file_fails_as_unchecked(self):
         out = io.StringIO()
         with redirect_stdout(out):
-            status = preflight.describe("/nonexistent/body.md")
+            status = preflight.describe("/nonexistent/body.md", [])
         self.assertEqual(status, 1)
         self.assertIn("cannot read", out.getvalue())
 
@@ -101,13 +101,13 @@ def test_clean_body_file_passes(self):
         original = description_check.check
         original_validate = preflight.validate_body
         description_check.check = lambda body: ("clean", [])
-        preflight.validate_body = lambda body_file: (0, ["PR body validation passed."])
+        preflight.validate_body = lambda body_file, changed_paths: (0, ["PR body validation passed."])
         try:
             with tempfile.NamedTemporaryFile("w", suffix=".md", delete=False) as handle:
                 handle.write(BODY)
             out = io.StringIO()
             with redirect_stdout(out):
-                status = preflight.describe(handle.name)
+                status = preflight.describe(handle.name, [])
         finally:
             description_check.check = original
             preflight.validate_body = original_validate
diff --git a/engine/skills/make-pr/tests/test_preflight.py b/engine/skills/make-pr/tests/test_preflight.py
index 7744a9c0f..928595e45 100644
--- a/engine/skills/make-pr/tests/test_preflight.py
+++ b/engine/skills/make-pr/tests/test_preflight.py
@@ -419,6 +419,32 @@ def test_a_body_the_validator_accepts_still_passes(self):
         self.assertIn("PR body validation passed.", out)
         self.assertIn("ok      preflight passed", out)
 
+    def test_a_changed_file_name_in_the_summary_fails_like_the_required_check(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            body_file = os.path.join(tmp, "body.md")
+            with open(body_file, "w", encoding="utf-8") as handle:
+                handle.write(VALID_BODY.replace("Now one script does it.", "Now the ecosystem page does it."))
+            status, out = run_preflight(body_file)
+        self.assertEqual(status, 1, out)
+        self.assertIn('"ecosystem" (changed file name)', out)
+        self.assertNotIn("ok      preflight passed", out)
+
+    def test_the_validator_is_handed_the_changed_files(self):
+        seen = {}
+
+        def records(cmd, **kwargs):
+            flag = cmd.index("--changed-files-file")
+            with open(cmd[flag + 1], encoding="utf-8") as handle:
+                seen["files"] = handle.read().split()
+            return subprocess.CompletedProcess(cmd, 0, stdout="PR body validation passed.", stderr="")
+
+        status, lines = pf.validate_body(
+            PR795_BODY, run=records, which=lambda name: "/usr/bin/node",
+            changed_paths=["a/b.py", "docs/ecosystem.md"],
+        )
+        self.assertEqual(status, 0, lines)
+        self.assertEqual(seen["files"], ["a/b.py", "docs/ecosystem.md"])
+
     def test_a_missing_node_is_unchecked_not_a_pass(self):
         status, lines = pf.validate_body(PR795_BODY, which=lambda name: None)
         self.assertEqual(status, 1, lines)

From ccddeb197c58a27e30fcad052a9fec3b1c5ec26f Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Thu, 24 Sep 2026 12:08:27 +0800
Subject: [PATCH 14/17] event-wait: read a final unterminated line, kill a
 stuck source, honor the deadline on write

Three review findings left open after #790 and #821:

- A lines source whose last complete record has no trailing newline is
  matched at EOF; a partial last line still ends as truncated_frame.
- A json_stream command that ignores SIGTERM is killed after 5s instead
  of being left running.
- Socket connect and the snapshot write use the wait's remaining time, so
  a peer that never reads ends the wait at its deadline.

Also carries #791's harness EOF test for drop_peers.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Change-Id: I42c99ca13c82dc0fd5c22bc3b56a991ddd38c77b
---
 .../skills/event-wait/scripts/wait_event.py   | 44 ++++++++--
 .../event-wait/tests/test_wait_event.py       | 86 ++++++++++++++++++-
 2 files changed, 121 insertions(+), 9 deletions(-)

diff --git a/product/skills/event-wait/scripts/wait_event.py b/product/skills/event-wait/scripts/wait_event.py
index f7dae8b2c..3d1550f60 100644
--- a/product/skills/event-wait/scripts/wait_event.py
+++ b/product/skills/event-wait/scripts/wait_event.py
@@ -455,6 +455,19 @@ def feed(self, chunk: bytes) -> Iterator[dict]:
             return self._feed_lines()
         return self._feed_length_prefix()
 
+    def finish(self) -> Iterator[dict]:
+        if self.framing["kind"] != "lines" or not self.buffer.strip():
+            return
+        tail = self.buffer
+        try:
+            record = self._decode_payload(tail)
+        except SourceError as exc:
+            raise SourceError(
+                "truncated_frame", f"source closed with {len(tail)} unread bytes"
+            ) from exc
+        self.buffer = b""
+        yield record
+
     def _decode_payload(self, payload: bytes) -> dict:
         try:
             record = json.loads(payload.decode("utf-8"))
@@ -507,10 +520,14 @@ def _feed_length_prefix(self) -> Iterator[dict]:
 class SocketChannel:
     """A connected Unix socket. Writable only when a snapshot is configured."""
 
-    def __init__(self, path: str) -> None:
+    def __init__(self, path: str, timeout: float) -> None:
         self.sock = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+        self.sock.settimeout(timeout)
         try:
             self.sock.connect(path)
+        except socket.timeout:
+            self.sock.close()
+            raise Timeout() from None
         except OSError as exc:
             self.sock.close()
             raise SourceError(
@@ -528,9 +545,12 @@ def recv(self, timeout: float) -> bytes:
                 "read_failed", f"socket read failed ({errno.errorcode.get(exc.errno, 'error')})"
             ) from exc
 
-    def send(self, payload: bytes) -> None:
+    def send(self, payload: bytes, timeout: float) -> None:
+        self.sock.settimeout(timeout)
         try:
             self.sock.sendall(payload)
+        except socket.timeout:
+            raise Timeout() from None
         except OSError as exc:
             raise SourceError(
                 "write_failed", f"snapshot request failed ({errno.errorcode.get(exc.errno, 'error')})"
@@ -579,7 +599,7 @@ def recv(self, timeout: float) -> bytes:
                 "read_failed", f"stream read failed ({errno.errorcode.get(exc.errno, 'error')})"
             ) from exc
 
-    def send(self, payload: bytes) -> None:
+    def send(self, payload: bytes, timeout: float) -> None:
         raise SourceError("read_only_source", "this source accepts no requests")
 
     def close(self) -> None:
@@ -587,7 +607,11 @@ def close(self) -> None:
             try:
                 self.process.terminate()
                 self.process.wait(timeout=5)
-            except (OSError, subprocess.TimeoutExpired) as exc:
+            except subprocess.TimeoutExpired:
+                print("event-wait: source command ignored terminate; killing it", file=sys.stderr)
+                self.process.kill()
+                self.process.wait()
+            except OSError as exc:
                 print(f"event-wait: stopping source command failed: {exc}", file=sys.stderr)
             finally:
                 if self.process.stdout is not None:
@@ -758,7 +782,7 @@ def wake_env(self, status: str | None = None) -> dict:
     def attach(self) -> None:
         source = self.spec["source"]
         if source["kind"] == "unix_socket":
-            self.channel = SocketChannel(source["path"])
+            self.channel = SocketChannel(source["path"], self.remaining())
         else:
             self.channel = StreamChannel(source["command"], source["path"])
 
@@ -779,7 +803,7 @@ def send_snapshot(self) -> None:
             frame = len(payload).to_bytes(framing["prefix_bytes"], framing["byte_order"]) + payload
         else:
             frame = payload + b"\n"
-        self.channel.send(frame)
+        self.channel.send(frame, self.remaining())
         self.requests_sent += 1
         self.snapshot_open = True
 
@@ -858,6 +882,10 @@ def consume(self) -> dict:
             if chunk is WOULD_BLOCK:
                 continue
             if not chunk:
+                for record in self.decoder.finish():
+                    outcome = self.handle(record)
+                    if outcome is not None:
+                        return outcome
                 if self.decoder.pending_bytes:
                     raise SourceError(
                         "truncated_frame",
@@ -1001,7 +1029,7 @@ def run(spec: dict, out=sys.stdout) -> int:
         callback_ready, callback_status = wait.callback_readiness()
         try:
             wait.attach()
-        except SourceError as exc:
+        except (SourceError, Timeout) as exc:
             wait.emit(
                 {
                     "record": "armed",
@@ -1010,7 +1038,7 @@ def run(spec: dict, out=sys.stdout) -> int:
                     "source_ready": False,
                     "callback_ready": callback_ready,
                     "callback_status": callback_status,
-                    "error_code": exc.code,
+                    "error_code": getattr(exc, "code", "deadline_exceeded"),
                 }
             )
             raise
diff --git a/product/skills/event-wait/tests/test_wait_event.py b/product/skills/event-wait/tests/test_wait_event.py
index 9457a72cd..47adba5c5 100644
--- a/product/skills/event-wait/tests/test_wait_event.py
+++ b/product/skills/event-wait/tests/test_wait_event.py
@@ -13,6 +13,7 @@
 """
 from __future__ import annotations
 
+import errno
 import json
 import os
 import shutil
@@ -159,7 +160,8 @@ def drop_peers(self, expect: int = 0) -> None:
             try:
                 peer.shutdown(socket.SHUT_RDWR)
             except OSError as exc:
-                print(f"producer: shutting down a peer failed: {exc}", file=sys.stderr)
+                if exc.errno != errno.ENOTCONN:
+                    print(f"producer: shutting down a peer failed: {exc}", file=sys.stderr)
             try:
                 peer.close()
             except OSError as exc:
@@ -582,6 +584,19 @@ def test_a_source_without_a_snapshot_is_never_written_to(self):
         self.assertEqual(producer.request_count(), 0)
 
 
+class ProducerHarnessTests(EventWaitTestCase):
+    def test_drop_peers_hands_a_connected_client_an_eof(self):
+        producer = self.producer("harness.sock")
+        client = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+        self.addCleanup(client.close)
+        client.connect(producer.path)
+
+        producer.drop_peers(expect=1)
+
+        client.settimeout(10.0)
+        self.assertEqual(client.recv(4096), b"")
+
+
 class StreamFailureTests(EventWaitTestCase):
     def test_a_malformed_frame_fails_closed_without_echoing_the_payload(self):
         producer = self.producer()
@@ -642,6 +657,34 @@ def test_a_match_sharing_a_read_with_a_bad_frame_is_still_delivered(self):
         self.assertEqual(receipt["status"], "completed")
         self.assertEqual(receipt["exit_code"], wait_event.EXIT_MATCHED)
 
+    def test_a_snapshot_write_to_a_peer_that_never_reads_ends_at_the_deadline(self):
+        path = os.path.join(self.dir, "deaf.sock")
+        server = socket.socket(socket.AF_UNIX, socket.SOCK_STREAM)
+        self.addCleanup(server.close)
+        server.bind(path)
+        server.listen(4)
+        held = []
+        acceptor = threading.Thread(target=lambda: held.append(server.accept()[0]), daemon=True)
+        acceptor.start()
+        self.addCleanup(lambda: [peer.close() for peer in held])
+        spec = self.socket_spec(
+            "deaf-peer",
+            "wf-1",
+            path,
+            deadline_seconds=2.0,
+            snapshot={
+                "request": {"kind": "req", "pad": "x" * 8_000_000},
+                "response_match_fields": {"kind": "res"},
+            },
+        )
+        wait = self.start(spec)
+        wait.read_armed()
+
+        receipt = wait.finish(timeout=12.0)
+        self.assertEqual(receipt["outcome"], "timeout", receipt)
+        self.assertEqual(receipt["error_code"], "deadline_exceeded")
+        self.assertEqual(receipt["exit_code"], wait_event.EXIT_TIMEOUT)
+
     def test_a_truncated_frame_at_disconnect_is_reported_as_truncated(self):
         producer = self.producer()
         spec = self.socket_spec("truncated", "wf-1", producer.path, deadline_seconds=8.0)
@@ -774,6 +817,47 @@ def test_a_stream_that_ends_before_the_match_is_a_source_error(self):
         self.assertEqual(receipt["error_code"], "source_closed")
         self.assertFalse(receipt["event_received"])
 
+    def test_a_final_record_without_a_newline_is_still_matched(self):
+        script = (
+            "import json,sys\n"
+            "sys.stdout.write(json.dumps({'workflowId':'wf-1','status':'running'}) + '\\n')\n"
+            "sys.stdout.write(json.dumps({'workflowId':'wf-1','status':'completed'}))\n"
+        )
+        spec = self.stream_spec("stream-no-newline", "wf-1", script)
+        wait = self.start(spec)
+        wait.read_armed()
+
+        receipt = wait.finish()
+        self.assertEqual(receipt["outcome"], "matched", receipt)
+        self.assertEqual(receipt["status"], "completed")
+
+    def test_a_partial_final_line_is_still_reported_as_truncated(self):
+        script = "import sys\nsys.stdout.write('{\"workflowId\": \"wf-1\", \"sta')\n"
+        spec = self.stream_spec("stream-partial", "wf-1", script)
+        wait = self.start(spec)
+        wait.read_armed()
+
+        receipt = wait.finish()
+        self.assertEqual(receipt["error_code"], "truncated_frame", receipt)
+
+    def test_a_source_command_that_ignores_terminate_is_killed_on_close(self):
+        command = [
+            sys.executable,
+            "-u",
+            "-c",
+            "import signal,time\n"
+            "signal.signal(signal.SIGTERM, signal.SIG_IGN)\n"
+            "print('ready')\n"
+            "time.sleep(60)\n",
+        ]
+        channel = wait_event.StreamChannel(command, None)
+        self.addCleanup(lambda: channel.process.poll() is None and channel.process.kill())
+        self.assertIn(b"ready", channel.recv(10.0))
+
+        channel.close()
+
+        self.assertIsNotNone(channel.process.poll())
+
     def test_a_stream_source_may_not_declare_a_snapshot(self):
         spec = self.stream_spec("stream-snap", "wf-1", "pass")
         spec["source"]["snapshot"] = {"request": {}}

From aa9412e3473da3005fb48f125af8af0ca623d14b Mon Sep 17 00:00:00 2001
From: Edbert Chan <edbert.chan@yahoo.com>
Date: Wed, 23 Sep 2026 22:13:45 +0800
Subject: [PATCH 15/17] llm-judge: find Codex's transcript by thread-id, and
 deliver subagent verdicts to the parent

Codex's notify payload carries no transcript path. A real capture from
`codex exec` has exactly: client, cwd, input-messages,
last-assistant-message, thread-id, turn-id, type. Both resolvers looked only
for transcript_path / conversation_id, so every Codex job was queued with an
empty transcript and its verdict was written where no drain looks. The Codex
notify tests passed only because their fixture supplied a transcript_path
the real harness never sends.

thread-id names the rollout file under ~/.codex/sessions, so both
resolvers now fall back to that file (shared in _sdk/transcripts.py).

A Claude subagent's verdict lands after the subagent has stopped, so no
later PostToolUse inside it ever drains it. The parent's drain now also
drains <session>/subagents/*.jsonl.

Real path: a real `codex exec` turn with this branch's notify queued against
its rollout file, the judge returned hit, and llm-judge's codex_notify
delivered the on_hit text; report.py --judge --check exited 0.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Change-Id: I5a619714e44f318c4abf315ebfb7783657f5baec
---
 engine/hooks/_sdk/transcripts.py              | 28 +++++++++
 engine/hooks/llm-judge/inbox.py               |  8 ++-
 engine/hooks/llm-judge/tests/test_inbox.py    | 58 +++++++++++++++++++
 engine/hooks/wrong-check-reflect/detect.py    |  3 +-
 .../wrong-check-reflect/tests/test_hooks.py   | 19 ++++++
 5 files changed, 113 insertions(+), 3 deletions(-)
 create mode 100644 engine/hooks/_sdk/transcripts.py

diff --git a/engine/hooks/_sdk/transcripts.py b/engine/hooks/_sdk/transcripts.py
new file mode 100644
index 000000000..5237a5924
--- /dev/null
+++ b/engine/hooks/_sdk/transcripts.py
@@ -0,0 +1,28 @@
+from __future__ import annotations
+
+import glob
+import os
+import re
+
+CODEX_SESSIONS_ENV = "CATSTACK_CODEX_SESSIONS_DIR"
+THREAD_ID = re.compile(r"^[0-9A-Za-z-]{8,64}$")
+
+
+def codex_sessions_root() -> str:
+    return os.environ.get(CODEX_SESSIONS_ENV) or os.path.join(os.path.expanduser("~"), ".codex", "sessions")
+
+
+def codex_rollout(payload: dict) -> str:
+    thread = payload.get("thread-id") or payload.get("thread_id")
+    if not isinstance(thread, str) or not THREAD_ID.match(thread):
+        return ""
+    pattern = os.path.join(codex_sessions_root(), "*", "*", "*", f"rollout-*-{thread}.jsonl")
+    matches = sorted(glob.glob(pattern))
+    return matches[-1] if matches else ""
+
+
+def subagent_transcripts(transcript: str) -> list[str]:
+    if not transcript.endswith(".jsonl"):
+        return []
+    folder = os.path.join(transcript[: -len(".jsonl")], "subagents")
+    return sorted(glob.glob(os.path.join(folder, "*.jsonl")))
diff --git a/engine/hooks/llm-judge/inbox.py b/engine/hooks/llm-judge/inbox.py
index 669ed910d..262739830 100644
--- a/engine/hooks/llm-judge/inbox.py
+++ b/engine/hooks/llm-judge/inbox.py
@@ -5,6 +5,7 @@
 import os
 
 import judge
+from transcripts import codex_rollout, subagent_transcripts
 
 NO_TRANSCRIPT = "llm-judge: {harness} payload has no transcript path, so finished verdicts were not checked"
 REPORT_LIMIT = 600
@@ -31,7 +32,7 @@ def resolve_transcript(payload: dict) -> str:
             candidate = os.path.join(root, project, "agent-transcripts", conv, f"{conv}.jsonl")
             if os.path.isfile(candidate):
                 return candidate
-    return ""
+    return codex_rollout(payload)
 
 
 def _turn_text(data: dict) -> str:
@@ -167,7 +168,10 @@ def unchecked_message(item: dict) -> str:
 
 def messages(transcript: str) -> list[str]:
     out = []
-    for item in judge.drain(transcript):
+    drained = []
+    for path in [transcript, *subagent_transcripts(transcript)]:
+        drained.extend(judge.drain(path))
+    for item in drained:
         outcome = item.get("outcome")
         if outcome == "clean":
             continue
diff --git a/engine/hooks/llm-judge/tests/test_inbox.py b/engine/hooks/llm-judge/tests/test_inbox.py
index a6ad26a91..057719bac 100644
--- a/engine/hooks/llm-judge/tests/test_inbox.py
+++ b/engine/hooks/llm-judge/tests/test_inbox.py
@@ -355,5 +355,63 @@ def test_no_transcript_says_unchecked_on_stderr(self):
         self.assertIn("no transcript path", self.run_codex([json.dumps({"type": "agent-turn-complete", "thread-id": "t1"})]))
 
 
+
+REAL_CODEX_NOTIFY_KEYS = ["client", "cwd", "input-messages", "last-assistant-message", "thread-id", "turn-id", "type"]
+THREAD = "01a0ce9a-301b-7d42-bb5c-b9a9a4cfe8c6"
+
+
+class TestTranscriptResolution(InboxTestCase):
+    def setUp(self):
+        super().setUp()
+        sessions = os.path.join(self.work.name, "codex-sessions")
+        day = os.path.join(sessions, "2026", "09", "23")
+        os.makedirs(day)
+        self.rollout = os.path.join(day, f"rollout-2026-09-23T22-10-06-{THREAD}.jsonl")
+        with open(self.rollout, "w", encoding="utf-8") as handle:
+            handle.write("{}\n")
+        self.sessions_env = patch.dict(os.environ, {"CATSTACK_CODEX_SESSIONS_DIR": sessions})
+        self.sessions_env.start()
+
+    def tearDown(self):
+        self.sessions_env.stop()
+        super().tearDown()
+
+    def real_codex_payload(self):
+        payload = {
+            "type": "agent-turn-complete",
+            "thread-id": THREAD,
+            "turn-id": "01a0ce9a-30dc-73f1-bfe6-ccce644446e1",
+            "cwd": self.work.name,
+            "client": "codex_exec",
+            "input-messages": ["Reply with the single word ok."],
+            "last-assistant-message": "ok",
+        }
+        self.assertEqual(sorted(payload), REAL_CODEX_NOTIFY_KEYS)
+        return payload
+
+    def test_real_codex_notify_payload_resolves_to_its_rollout_file(self):
+        self.assertEqual(inbox.resolve_transcript(self.real_codex_payload()), self.rollout)
+
+    def test_hit_for_a_codex_rollout_reaches_the_next_codex_notify(self):
+        self.transcript = self.rollout
+        self.seed(ANSWERS_TRUE)
+        self.assertEqual(self.run_codex([json.dumps(self.real_codex_payload())]), ON_HIT + "\n")
+
+    def test_unknown_or_unsafe_thread_id_resolves_to_nothing(self):
+        for thread in ("0000000-0000-not-there", "../../etc", "*"):
+            with self.subTest(thread=thread):
+                self.assertEqual(inbox.resolve_transcript({"thread-id": thread}), "")
+
+    def test_a_subagent_verdict_is_delivered_to_the_parent_session(self):
+        subagent = os.path.join(self.transcript[: -len(".jsonl")], "subagents", "agent-a1.jsonl")
+        os.makedirs(os.path.dirname(subagent))
+        with open(subagent, "w", encoding="utf-8") as handle:
+            handle.write("{}\n")
+        parent = self.transcript
+        self.transcript = subagent
+        self.seed(ANSWERS_TRUE)
+        self.assertEqual(inbox.messages(parent), [ON_HIT])
+        self.assertEqual(inbox.messages(parent), [])
+
 if __name__ == "__main__":
     unittest.main()
diff --git a/engine/hooks/wrong-check-reflect/detect.py b/engine/hooks/wrong-check-reflect/detect.py
index 0bb9339b3..e1af76c89 100644
--- a/engine/hooks/wrong-check-reflect/detect.py
+++ b/engine/hooks/wrong-check-reflect/detect.py
@@ -15,6 +15,7 @@
     os.path.dirname(os.path.dirname(os.path.abspath(__file__))), "_sdk"))
 
 from events import write_stage_event  # noqa: E402
+from transcripts import codex_rollout  # noqa: E402
 from flags import enforcement_gate  # noqa: E402
 
 HOOKS_DIR = os.path.dirname(os.path.abspath(__file__))
@@ -258,7 +259,7 @@ def resolve_transcript(payload: dict) -> str:
                     return candidate
         except OSError:
             pass
-    return ""
+    return codex_rollout(payload)
 
 
 def _is_assistant_line(data: dict) -> bool:
diff --git a/engine/hooks/wrong-check-reflect/tests/test_hooks.py b/engine/hooks/wrong-check-reflect/tests/test_hooks.py
index 27a62d0c6..c647eebd9 100644
--- a/engine/hooks/wrong-check-reflect/tests/test_hooks.py
+++ b/engine/hooks/wrong-check-reflect/tests/test_hooks.py
@@ -545,6 +545,25 @@ def test_claude_stop_records_the_queued_job_under_the_claude_harness(self):
                          [("wrong-check-reflect", "claude", "transcript")])
         self.assertEqual(queued[0]["finding_id"], self.wait_for_jobs(1)[0][: -len(".json")])
 
+    def test_real_codex_notify_queues_against_its_rollout_file(self):
+        self.use_runners(SLOW_CLEAN)
+        thread = "01a0ce9a-301b-7d42-bb5c-b9a9a4cfe8c6"
+        day = os.path.join(self.reflect_state.name, "codex-sessions", "2026", "09", "23")
+        os.makedirs(day)
+        rollout = os.path.join(day, f"rollout-2026-09-23T22-10-06-{thread}.jsonl")
+        with open(rollout, "w", encoding="utf-8") as handle:
+            handle.write("{}\n")
+        payload = {"type": "agent-turn-complete", "thread-id": thread, "turn-id": "t", "cwd": "/",
+                   "client": "codex_exec", "input-messages": ["hi"], "last-assistant-message": HIT_TEXT}
+        with patch.dict(os.environ, {"CATSTACK_CODEX_SESSIONS_DIR": os.path.dirname(os.path.dirname(os.path.dirname(day)))}):
+            err = run_codex_notify([json.dumps(payload)])
+        self.assertEqual(err, "")
+        [job] = self.wait_for_jobs(1)
+        with open(os.path.join(self.state.name, "jobs", job), encoding="utf-8") as handle:
+            self.assertEqual(json.load(handle)["transcript"], rollout)
+        queued = [r for r in self.stage_rows() if r["action"] == "judge_queued"]
+        self.assertEqual([(r["harness"], r["reason"]) for r in queued], [("codex", "transcript")])
+
     def test_claude_malformed_stdin_fail_open(self):
         err = io.StringIO()
         with patch.object(sys, "stdin", io.StringIO("not-json")):

From 9df0641396c8000dd5c525ee803cf3df10ebb2f3 Mon Sep 17 00:00:00 2001
From: CI Bot <ci@invoker.dev>
Date: Wed, 23 Sep 2026 14:39:56 +0000
Subject: [PATCH 16/17] llm-judge: look for Codex rollouts under CODEX_HOME
 when it is set

Codex keeps its rollout files under $CODEX_HOME/sessions whenever that
variable is set, and the notify hook inherits the variable from the Codex
process that spawns it. codex_sessions_root() only looked at
CATSTACK_CODEX_SESSIONS_DIR and the hard-coded ~/.codex/sessions, so a
session run with a non-default Codex home resolved to no file at all: the
job was queued under an empty path and its verdict could never be
delivered.

The explicit CATSTACK_CODEX_SESSIONS_DIR override still wins, so the
tests keep pinning the sessions tree directly. Otherwise the root is
derived from CODEX_HOME, falling back to ~/.codex as before.

The new test fails against the previous helper with '' != the rollout
path, and passes with it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
---
 engine/hooks/_sdk/transcripts.py           |  7 ++++++-
 engine/hooks/llm-judge/tests/test_inbox.py | 15 +++++++++++++++
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/engine/hooks/_sdk/transcripts.py b/engine/hooks/_sdk/transcripts.py
index 5237a5924..2629a2ba3 100644
--- a/engine/hooks/_sdk/transcripts.py
+++ b/engine/hooks/_sdk/transcripts.py
@@ -5,11 +5,16 @@
 import re
 
 CODEX_SESSIONS_ENV = "CATSTACK_CODEX_SESSIONS_DIR"
+CODEX_HOME_ENV = "CODEX_HOME"
 THREAD_ID = re.compile(r"^[0-9A-Za-z-]{8,64}$")
 
 
 def codex_sessions_root() -> str:
-    return os.environ.get(CODEX_SESSIONS_ENV) or os.path.join(os.path.expanduser("~"), ".codex", "sessions")
+    override = os.environ.get(CODEX_SESSIONS_ENV)
+    if override:
+        return override
+    home = os.environ.get(CODEX_HOME_ENV) or os.path.join(os.path.expanduser("~"), ".codex")
+    return os.path.join(home, "sessions")
 
 
 def codex_rollout(payload: dict) -> str:
diff --git a/engine/hooks/llm-judge/tests/test_inbox.py b/engine/hooks/llm-judge/tests/test_inbox.py
index 057719bac..8787882f7 100644
--- a/engine/hooks/llm-judge/tests/test_inbox.py
+++ b/engine/hooks/llm-judge/tests/test_inbox.py
@@ -397,6 +397,21 @@ def test_hit_for_a_codex_rollout_reaches_the_next_codex_notify(self):
         self.seed(ANSWERS_TRUE)
         self.assertEqual(self.run_codex([json.dumps(self.real_codex_payload())]), ON_HIT + "\n")
 
+    def test_a_codex_home_override_is_where_the_rollout_is_looked_up(self):
+        home = os.path.join(self.work.name, "codex-home")
+        day = os.path.join(home, "sessions", "2026", "09", "23")
+        os.makedirs(day)
+        rollout = os.path.join(day, f"rollout-2026-09-23T22-40-00-{THREAD}.jsonl")
+        with open(rollout, "w", encoding="utf-8") as handle:
+            handle.write("{}\n")
+        self.sessions_env.stop()
+        try:
+            with patch.dict(os.environ, {"CODEX_HOME": home}, clear=False):
+                os.environ.pop("CATSTACK_CODEX_SESSIONS_DIR", None)
+                self.assertEqual(inbox.resolve_transcript(self.real_codex_payload()), rollout)
+        finally:
+            self.sessions_env.start()
+
     def test_unknown_or_unsafe_thread_id_resolves_to_nothing(self):
         for thread in ("0000000-0000-not-there", "../../etc", "*"):
             with self.subTest(thread=thread):

From 619f890a4a086996f47b3e871cda840fb83dbf02 Mon Sep 17 00:00:00 2001
From: CI Bot <ci@invoker.dev>
Date: Wed, 23 Sep 2026 14:41:12 +0000
Subject: [PATCH 17/17] =?UTF-8?q?invoker:=20wf-1790173197097-223/repair=20?=
 =?UTF-8?q?=E2=80=94=20Repair=20PR=20#819:=20failed=5Fchecks:=20validate;?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Exit code: 0