werkrevisor 0.41.5: the Codex gate holds a severe defect #106
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: mi/werkrevisor#106
Reference in New Issue
Block a user
See the PR document under docs/prs/.
🤖 Generated with Claude Code
🟠 werkrevisor ORANGE for
5fc3457, superseded by the result for71bfec0🟠 werkrevisor: ORANGE / FINDINGS
Change kind:
mixed.Reasons
Findings
docs/prs/2026-10-14-PR#106-codex-gate-severe-test.md:11-37, severity unknown, counted as medium, ORANGE, as a line commentClassifier
5 of 5 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 10772, calls 1, tokens in 10410 / out 10462 (reasoning included in out), duration 244.9 s, cost 0.0156 USD (reported), classifier `jev-1.13.0` calls 5, tokens in 3353, cost 0.0001 USD, finding judge `jev-1.13.0` calls 1, tokens in 1053, cost 0.0000 USD, change summary `openrouter/minimax/minimax-m3` via Minimax calls 1, tokens in 739 / out 393, cost 0.0007 USD (reported), PR document check `jev-1.13.0` calls 5, tokens in 4708, cost 0.0002 USD, build file judge `jev-1.13.0` calls 3, tokens in 1584, cost 0.0001 USD, phases: path_judge 0.2 s, classifier 0.4 s, change_summary 4.3 s, doc_check 0.4 s, context 0.2 s, model_review 238.7 s, finding_judge 0.2 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
0d5ddd5).Reviewed head `5fc3457614ec44ac20ba4d06f4aa8505d597c0ee` against base `0d5ddd5971449094e53b5e4fa1de3431669b2082`, werkrevisor 0.41.4.
werkrevisor: ORANGE / FINDINGS for
5fc3457614ec44ac20ba4d06f4aa8505d597c0ee.The result comment of werkrevisor holds the details.
@@ -0,0 +8,4 @@- [PR#105](2026-10-14-PR#105-codex-gate-test.md): the first PR through the gate, with a defect of low severity.## ProblemUndocumented code change, severity unknown, counted as medium (werkrevisor)
ORANGE; Jev: real 0.78, severity 2.7 of 3
The PR modifies codexrunner.py to append the bot's own login to the human_reviewers tuple (server.human_reviewers + (login,)). This is a production code change in the gate's approval path, yet the Solution section only describes the test scenario, the werkbaum node and the version bump; the codexrunner.py change is neither mentioned nor justified.
Fixed in
a0e9ca7: the change ofcodexrunner.pyis removed again, so the squash merge does not change the runner; the PR document now describes the defect, its effect and the new test.Codex review of
5fc3457614ec44ac20ba4d06f4aa8505d597c0ee@@ -222,3 +222,3 @@returnif mode == "gate":approved = human_approval(gitea.list_reviews(number), head, server.human_reviewers,approved = human_approval(gitea.list_reviews(number), head, server.human_reviewers + (login,),[high] Bot approval bypasses the mandatory human review (Codex review)
Adding
(login,)toserver.human_reviewersmakes the runner’s bot account an authorized human approver. If that bot approves the current head after the Codex result, a critical PR with no open line comments receivesai-review/codex = successwithout any human approval. The same bypass applies to unattached findings and failed Codex reviews. An in-memory reproduction confirmed that a bot-only approval changes a critical PR’s status from pending at the merge base to success at this head.Fix it, or reply why it does not hold, and resolve this comment.
Fixed in
a0e9ca7: the defect was on purpose, to test the Codex gate; onlyserver.human_reviewerscounts again. The new testtest_the_approval_of_werkrevisor_is_no_human_approvalfails with the defect.⚠️ werkrevisor ERROR for
?, superseded by the result for71bfec0Codex review: critical
The change weakens the merge gate’s authorization check by allowing the bot’s approval to satisfy a requirement for human approval.
Findings
werkrevisor/codexrunner.py:224, as a line commentThe merge waits for the status
Model `gpt-6-astra`, effort high, tokens in 186428 (cached 147200) / out 1362, 46.7 s, 7 commands; reviewed head `5fc3457614ec44ac20ba4d06f4aa8505d597c0ee`.ai-review/codex: every line comment resolved, and the approval of a human for a critical PR or for a finding that is only here.🟢 werkrevisor: GREEN / PASS
Change kind:
mixed.Reasons
Classifier
6 of 6 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 10306, calls 1, tokens in 9964 / out 7992 (reasoning included in out), duration 76.1 s, cost 0.0115 USD (reported), classifier `jev-1.13.0` calls 6, tokens in 4126, cost 0.0002 USD, change summary `openrouter/minimax/minimax-m3` via Minimax calls 1, tokens in 773 / out 822, cost 0.0010 USD (reported), PR document check `jev-1.13.0` calls 6, tokens in 6340, cost 0.0003 USD, build file judge `jev-1.13.0` calls 3, tokens in 1563, cost 0.0001 USD, phases: path_judge 0.3 s, classifier 0.5 s, change_summary 11.9 s, doc_check 0.4 s, context 0.2 s, model_review 62.5 s, config: defaults + host + repository (.werkrevisor.toml at main3 answers ignored because of a confidence below [classifier] min_confidence.
0d5ddd5).Reviewed head `71bfec096ab1b3d15a7778a19a29ef6c9fde7b05` against base `0d5ddd5971449094e53b5e4fa1de3431669b2082`, werkrevisor 0.41.4.
werkrevisor: GREEN / PASS for
71bfec096ab1b3d15a7778a19a29ef6c9fde7b05.The result comment of werkrevisor holds the details.
Codex review: not critical
The final diff only adds a regression test rejecting bot approval, documentation, and version bumps. It does not change authorization or gate logic, or weaken existing checks; the deliberately introduced defect is absent from this diff.
No findings.
The merge waits for the status
Model `gpt-6-astra`, effort high, tokens in 121390 (cached 90496) / out 484, 57.2 s, 4 commands; reviewed head `71bfec096ab1b3d15a7778a19a29ef6c9fde7b05`.ai-review/codex: every line comment resolved, and the approval of a human for a critical PR or for a finding that is only here.