werkrevisor 0.65.2: the status ai-review/codex names who acts next #138
No Assignees
Notifications
Due Date
No due date set.
Dependencies
No dependencies set.
Reference: mi/werkrevisor#138
Reference in New Issue
Block a user
See the PR document under docs/prs/.
🤖 Generated with Claude Code
🟢 werkrevisor GREEN for
59bc6a3, superseded by the result for5caafdf🟢 werkrevisor: GREEN / PASS
Gate: approved by werkrevisor; merge waits for 2 open hint(s): reply to each line comment and resolve it, then restart the review or push a commit.
Change kind:
mixed.Reasons
Findings
[high] Weakened Assertions at
tests/test_codexrunner.py:167, hint, as a line comment[medium] Misleading Queue at
werkrevisor/codexrunner.py:25, hint, as a line commentClassifier
6 of 6 units passed all questions (1 PR document(s)).
Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 29374, calls 1, tokens in 28202 / out 47488 (reasoning included in out), duration 625.6 s, cost 0.0654 USD (reported), classifier `jev-1.14.0` calls 6, tokens in 4969, cost 0.0002 USD, finding judge `jev-1.14.0` calls 2, tokens in 3094, cost 0.0001 USD, PR document check `jev-1.14.0` calls 6, tokens in 15869, cost 0.0007 USD, build file judge `jev-1.14.0` calls 4, tokens in 5659, cost 0.0002 USD, phases: path_judge 0.3 s, classifier 0.4 s, doc_check 0.4 s, context 0.2 s, model_review 623.7 s, finding_judge 0.2 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
284f80e).Reviewed head `59bc6a367185395f17a0bcef3141c9a043fcaa3c` against base `284f80eeae1941db4f8ef2b56088533e1b710d98`, werkrevisor 0.65.0.
werkrevisor: GREEN / PASS for
59bc6a367185395f17a0bcef3141c9a043fcaa3c.The result comment of werkrevisor holds the details.
@@ -165,3 +165,3 @@c["resolver"] = {"login": "claude-code"}h.run()self.assertEqual(h.status(), ("success", "Codex review passed"))self.assertEqual(h.status(), ("success", codexrunner.PASSED))[high] Weakened Assertions (werkrevisor)
hint; Jev: real 0.59, severity 2.2 of 3
Several existing assertions that previously pinned literal status text now use
codexrunner.PASSED,QUEUED,RUNNING,HUMAN,FAILED, orFAILED_APPROVED. The runner emits those same constants, so changing a constant's wording or assigning the wrong allowed actor prefix can leave the suite green and no longer protects the externally visible status contract. Keep literal expected descriptions or separate immutable test fixtures.Fixed in
5caafdf:test_the_descriptions_are_the_documented_onespins each text literally; the other tests keep the constants, so one place holds the contract.@@ -22,2 +22,4 @@MAX_ATTEMPTS = 3MODES = ("off", "advise", "gate")# Gitea shows each pending status alike, so the description starts with who acts next.QUEUED = "queued ⏳: Codex reviews after werkrevisor"[medium] Misleading Queue (werkrevisor)
hint; Jev: real 0.75, severity 2.3 of 3
QUEUEDis set whenself.sandbox_faultprevents a review, butfail_sandboxrequires an operator to repair the workstation before Codex can run again. The new description says Codex reviews after werkrevisor, so a blocked PR identifies Codex as the next actor even though the operator is. This contradicts the who-acts-next contract; use a sandbox-specific or otherwise accurate description.Fixed in
5caafdf: a broken sandbox now showsoperator 🔧: the Codex sandbox fails on the workstation; see the runner log, a queued PR keepsqueued ⏳:.Codex review: not critical
The change only updates status descriptions, documentation, tests, and version numbers; merge states and approval requirements remain unchanged. Read-only checks confirmed unchanged states across 48 combinations; the full test suite could not run because Docker is unavailable.
No findings.
The merge waits for the status
Model `gpt-6-astra`, effort high, tokens in 178103 (cached 137728) / out 1630, 50.0 s, 8 commands; reviewed head `59bc6a367185395f17a0bcef3141c9a043fcaa3c`.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.New commits pushed, approval review dismissed automatically according to repository settings
🟠 werkrevisor ORANGE for
5caafdf, superseded by the result foraedea5d🟠 werkrevisor: ORANGE / FINDINGS
Change kind:
mixed.Reasons
59bc6a3not carried over: 8 approved block(s) are missingFindings
[medium] RED Status at
werkrevisor/codexrunner.py:307, ORANGE, as a line comment[low] Rollout Claim at
docs/prs/2026-10-23-PR#138-codex-status-who.md:71hint
The claim that the first run refreshes every status is false for WIP PRs:
run_repository()skips them beforecollect()can callset_status(). If a PR became WIP after receiving anai-review/codexstatus, that status retains its old description until the title changes, leaving the rollout incomplete. Qualify the claim or refresh these statuses as well.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 29871, calls 1, tokens in 28664 / out 14454 (reasoning included in out), duration 187.4 s, cost 0.0259 USD (reported), classifier `jev-1.14.0` calls 6, tokens in 5064, cost 0.0002 USD, finding judge `jev-1.14.0` calls 1, tokens in 1695, cost 0.0001 USD, PR document check `jev-1.14.0` calls 6, tokens in 16977, cost 0.0007 USD, build file judge `jev-1.14.0` calls 4, tokens in 5811, cost 0.0002 USD, phases: path_judge 0.2 s, classifier 0.3 s, doc_check 0.4 s, context 0.2 s, model_review 184.9 s, finding_judge 0.2 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
284f80e).Reviewed head `5caafdfea0c842115f1898057a4b8a110d13ab16` against base `284f80eeae1941db4f8ef2b56088533e1b710d98`, werkrevisor 0.65.0.
Diagrams
Drawn by a model from the diff with PR Lens: a view of what the PR touches, not a statement of what it does.
werkrevisor: ORANGE / FINDINGS for
5caafdfea0c842115f1898057a4b8a110d13ab16.The result comment of werkrevisor holds the details.
@@ -297,3 +305,3 @@if ready is None:if mode == "gate":self.set_status(gitea, head, "pending", "waiting for the Codex review")self.set_status(gitea, head, "pending", QUEUED)[medium] RED Status (werkrevisor)
ORANGE; Jev: real 0.81, severity 3.0 of 3
werkrevisor_result()returnsNoneboth before a result exists and for aREDverdict. For a non-WIP gate PR with a RED head, the new branch therefore setsQUEUED, although Codex deliberately skips RED heads and only a new author push can advance the workflow. This makes the status claim that Codex will run by itself and violates the documented actor-next behavior; distinguish RED from an absent result or use an author-facing status.Fixed in
aedea5d: a RED head now showsauthor ✍️: werkrevisor found RED; push a commit that fixes the cause; only a missing result showsqueued ⏳:.test_a_red_head_gets_no_codex_reviewchecks it.Codex review: not critical
The change updates status descriptions without changing approval requirements or merge-gate decisions, and does not weaken tests or security controls. Gate states matched the base version across all 32 decision-input combinations; the prescribed tests could not run because Docker is unavailable.
No findings.
The merge waits for the status
Model `gpt-6-astra`, effort high, tokens in 206027 (cached 180992) / out 1815, 61.2 s, 10 commands; reviewed head `5caafdfea0c842115f1898057a4b8a110d13ab16`.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.Answer to the review of
5caafdf(round 2), fixed inaedea5d:🟢 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 failed (TimeoutError), tokens counted before the call 30328, duration 724.5 s, cost unknown, classifier `jev-1.14.0` calls 6, tokens in 5160, cost 0.0002 USD, PR document check `jev-1.14.0` calls 6, tokens in 17907, cost 0.0008 USD, build file judge `jev-1.14.0` calls 4, tokens in 5973, cost 0.0003 USD, phases: path_judge 0.2 s, classifier 0.3 s, doc_check 0.3 s, context 0.2 s, model_review 722.7 s, config: defaults + host + repository (.werkrevisor.toml at main1 answer ignored because of a confidence below [classifier] min_confidence.
284f80e).Reviewed head `aedea5d10809cf5f265d0bffafcd2bb67063606d` against base `284f80eeae1941db4f8ef2b56088533e1b710d98`, werkrevisor 0.65.0.
werkrevisor: GREEN / PASS for
aedea5d10809cf5f265d0bffafcd2bb67063606d.The result comment of werkrevisor holds the details.
Codex review: not critical
The change updates status descriptions without changing merge-gate states, approval requirements, or security controls. The local description test passed; the required container tests could not run because Docker is unavailable.
No findings.
The merge waits for the status
Model `gpt-6-astra`, effort high, tokens in 223727 (cached 194560) / out 1072, 45.4 s, 10 commands; reviewed head `aedea5d10809cf5f265d0bffafcd2bb67063606d`.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.