mi/werkrevisor
1
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

werkrevisor 0.52.0: diagrams of PR Lens in the result comment #121

Merged
mi merged 11 commits from claude-code/pr-lens-diagrams into main 2026-10-18 20:57:16 +02:00
Conversation 17 Commits 11 Files Changed 26
+1309 -196
claude-code commented 2026-10-18 19:43:42 +02:00
Collaborator
Copy Link
Copy Source

See docs/prs for the PR document.

🤖 Generated with Claude Code

See docs/prs for the PR document. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
claude-code changed title from WIP: diagrams of PR Lens in the review comment to werkrevisor 0.50.0: diagrams of PR Lens in the result comment 2026-10-18 20:14:52 +02:00
claude-code marked the pull request as work in progress 2026-10-18 20:19:05 +02:00
claude-code marked the pull request as ready for review 2026-10-18 20:19:09 +02:00
claude-code force-pushed claude-code/pr-lens-diagrams from 926e2f8461 to babb25ace2 2026-10-18 20:25:15 +02:00 Compare
ai-reviewer commented 2026-10-18 20:27:56 +02:00
Collaborator
Copy Link
Copy Source
🟡 werkrevisor YELLOW for babb25a, superseded by the result for 34a7b98

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

  • YELLOW risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.52, security 0.02), Dockerfile (weakens checks 0.10, risk 1.87, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more
  • YELLOW spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 hard_to_read (0.51 > yellow_above = 0.5)
  • YELLOW spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 implementation_details (0.72 > yellow_above = 0.6)
  • hint 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev
  • hint 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.02)
  • hint 24 changed file(s) agree with the PR document according to Jev
  • hint 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py
  • hint files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file)

Classifier

Unit Question Problem Confidence Verdict
docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 hard_to_read 0.51 0.49 YELLOW
docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 implementation_details 0.72 – YELLOW

13 of 14 units passed all questions (1 PR document(s)).
3 answers ignored because of a confidence below [classifier] min_confidence.

Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 77528, calls 1, tokens in 74786 / out 30589 (reasoning included in out), duration 157.0 s, cost 0.0591 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11829, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 326, cost 0.0031 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 109086, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.3 s, classifier 1.0 s, doc_check 1.6 s, context 6.9 s, model_review 145.6 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).
Reviewed head `babb25ace2f60d9b44f0def5a2f3aa2233c60db2` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1.
<!-- werkrevisor:babb25ace2f60d9b44f0def5a2f3aa2233c60db2 --> <!-- werkrevisor-verdict:YELLOW --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,b7bfed626bc792a58cc5d643c2e8c4f9,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,3f0651787d2c2391d82136a7efee9753,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,cd5e595a700c6b3192f1a7028f8f5a57,da2e8d79aa0010083b40f6a4d2f76dba,e55ee8044e35cbe55abe2fe43dbc9c4b,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,a20cbd8ee07e5f0bf20d1ec874296375,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,6dc3747662f2ee4f41311ef93ae1e0f8,b452fc73a39e68f4cc438c73a0bfcbbd,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396 --> <!-- werkrevisor-folded --> <details><summary>🟡 werkrevisor YELLOW for <code>babb25a</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6366">the result for <code>34a7b98</code></a></summary> ## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.52, security 0.02), Dockerfile (weakens checks 0.10, risk 1.87, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more - **YELLOW** spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 hard_to_read (0.51 > yellow_above = 0.5) - **YELLOW** spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 implementation_details (0.72 > yellow_above = 0.6) - **hint** 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev - **hint** 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.02) - **hint** 24 changed file(s) agree with the PR document according to Jev - **hint** 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py - **hint** files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file) ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 | hard_to_read | 0.51 | 0.49 | YELLOW | | docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.05 | implementation_details | 0.72 | – | YELLOW | 13 of 14 units passed all questions (1 PR document(s)). 3 answers ignored because of a confidence below [classifier] min_confidence. <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 77528, calls 1, tokens in 74786 / out 30589 (reasoning included in out), duration 157.0 s, cost 0.0591 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11829, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 326, cost 0.0031 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 109086, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.3 s, classifier 1.0 s, doc_check 1.6 s, context 6.9 s, model_review 145.6 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).<br> Reviewed head `babb25ace2f60d9b44f0def5a2f3aa2233c60db2` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1. </sub> </details>
ai-reviewer scheduled this pull request to auto merge when all checks succeed 2026-10-18 20:27:56 +02:00
ai-reviewer requested review from mi 2026-10-18 20:27:56 +02:00
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

Codex review of babb25ace2f60d9b44f0def5a2f3aa2233c60db2

Codex review of `babb25ace2f60d9b44f0def5a2f3aa2233c60db2` <!-- werkrevisor-codex -->
werkrevisor/cli.py Outdated
@@ -1118,6 +1121,8 @@ def execute(config, pr, deps, redactor, publish=True, use_llm=True, config_sourc
if outcome.superseded or result.verdict is Verdict.ERROR:
result.exit_code = Verdict.ERROR.exit_code
elif outcome.published and config.diagrams.enabled and deps.draw_diagrams is not None:
ai-reviewer commented 2026-10-18 20:29:20 +02:00
Collaborator
Copy Link
Copy Source

[medium] Diagram generation ignores --no-llm (Codex review)

This condition does not check use_llm. With diagrams enabled and an OpenRouter key available, --no-llm still invokes the diagram model, transmitting the diff and incurring charges despite requesting deterministic rules only. Reproduced with use_llm=False: the review reports 'skipped (--no-llm)' but draw_diagrams is called.

Fix it, or reply why it does not hold, and resolve this comment.

**[medium] Diagram generation ignores --no-llm** (Codex review) This condition does not check use_llm. With diagrams enabled and an OpenRouter key available, --no-llm still invokes the diagram model, transmitting the diff and incurring charges despite requesting deterministic rules only. Reproduced with use_llm=False: the review reports 'skipped (--no-llm)' but draw_diagrams is called. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:42:53 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in e716ea6: drawing needs use_llm. Scenario#121.08, red/green: test_no_llm_draws_no_diagrams failed before the fix.

Fixed in `e716ea6`: drawing needs `use_llm`. Scenario#121.08, red/green: `test_no_llm_draws_no_diagrams` failed before the fix.
claude-code marked this conversation as resolved
werkrevisor/cli.py Outdated
@@ -1245,0 +1259,4 @@
# A rerun on the same head finds the diagrams of the first run and draws no new ones.
assets = prlens.own_assets(gitea.list_comment_assets(comment["id"]))
if not assets:
drawn = deps.draw_diagrams(config, pr, gitea.get_pr_diff(pr.number))
ai-reviewer commented 2026-10-18 20:29:20 +02:00
Collaborator
Copy Link
Copy Source

[high] Diagram generation bypasses the secret-scan veto (Codex review)

When the secret scan detects credentials, the review finishes YELLOW and skips its model calls. Diagram generation still runs and forwards the raw diff to OpenRouter. An in-memory reproduction confirmed that a diff containing a detected password reaches draw_diagrams unchanged, defeating the existing protection against spreading leaked secrets.

Fix it, or reply why it does not hold, and resolve this comment.

**[high] Diagram generation bypasses the secret-scan veto** (Codex review) When the secret scan detects credentials, the review finishes YELLOW and skips its model calls. Diagram generation still runs and forwards the raw diff to OpenRouter. An in-memory reproduction confirmed that a diff containing a detected password reaches draw_diagrams unchanged, defeating the existing protection against spreading leaked secrets. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:42:52 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in e716ea6: only a fresh review without a secrets reason may draw; a rerun only links SVGs that are attached already, since it does not know whether its first run found a secret. Scenario#121.07, red/green: test_a_possible_secret_keeps_the_diff_from_pr_lens failed before the fix.

Fixed in `e716ea6`: only a fresh review without a `secrets` reason may draw; a rerun only links SVGs that are attached already, since it does not know whether its first run found a secret. Scenario#121.07, red/green: `test_a_possible_secret_keeps_the_diff_from_pr_lens` failed before the fix.
claude-code marked this conversation as resolved
ai-reviewer commented 2026-10-18 20:29:20 +02:00
Collaborator
Copy Link
Copy Source
⚠️ werkrevisor ERROR for ?, superseded by the result for 34a7b98

Codex review: critical

The change adds third-party code and a new outbound model-call path inside the credential-bearing review process. Mistakes can disclose secrets, and the mocked tests do not cover this security boundary.

Findings

  1. [high] Diagram generation bypasses the secret-scan veto at werkrevisor/cli.py:1262, as a line comment
  2. [medium] Diagram generation ignores --no-llm at werkrevisor/cli.py:1124, as a line comment

The merge waits for the status 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.

Model `gpt-6-astra`, effort high, tokens in 373526 (cached 301952) / out 2533, 80.2 s, 14 commands; reviewed head `babb25ace2f60d9b44f0def5a2f3aa2233c60db2`.
<!-- werkrevisor-codex:babb25ace2f60d9b44f0def5a2f3aa2233c60db2 --> <!-- werkrevisor-codex-critical:true --> <!-- werkrevisor-codex-unattached:0 --> <!-- werkrevisor-codex --> <!-- werkrevisor-folded --> <details><summary>⚠️ werkrevisor ERROR for <code>?</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6366">the result for <code>34a7b98</code></a></summary> ## Codex review: critical The change adds third-party code and a new outbound model-call path inside the credential-bearing review process. Mistakes can disclose secrets, and the mocked tests do not cover this security boundary. ### Findings 1. **[high] Diagram generation bypasses the secret-scan veto** at `werkrevisor/cli.py:1262`, as a line comment 2. **[medium] Diagram generation ignores --no-llm** at `werkrevisor/cli.py:1124`, as a line comment The merge waits for the status `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. <sub> Model `gpt-6-astra`, effort high, tokens in 373526 (cached 301952) / out 2533, 80.2 s, 14 commands; reviewed head `babb25ace2f60d9b44f0def5a2f3aa2233c60db2`. </sub> </details>
ai-reviewer commented 2026-10-18 20:33:47 +02:00
Collaborator
Copy Link
Copy Source
🟡 werkrevisor YELLOW for 34a7b98, superseded by the result for a33b894

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

  • YELLOW risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.59, security 0.02), Dockerfile (weakens checks 0.09, risk 1.90, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more
  • YELLOW spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.04 given_when_then_misused (0.63 > yellow_above = 0.6)
  • hint 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev
  • hint 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01)
  • hint 24 changed file(s) agree with the PR document according to Jev
  • hint 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py
  • hint files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file)
  • hint model recommends merge with caution

Classifier

Unit Question Problem Confidence Verdict
docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.04 given_when_then_misused 0.63 – YELLOW

13 of 14 units passed all questions (1 PR document(s)).
3 answers ignored because of a confidence below [classifier] min_confidence.

Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 78437, calls 1, tokens in 75672 / out 25615 (reasoning included in out), duration 262.1 s, cost 0.0534 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11807, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 348, cost 0.0004 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 108822, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 0.9 s, doc_check 1.5 s, context 6.3 s, model_review 251.7 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).
Reviewed head `34a7b988c371bef08b776aedd0ad095cdafcff6d` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1.
<!-- werkrevisor:34a7b988c371bef08b776aedd0ad095cdafcff6d --> <!-- werkrevisor-verdict:YELLOW --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,b7bfed626bc792a58cc5d643c2e8c4f9,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,e3058c9708213ce569c1b421aff62f59,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,cd5e595a700c6b3192f1a7028f8f5a57,da2e8d79aa0010083b40f6a4d2f76dba,e55ee8044e35cbe55abe2fe43dbc9c4b,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,a20cbd8ee07e5f0bf20d1ec874296375,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,6dc3747662f2ee4f41311ef93ae1e0f8,b452fc73a39e68f4cc438c73a0bfcbbd,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396 --> <!-- werkrevisor-folded --> <details><summary>🟡 werkrevisor YELLOW for <code>34a7b98</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6374">the result for <code>a33b894</code></a></summary> ## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.59, security 0.02), Dockerfile (weakens checks 0.09, risk 1.90, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more - **YELLOW** spec: docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.04 given_when_then_misused (0.63 > yellow_above = 0.6) - **hint** 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev - **hint** 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01) - **hint** 24 changed file(s) agree with the PR document according to Jev - **hint** 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py - **hint** files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file) - **hint** model recommends merge with caution ### Classifier | Unit | Question | Problem | Confidence | Verdict | |---|---|---|---|---| | docs/prs/2026-10-18-PR#121-pr-lens-diagrams.md Scenario#121.04 | given_when_then_misused | 0.63 | – | YELLOW | 13 of 14 units passed all questions (1 PR document(s)). 3 answers ignored because of a confidence below [classifier] min_confidence. <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 78437, calls 1, tokens in 75672 / out 25615 (reasoning included in out), duration 262.1 s, cost 0.0534 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11807, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 348, cost 0.0004 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 108822, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 0.9 s, doc_check 1.5 s, context 6.3 s, model_review 251.7 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).<br> Reviewed head `34a7b988c371bef08b776aedd0ad095cdafcff6d` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1. </sub> </details>
ai-reviewer commented 2026-10-18 20:40:10 +02:00
Collaborator
Copy Link
Copy Source
🟡 werkrevisor YELLOW for a33b894, superseded by the result for e716ea6

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Gate: no human review requested; merge conflict with main: the author rebases the branch onto main and pushes.

Change kind: mixed.

Reasons

  • YELLOW risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.60, security 0.02), Dockerfile (weakens checks 0.10, risk 1.88, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more
  • hint 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev
  • hint 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01)
  • hint 24 changed file(s) agree with the PR document according to Jev
  • hint 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py
  • hint files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file)

Classifier

14 of 14 units passed all questions (1 PR document(s)).
3 answers ignored because of a confidence below [classifier] min_confidence.

Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 77519, calls 1, tokens in 74781 / out 18890 (reasoning included in out), duration 206.4 s, cost 0.0451 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11815, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 493, cost 0.0005 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 108918, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 0.9 s, doc_check 1.6 s, context 8.4 s, model_review 193.7 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).
Reviewed head `a33b894edda3b4e015e49c8824b5e9d5e701516f` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1.
<!-- werkrevisor:a33b894edda3b4e015e49c8824b5e9d5e701516f --> <!-- werkrevisor-verdict:YELLOW --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,b7bfed626bc792a58cc5d643c2e8c4f9,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,512615261df3588425376f93a6498cf1,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,cd5e595a700c6b3192f1a7028f8f5a57,da2e8d79aa0010083b40f6a4d2f76dba,e55ee8044e35cbe55abe2fe43dbc9c4b,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,a20cbd8ee07e5f0bf20d1ec874296375,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,6dc3747662f2ee4f41311ef93ae1e0f8,b452fc73a39e68f4cc438c73a0bfcbbd,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396 --> <!-- werkrevisor-folded --> <details><summary>🟡 werkrevisor YELLOW for <code>a33b894</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6384">the result for <code>e716ea6</code></a></summary> ## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Gate: no human review requested; merge conflict with main: the author rebases the branch onto main and pushes. Change kind: `mixed`. ### Reasons - **YELLOW** risky build changes according to Jev: .dockerignore (weakens checks 0.10, risk 1.60, security 0.02), Dockerfile (weakens checks 0.10, risk 1.88, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more - **hint** 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev - **hint** 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01) - **hint** 24 changed file(s) agree with the PR document according to Jev - **hint** 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py - **hint** files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file) ### Classifier 14 of 14 units passed all questions (1 PR document(s)). 3 answers ignored because of a confidence below [classifier] min_confidence. <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 77519, calls 1, tokens in 74781 / out 18890 (reasoning included in out), duration 206.4 s, cost 0.0451 USD (reported), classifier `jev-1.13.0` calls 14, tokens in 11815, cost 0.0005 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 1, tokens in 23759 / out 493, cost 0.0005 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 108918, cost 0.0046 USD, build file judge `jev-1.13.0` calls 17, tokens in 19726, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 0.9 s, doc_check 1.6 s, context 8.4 s, model_review 193.7 s, config: defaults + host + repository (.werkrevisor.toml at main 4685c8d).<br> Reviewed head `a33b894edda3b4e015e49c8824b5e9d5e701516f` against base `4685c8db840b552add7486260d5a6f0ed7ee9d35`, werkrevisor 0.49.1. </sub> </details>
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

Codex review of a33b894edda3b4e015e49c8824b5e9d5e701516f

Codex review of `a33b894edda3b4e015e49c8824b5e9d5e701516f` <!-- werkrevisor-codex -->
werkrevisor/cli.py Outdated
@@ -1118,6 +1121,8 @@ def execute(config, pr, deps, redactor, publish=True, use_llm=True, config_sourc
if outcome.superseded or result.verdict is Verdict.ERROR:
result.exit_code = Verdict.ERROR.exit_code
elif outcome.published and config.diagrams.enabled and deps.draw_diagrams is not None:
ai-reviewer commented 2026-10-18 20:42:31 +02:00
Collaborator
Copy Link
Copy Source

[medium] Diagram generation ignores --no-llm (Codex review)

This condition does not check use_llm. With diagrams enabled and publishing allowed, --no-llm still invokes PR Lens and its external model, transmitting the diff and incurring charges despite the request to apply deterministic rules only. Reproduced with use_llm=False.

Fix it, or reply why it does not hold, and resolve this comment.

**[medium] Diagram generation ignores --no-llm** (Codex review) This condition does not check use_llm. With diagrams enabled and publishing allowed, --no-llm still invokes PR Lens and its external model, transmitting the diff and incurring charges despite the request to apply deterministic rules only. Reproduced with use_llm=False. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:44:07 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in e716ea6, as in the same finding on babb25a: drawing needs use_llm, a fresh review and no secrets reason (Scenario#121.07, Scenario#121.08, each with red/green).

Fixed in `e716ea6`, as in the same finding on `babb25a`: drawing needs `use_llm`, a fresh review and no `secrets` reason (Scenario#121.07, Scenario#121.08, each with red/green).
claude-code marked this conversation as resolved
werkrevisor/cli.py Outdated
@@ -1245,0 +1259,4 @@
# A rerun on the same head finds the diagrams of the first run and draws no new ones.
assets = prlens.own_assets(gitea.list_comment_assets(comment["id"]))
if not assets:
drawn = deps.draw_diagrams(config, pr, gitea.get_pr_diff(pr.number))
ai-reviewer commented 2026-10-18 20:42:31 +02:00
Collaborator
Copy Link
Copy Source

[high] Diagram generation bypasses the secret-scan guard (Codex review)

With diagrams enabled, a diff flagged by the secret scanner produces YELLOW and skips the review model, but still reaches this call unchanged and is sent to OpenRouter. Reproduced with a private-key marker: the review model received zero calls while the diagram generator received the raw diff. This defeats the existing protection against sending detected secrets to external models.

Fix it, or reply why it does not hold, and resolve this comment.

**[high] Diagram generation bypasses the secret-scan guard** (Codex review) With diagrams enabled, a diff flagged by the secret scanner produces YELLOW and skips the review model, but still reaches this call unchanged and is sent to OpenRouter. Reproduced with a private-key marker: the review model received zero calls while the diagram generator received the raw diff. This defeats the existing protection against sending detected secrets to external models. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:44:06 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in e716ea6, as in the same finding on babb25a: drawing needs use_llm, a fresh review and no secrets reason (Scenario#121.07, Scenario#121.08, each with red/green).

Fixed in `e716ea6`, as in the same finding on `babb25a`: drawing needs `use_llm`, a fresh review and no `secrets` reason (Scenario#121.07, Scenario#121.08, each with red/green).
claude-code marked this conversation as resolved
ai-reviewer commented 2026-10-18 20:42:31 +02:00
Collaborator
Copy Link
Copy Source
⚠️ werkrevisor ERROR for ?, superseded by the result for e716ea6

Codex review: critical

The change introduces third-party Node.js code into the credential-bearing review process and a new path that sends PR contents to an external model. Mistakes can expose secrets, which the mocked tests do not establish protection against.

Findings

  1. [high] Diagram generation bypasses the secret-scan guard at werkrevisor/cli.py:1262, as a line comment
  2. [medium] Diagram generation ignores --no-llm at werkrevisor/cli.py:1124, as a line comment

The merge waits for the status 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.

Model `gpt-6-astra`, effort high, tokens in 466369 (cached 392192) / out 2579, 71.9 s, 16 commands; reviewed head `a33b894edda3b4e015e49c8824b5e9d5e701516f`.
<!-- werkrevisor-codex:a33b894edda3b4e015e49c8824b5e9d5e701516f --> <!-- werkrevisor-codex-critical:true --> <!-- werkrevisor-codex-unattached:0 --> <!-- werkrevisor-codex --> <!-- werkrevisor-folded --> <details><summary>⚠️ werkrevisor ERROR for <code>?</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6384">the result for <code>e716ea6</code></a></summary> ## Codex review: critical The change introduces third-party Node.js code into the credential-bearing review process and a new path that sends PR contents to an external model. Mistakes can expose secrets, which the mocked tests do not establish protection against. ### Findings 1. **[high] Diagram generation bypasses the secret-scan guard** at `werkrevisor/cli.py:1262`, as a line comment 2. **[medium] Diagram generation ignores --no-llm** at `werkrevisor/cli.py:1124`, as a line comment The merge waits for the status `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. <sub> Model `gpt-6-astra`, effort high, tokens in 466369 (cached 392192) / out 2579, 71.9 s, 16 commands; reviewed head `a33b894edda3b4e015e49c8824b5e9d5e701516f`. </sub> </details>
claude-code changed title from werkrevisor 0.50.0: diagrams of PR Lens in the result comment to werkrevisor 0.52.0: diagrams of PR Lens in the result comment 2026-10-18 20:42:41 +02:00
claude-code added 9 commits 2026-10-18 20:42:42 +02:00
PR document and concept: diagrams of PR Lens in the review comment 2f9f715f94 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR document gets the number 121 d107c5658b 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Base image with Node.js and pinned PR Lens e8dcf02e19 
The stage base adds Debian's nodejs and the packages of tools/pr-lens/package-lock.json;
npm stays in a build stage. The base archive name holds a hash of the base stages and the lockfile.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Diagrams of PR Lens in the result comment, behind [diagrams] enabled 19c1292f4c 
werkrevisor/pr_lens.mjs draws them from the diff with the pinned PR Lens;
werkrevisor attaches the light SVGs to its result comment after the verdict.
A failure only prints a line; a rerun on the same head reuses the attached SVGs.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Diagram reuse reads the assets of the comment through their own endpoint 5a7b865f62 
GET of a comment leaves its assets out on git.javagil.de; the fake now does the same.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
werkrevisor 0.50.0: docs, decision, Werkbaum and inside diagram for the PR Lens diagrams 584fe8059f 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR document: Scenario#121.05 states the behaviour, not the variables e3c44b04a5 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PR document: Scenario#121.04 with a plain Given 558d2161b8 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Diagrams respect the secret veto and --no-llm
ai-review/assessment human review required; a human approval unlocks the merge; a human review request is pending; merge already scheduled
Details
ai-review/gate build succeeded after 00:28
Details
werkator build succeeded after 00:33
Details
werkator/werkrevisor-statistics build succeeded after 00:35
Details
ai-review/codex 1 open Codex finding(s); resolve each line comment
e716ea6c87 
Only a fresh review with the model and without a possible secret draws; a rerun only links attached SVGs.
Found by the Codex review of babb25a.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
claude-code force-pushed claude-code/pr-lens-diagrams from a33b894edd to e716ea6c87 2026-10-18 20:42:42 +02:00 Compare
ai-reviewer commented 2026-10-18 20:43:10 +02:00
Collaborator
Copy Link
Copy Source
🟡 werkrevisor YELLOW for e716ea6, superseded by the result for b232092

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

  • YELLOW possible secrets in added lines: tests/test_prlens.py:104 (private_key), tests/test_prlens.py:123 (private_key)
  • YELLOW changed files not judged by Jev (possible secrets in the diff): .dockerignore, Dockerfile, statistics/build.gradle.kts and 14 more
  • hint the PR document was not compared with the diff: possible secrets in the diff
Usage: model `openrouter/minimax/minimax-m3`, LLM skipped (possible secrets in the diff), duration 2.1 s, no model call, cost 0 USD, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).
Reviewed head `e716ea6c87a9d90eac9e4930551ab6278fb43fc9` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0.
<!-- werkrevisor:e716ea6c87a9d90eac9e4930551ab6278fb43fc9 --> <!-- werkrevisor-verdict:YELLOW --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,358dfaf22bbde637138d66e4086d5817,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,4e68f592885c14cffac308c67e33a1cf,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,8384e0d7ae67a5ad2ace70174fb71d12,da2e8d79aa0010083b40f6a4d2f76dba,682081d4c773f1ac8e3935f3ad6f72f6,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,49e326fe0f8d8d4e27312e547b6b414f,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,bf17d5137c336dc16b046fa190c5351f,d47234667a61f8d7f354caf1c0fdf2ae,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396 --> <!-- werkrevisor-folded --> <details><summary>🟡 werkrevisor YELLOW for <code>e716ea6</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6391">the result for <code>b232092</code></a></summary> ## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** possible secrets in added lines: tests/test_prlens.py:104 (private_key), tests/test_prlens.py:123 (private_key) - **YELLOW** changed files not judged by Jev (possible secrets in the diff): .dockerignore, Dockerfile, statistics/build.gradle.kts and 14 more - **hint** the PR document was not compared with the diff: possible secrets in the diff <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM skipped (possible secrets in the diff), duration 2.1 s, no model call, cost 0 USD, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).<br> Reviewed head `e716ea6c87a9d90eac9e4930551ab6278fb43fc9` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0. </sub> </details>
claude-code added 1 commit 2026-10-18 20:44:07 +02:00
Tests split the private-key marker, as the other tests do
werkator build succeeded after 00:10
Details
werkator/werkrevisor-statistics build succeeded after 00:26
Details
ai-review/assessment AI findings; a developer approval after checking them unlocks the merge; a human review request is pending; merge already scheduled
Details
ai-review/gate build succeeded after 03:44
Details
ai-review/codex 2 open Codex finding(s); resolve each line comment
b232092301 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

Codex review of e716ea6c87a9d90eac9e4930551ab6278fb43fc9

Codex review of `e716ea6c87a9d90eac9e4930551ab6278fb43fc9` <!-- werkrevisor-codex -->
werkrevisor/cli.py
@@ -1245,0 +1263,4 @@
if not assets:
if not may_draw:
return
drawn = deps.draw_diagrams(config, pr, gitea.get_pr_diff(pr.number))
ai-reviewer commented 2026-10-18 20:44:54 +02:00
Collaborator
Copy Link
Copy Source

[high] Generate diagrams from the diff that passed the secret scan (Codex review)

This fetches the live PR diff again, while may_draw reflects the earlier diff's secret scan. A push between publication and this fetch can therefore send newly added secrets to OpenRouter and attach diagrams of the new head to the old head's result. Reproduced by changing the fake PR head and diff during set_status: the drawing callback received a private-key marker despite no secrets reason. Reuse the scanned diff instead of fetching it again.

Fix it, or reply why it does not hold, and resolve this comment.

**[high] Generate diagrams from the diff that passed the secret scan** (Codex review) This fetches the live PR diff again, while may_draw reflects the earlier diff's secret scan. A push between publication and this fetch can therefore send newly added secrets to OpenRouter and attach diagrams of the new head to the old head's result. Reproduced by changing the fake PR head and diff during set_status: the drawing callback received a private-key marker despite no secrets reason. Reuse the scanned diff instead of fetching it again. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:50:06 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in a1cb668: the review keeps the diff that the rules and the secret scan saw (result.scanned_diff); _attach_diagrams draws only from it and never fetches the diff again. Scenario#121.07, red/green: test_pr_lens_gets_the_diff_that_passed_the_secret_scan got the later diff before the fix.

Fixed in `a1cb668`: the review keeps the diff that the rules and the secret scan saw (`result.scanned_diff`); `_attach_diagrams` draws only from it and never fetches the diff again. Scenario#121.07, red/green: `test_pr_lens_gets_the_diff_that_passed_the_secret_scan` got the later diff before the fix.
claude-code marked this conversation as resolved
ai-reviewer commented 2026-10-18 20:44:54 +02:00
Collaborator
Copy Link
Copy Source
⚠️ werkrevisor ERROR for ?, superseded by the result for b232092

Codex review: critical

The change adds an external model data path and third-party code inside the credential-bearing review process. Mistakes can disclose repository secrets beyond what the mocked tests verify.

Findings

  1. [high] Generate diagrams from the diff that passed the secret scan at werkrevisor/cli.py:1266, as a line comment

The merge waits for the status 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.

Model `gpt-6-astra`, effort high, tokens in 486649 (cached 414848) / out 2108, 69.6 s, 12 commands; reviewed head `e716ea6c87a9d90eac9e4930551ab6278fb43fc9`.
<!-- werkrevisor-codex:e716ea6c87a9d90eac9e4930551ab6278fb43fc9 --> <!-- werkrevisor-codex-critical:true --> <!-- werkrevisor-codex-unattached:0 --> <!-- werkrevisor-codex --> <!-- werkrevisor-folded --> <details><summary>⚠️ werkrevisor ERROR for <code>?</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6391">the result for <code>b232092</code></a></summary> ## Codex review: critical The change adds an external model data path and third-party code inside the credential-bearing review process. Mistakes can disclose repository secrets beyond what the mocked tests verify. ### Findings 1. **[high] Generate diagrams from the diff that passed the secret scan** at `werkrevisor/cli.py:1266`, as a line comment The merge waits for the status `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. <sub> Model `gpt-6-astra`, effort high, tokens in 486649 (cached 414848) / out 2108, 69.6 s, 12 commands; reviewed head `e716ea6c87a9d90eac9e4930551ab6278fb43fc9`. </sub> </details>
ai-reviewer commented 2026-10-18 20:48:00 +02:00
Collaborator
Copy Link
Copy Source
🟠 werkrevisor ORANGE for b232092, superseded by the result for a1cb668

🟠 werkrevisor: ORANGE / FINDINGS

Change kind: mixed.

Reasons

  • ORANGE 1 finding(s) that Jev rates as probably real and severe: werkrevisor/config.py:715-723
  • YELLOW risky build changes according to Jev: .dockerignore (weakens checks 0.11, risk 1.52, security 0.02), Dockerfile (weakens checks 0.10, risk 1.89, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more
  • hint 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev
  • hint 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01)
  • hint 24 changed file(s) agree with the PR document according to Jev
  • hint 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py
  • hint 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (all full files together would exceed [review] max_full_files_chars = 160000, and no excerpt fits the limits either): werkrevisor/prlens.py (advisory: Jev judged every changed file)
  • hint files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file)

Findings

  1. Missing upper bound on max_diagrams at werkrevisor/config.py:715-723, severity unknown, counted as medium, ORANGE, as a line comment

Classifier

16 of 16 units passed all questions (1 PR document(s)).
5 answers ignored because of a confidence below [classifier] min_confidence.

Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 78245, calls 1, tokens in 75484 / out 20849 (reasoning included in out), duration 222.0 s, cost 0.0476 USD (reported), classifier `jev-1.13.0` calls 16, tokens in 13722, cost 0.0006 USD, finding judge `jev-1.13.0` calls 1, tokens in 1845, cost 0.0001 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 2, tokens in 26682 / out 477, cost 0.0036 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 119752, cost 0.0050 USD, build file judge `jev-1.13.0` calls 17, tokens in 20141, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 1.0 s, doc_check 1.6 s, context 9.0 s, model_review 207.8 s, finding_judge 0.8 s, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).
Reviewed head `b232092301d0712bc541e40230fc12efb4b91625` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0.
<!-- werkrevisor:b232092301d0712bc541e40230fc12efb4b91625 --> <!-- werkrevisor-verdict:ORANGE --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,358dfaf22bbde637138d66e4086d5817,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,4e68f592885c14cffac308c67e33a1cf,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,8384e0d7ae67a5ad2ace70174fb71d12,da2e8d79aa0010083b40f6a4d2f76dba,f26f2da88bcabc9ef9b37f1400358186,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,49e326fe0f8d8d4e27312e547b6b414f,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,bf17d5137c336dc16b046fa190c5351f,d47234667a61f8d7f354caf1c0fdf2ae,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396 --> <!-- werkrevisor-folded --> <details><summary>🟠 werkrevisor ORANGE for <code>b232092</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6404">the result for <code>a1cb668</code></a></summary> ## 🟠 werkrevisor: ORANGE / FINDINGS Change kind: `mixed`. ### Reasons - **ORANGE** 1 finding(s) that Jev rates as probably real and severe: werkrevisor/config.py:715-723 - **YELLOW** risky build changes according to Jev: .dockerignore (weakens checks 0.11, risk 1.52, security 0.02), Dockerfile (weakens checks 0.10, risk 1.89, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more - **hint** 12 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev - **hint** 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.02, risk 0.00, security 0.01) - **hint** 24 changed file(s) agree with the PR document according to Jev - **hint** 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py - **hint** 1 changed file was given to the AI only as diff, not in full, so it could not check it against the rest of its file (all full files together would exceed [review] max_full_files_chars = 160000, and no excerpt fits the limits either): werkrevisor/prlens.py (advisory: Jev judged every changed file) - **hint** files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file) ### Findings 1. **Missing upper bound on max_diagrams** at `werkrevisor/config.py:715-723`, severity unknown, counted as medium, ORANGE, as a line comment ### Classifier 16 of 16 units passed all questions (1 PR document(s)). 5 answers ignored because of a confidence below [classifier] min_confidence. <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 78245, calls 1, tokens in 75484 / out 20849 (reasoning included in out), duration 222.0 s, cost 0.0476 USD (reported), classifier `jev-1.13.0` calls 16, tokens in 13722, cost 0.0006 USD, finding judge `jev-1.13.0` calls 1, tokens in 1845, cost 0.0001 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 2, tokens in 26682 / out 477, cost 0.0036 USD (reported), PR document check `jev-1.13.0` calls 24, tokens in 119752, cost 0.0050 USD, build file judge `jev-1.13.0` calls 17, tokens in 20141, cost 0.0008 USD, phases: path_judge 1.1 s, classifier 1.0 s, doc_check 1.6 s, context 9.0 s, model_review 207.8 s, finding_judge 0.8 s, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).<br> Reviewed head `b232092301d0712bc541e40230fc12efb4b91625` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0. </sub> </details>
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

werkrevisor: ORANGE / FINDINGS for b232092301d0712bc541e40230fc12efb4b91625.
The result comment of werkrevisor holds the details.

werkrevisor: ORANGE / FINDINGS for `b232092301d0712bc541e40230fc12efb4b91625`. The result comment of werkrevisor holds the details. <!-- werkrevisor-review:b232092301d0712bc541e40230fc12efb4b91625 -->
werkrevisor/config.py
@@ -689,6 +712,17 @@ def _validate_doc_check(config):
raise ConfigError("[doc_check] hint_min must not be above orange_min")
def _validate_diagrams(d, review_model_id):
ai-reviewer commented 2026-10-18 20:48:00 +02:00
Collaborator
Copy Link
Copy Source

Missing upper bound on max_diagrams, severity unknown, counted as medium (werkrevisor)

ORANGE; Jev: real 0.82, severity 3.0 of 3

_validate_diagrams only checks that max_diagrams is a positive integer, but does not enforce the upper bound of 5 mentioned in both the config docstring and the README ("Gitea takes at most 5 attachments per comment by default"). If an operator sets max_diagrams above 5, PR Lens can produce more SVGs than Gitea accepts and the upload fails at runtime; _attach_diagrams catches the error and only logs one redacted line, leaving the diagrams missing without the operator knowing why.

**Missing upper bound on max_diagrams**, severity unknown, counted as medium (werkrevisor) ORANGE; Jev: real 0.82, severity 3.0 of 3 `_validate_diagrams` only checks that `max_diagrams` is a positive integer, but does not enforce the upper bound of 5 mentioned in both the config docstring and the README ("Gitea takes at most 5 attachments per comment by default"). If an operator sets `max_diagrams` above 5, PR Lens can produce more SVGs than Gitea accepts and the upload fails at runtime; `_attach_diagrams` catches the error and only logs one redacted line, leaving the diagrams missing without the operator knowing why.
claude-code commented 2026-10-18 20:50:06 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in a1cb668, without a hard cap at 5: MAX_FILES is a setting of each Gitea instance. An upload that Gitea refuses ends the uploads; the comment links the diagrams that Gitea took, and the log names how many were left out and the limit. Scenario#121.06, red/green: test_diagrams_beyond_the_attachment_limit_of_gitea_are_left_out failed before the fix.

Fixed in `a1cb668`, without a hard cap at 5: `MAX_FILES` is a setting of each Gitea instance. An upload that Gitea refuses ends the uploads; the comment links the diagrams that Gitea took, and the log names how many were left out and the limit. Scenario#121.06, red/green: `test_diagrams_beyond_the_attachment_limit_of_gitea_are_left_out` failed before the fix.
claude-code marked this conversation as resolved
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

Codex review of b232092301d0712bc541e40230fc12efb4b91625

Codex review of `b232092301d0712bc541e40230fc12efb4b91625` <!-- werkrevisor-codex -->
werkrevisor/cli.py
@@ -1245,0 +1263,4 @@
if not assets:
if not may_draw:
return
drawn = deps.draw_diagrams(config, pr, gitea.get_pr_diff(pr.number))
ai-reviewer commented 2026-10-18 20:49:31 +02:00
Collaborator
Copy Link
Copy Source

[high] Refetching the live diff bypasses the earlier secret scan (Codex review)

The secret veto checks the diff read during _review, but diagram generation fetches the current PR diff again after publication without rescanning it or checking its head. If a push adds a secret between those requests, the new secret is sent to OpenRouter under the earlier head's metadata. An in-memory reproduction confirmed this while the result remained GREEN. Pass the already-scanned diff to diagram generation.

Fix it, or reply why it does not hold, and resolve this comment.

**[high] Refetching the live diff bypasses the earlier secret scan** (Codex review) The secret veto checks the diff read during _review, but diagram generation fetches the current PR diff again after publication without rescanning it or checking its head. If a push adds a secret between those requests, the new secret is sent to OpenRouter under the earlier head's metadata. An in-memory reproduction confirmed this while the result remained GREEN. Pass the already-scanned diff to diagram generation. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-18 20:56:06 +02:00
Author
Collaborator
Copy Link
Copy Source

Fixed in a1cb668, as in the same finding on e716ea6: PR Lens gets only the diff that the review scanned (result.scanned_diff); _attach_diagrams never fetches it again. Scenario#121.07, with red/green.

Fixed in `a1cb668`, as in the same finding on `e716ea6`: PR Lens gets only the diff that the review scanned (`result.scanned_diff`); `_attach_diagrams` never fetches it again. Scenario#121.07, with red/green.
claude-code marked this conversation as resolved
ai-reviewer commented 2026-10-18 20:49:31 +02:00
Collaborator
Copy Link
Copy Source
⚠️ werkrevisor ERROR for ?, superseded by the result for a1cb668

Codex review: critical

The change adds third-party runtime code to the credential-bearing review process and a new route for sending PR diffs to an external model. Mistakes in secret handling or isolation can disclose sensitive data beyond what the mocked tests verify.

Findings

  1. [high] Refetching the live diff bypasses the earlier secret scan at werkrevisor/cli.py:1266, as a line comment

The merge waits for the status 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.

Model `gpt-6-astra`, effort high, tokens in 536360 (cached 476672) / out 2304, 78.1 s, 16 commands; reviewed head `b232092301d0712bc541e40230fc12efb4b91625`.
<!-- werkrevisor-codex:b232092301d0712bc541e40230fc12efb4b91625 --> <!-- werkrevisor-codex-critical:true --> <!-- werkrevisor-codex-unattached:0 --> <!-- werkrevisor-codex --> <!-- werkrevisor-folded --> <details><summary>⚠️ werkrevisor ERROR for <code>?</code>, superseded by <a href="http://git.javagil.de/mi/werkrevisor/pulls/121#issuecomment-6404">the result for <code>a1cb668</code></a></summary> ## Codex review: critical The change adds third-party runtime code to the credential-bearing review process and a new route for sending PR diffs to an external model. Mistakes in secret handling or isolation can disclose sensitive data beyond what the mocked tests verify. ### Findings 1. **[high] Refetching the live diff bypasses the earlier secret scan** at `werkrevisor/cli.py:1266`, as a line comment The merge waits for the status `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. <sub> Model `gpt-6-astra`, effort high, tokens in 536360 (cached 476672) / out 2304, 78.1 s, 16 commands; reviewed head `b232092301d0712bc541e40230fc12efb4b91625`. </sub> </details>
claude-code added 1 commit 2026-10-18 20:50:06 +02:00
Diagrams from the scanned diff only; a refused upload keeps the diagrams Gitea took
werkator build succeeded after 00:11
Details
werkator/werkrevisor-statistics build succeeded after 00:26
Details
ai-review/assessment human review required; a human approval unlocks the merge; a human review request is pending; merge already scheduled
Details
ai-review/gate build succeeded after 05:28
Details
ai-review/codex 1 open Codex finding(s); resolve each line comment
a1cb668dc4 
Found by the Codex review of e716ea6 and the werkrevisor review of b232092.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
ai-reviewer commented 2026-10-18 20:55:44 +02:00
Collaborator
Copy Link
Copy Source

🟡 werkrevisor: YELLOW / HUMAN_REVIEW

Change kind: mixed.

Reasons

  • YELLOW risky build changes according to Jev: .dockerignore (weakens checks 0.11, risk 1.55, security 0.02), Dockerfile (weakens checks 0.10, risk 1.89, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more
  • hint 13 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev
  • hint 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.03, risk 0.00, security 0.01)
  • hint 25 changed file(s) agree with the PR document according to Jev
  • hint 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py
  • hint 3 changed files were given to the AI only as diff, not in full, so it could not check them against the rest of their file (all full files together would exceed [review] max_full_files_chars = 160000, and no excerpt fits the limits either): werkrevisor/pr_lens.mjs, werkrevisor/prlens.py, werkrevisor/report.py (advisory: Jev judged every changed file)
  • hint files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file)

Classifier

16 of 16 units passed all questions (1 PR document(s)).
5 answers ignored because of a confidence below [classifier] min_confidence.

Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 80033, calls 1, tokens in 77193 / out 28781 (reasoning included in out), duration 325.9 s, cost 0.0577 USD (reported), classifier `jev-1.13.0` calls 16, tokens in 13926, cost 0.0006 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 4, tokens in 36001 / out 1049, cost 0.0047 USD (reported), PR document check `jev-1.13.0` calls 25, tokens in 130877, cost 0.0055 USD, build file judge `jev-1.13.0` calls 18, tokens in 21328, cost 0.0009 USD, phases: path_judge 1.3 s, classifier 1.0 s, doc_check 1.7 s, context 19.8 s, model_review 301.3 s, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).
Reviewed head `a1cb668dc4434beff137a30b342d80223e2c7252` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0.
<!-- werkrevisor:a1cb668dc4434beff137a30b342d80223e2c7252 --> <!-- werkrevisor-verdict:YELLOW --> <!-- werkrevisor-blocks:6a5dd6d15a75ce527b81c68c2818c813,d08f070eb191140febbed9f8337e84c4,36931004656593df9fa3e3ebc963e9d2,358dfaf22bbde637138d66e4086d5817,fd005bda47e317b42d39a976c2a11daf,cfb5b10cb3cf703bf6ed41383d265d22,23c03ab2929fa10aa09d24ae9bce8090,97ff1748d79afbd3818e3a037f781eeb,db20165e1d03111a86c2c21f791b17ff,c4a55b1253ce650869e43668f1ca33a8,86b61de5e5dd9a3913e228b031e6b815,c7426376be99af7fb99445a20381c4e3,27a941d5e6c8f8b226548d1b038b45f2,83cf952851e2a6f8031616af674d1a53,5bdab77a421047471ebd4aece905c38c,03196a062679ab178fd54ab43daf0474,61f9f80cd1e1fcda02b28d6474c9c8ce,5fdc271260963c60959facd73a6b555a,768b1b55a99fb5326c78479d06997b98,0bcba3f2d2e7b4080a0d510eb3d210dd,8384e0d7ae67a5ad2ace70174fb71d12,da2e8d79aa0010083b40f6a4d2f76dba,0eb9efb660d6fa04ee38f2c8ecf9b419,787503c8d9ba5e44f388d566b30d3849,1513afe6c5523825ba9d22cf2dd52dc1,7cb5e52706b3ff04396ac03c416cc09a,91f33c237e1db63164dfb39fb44f4c9c,4ae0f702d1e3515bbbe58f96db668098,438e9746e9b295dfe8a3685c48935f4f,7d7bb43f72075a5e171dc02dcf23780d,373d148295aa718cde37a9d93f8cca3f,7822582c202c2521fea3fd667c436a31,d0a25d8c07e95d646d21bec9d51577fa,a6ebd6850f5ced07f5ce1f811a44831c,e1529479c46d63b7947586c7a37ba709,28b4c5d2aad07665bce904a681cbfdd4,3c60cbdc4dd48e2a5b08e7d9de4865c5,47d792ef5e5d855f89c030da011a799b,575088c5a9a867b44a5f854105399a33,924fe5dc830391b398ff7b9fc2d8bd22,8f3239045b94d86ba67cd4a7f53d0bae,608881546380d8db17664576d17a2482,49e326fe0f8d8d4e27312e547b6b414f,7fa1cdc645f02745f6fb26abe312e75e,9c9b688f26e5a3b3b44eb0f5336814f4,181205433044243e622f2d8e9e0a9d7f,43580384cba6f832d8871d649d923501,d8f5a31f10a8fe7db09536973ec42348,ef2dd26a0d22dd31e9de9c71c24e43cb,c272aa0bbf617c63f736cc544a985f0e,12a0e07c4646558ca0e39445bae26427,9d54c899562ad34810f0c2118a0fff42,2086857163b30bf27a8311ca587e8939,5fd168200f4a61084cb520b65391233c,ace9bd022e0ffd6415788fae898bd5d0,36461bb4f9ef83927d1626224cfb8c48,e8a0e2482fd99a2898e9d7e3f6c698da,fd849b2a094720de035c6645665938cd,e6c0b9a6aecd73a81379140cb953995e,bddbf4148d3c48a87349acbb8f048c8d,ee659a7a8a8168f1cd8b768f6eb3a784,2a526341ae388b9a485a392b293e25cb,351fc8c9957d07c225622497a4ae8135,846f9429cc2b163fd4c6016cc4804887,7acab543a4c8952784ef552794aa3396,720d305fea067e7778f51287b7e43d99 --> ## 🟡 werkrevisor: YELLOW / HUMAN_REVIEW Change kind: `mixed`. ### Reasons - **YELLOW** risky build changes according to Jev: .dockerignore (weakens checks 0.11, risk 1.55, security 0.02), Dockerfile (weakens checks 0.10, risk 1.89, security 0.04), tools/build-rootfs (weakens checks 0.05, risk 2.00, security 0.02) and 1 more - **hint** 13 changed file(s) that neither weaken the checks, touch security nor show a risk category of the repository config according to Jev - **hint** 1 build change(s) of low risk according to Jev: statistics/build.gradle.kts (weakens checks 0.03, risk 0.00, security 0.01) - **hint** 25 changed file(s) agree with the PR document according to Jev - **hint** 1 changed file was given to the AI as excerpt of the selected lines and 5 lines around each changed line, not in full (each larger than [review] max_full_file_chars = 80000): werkrevisor/cli.py - **hint** 3 changed files were given to the AI only as diff, not in full, so it could not check them against the rest of their file (all full files together would exceed [review] max_full_files_chars = 160000, and no excerpt fits the limits either): werkrevisor/pr_lens.mjs, werkrevisor/prlens.py, werkrevisor/report.py (advisory: Jev judged every changed file) - **hint** files the model cannot review: docs/werkrevisor-review-inside-dark.svg, docs/werkrevisor-review-inside.svg, tools/pr-lens/package-lock.json (advisory: Jev judged every changed file) ### Classifier 16 of 16 units passed all questions (1 PR document(s)). 5 answers ignored because of a confidence below [classifier] min_confidence. <sub> Usage: model `openrouter/minimax/minimax-m3`, LLM done, tokens counted before the call 80033, calls 1, tokens in 77193 / out 28781 (reasoning included in out), duration 325.9 s, cost 0.0577 USD (reported), classifier `jev-1.13.0` calls 16, tokens in 13926, cost 0.0006 USD, excerpt selection `openrouter/openai/gpt-6-luna` calls 4, tokens in 36001 / out 1049, cost 0.0047 USD (reported), PR document check `jev-1.13.0` calls 25, tokens in 130877, cost 0.0055 USD, build file judge `jev-1.13.0` calls 18, tokens in 21328, cost 0.0009 USD, phases: path_judge 1.3 s, classifier 1.0 s, doc_check 1.7 s, context 19.8 s, model_review 301.3 s, config: defaults + host + repository (.werkrevisor.toml at main a6185f0).<br> Reviewed head `a1cb668dc4434beff137a30b342d80223e2c7252` against base `a6185f094958d55dd8c59118a8413ad13d8c198a`, werkrevisor 0.51.0. </sub>
mi merged commit db2fba45e7 into main 2026-10-18 20:57:16 +02:00
mi deleted branch claude-code/pr-lens-diagrams 2026-10-18 20:57:16 +02:00
ai-reviewer left a comment
Collaborator
Copy Link
Copy Source

Codex review of a1cb668dc4434beff137a30b342d80223e2c7252

Codex review of `a1cb668dc4434beff137a30b342d80223e2c7252` <!-- werkrevisor-codex -->
werkrevisor/cli.py
@@ -1245,0 +1278,4 @@
file=sys.stderr)
break
if assets:
gitea.edit_issue_comment(comment["id"], body + prlens.section(assets))
ai-reviewer commented 2026-10-18 20:58:12 +02:00
Collaborator
Copy Link
Copy Source

[medium] Diagram attachment overwrites concurrent comment updates (Codex review)

The comment body is read before diagram generation and written back afterward. If a newer head finishes review meanwhile, _fold_earlier marks this comment as superseded, but this write restores the stale body and removes that notice and link. This was reproduced using the real folding and attachment functions. Re-read and preserve the current comment body after drawing.

Fix it, or reply why it does not hold, and resolve this comment.

**[medium] Diagram attachment overwrites concurrent comment updates** (Codex review) The comment body is read before diagram generation and written back afterward. If a newer head finishes review meanwhile, _fold_earlier marks this comment as superseded, but this write restores the stale body and removes that notice and link. This was reproduced using the real folding and attachment functions. Re-read and preserve the current comment body after drawing. Fix it, or reply why it does not hold, and resolve this comment. <!-- werkrevisor-codex -->
claude-code commented 2026-10-20 16:24:18 +02:00
Author
Collaborator
Copy Link
Copy Source

Holds; PR#121 was merged a minute before this comment. Fixed in PR#127 (786ca61): _attach_diagrams reads the body again after the drawing and puts the diagrams inside a fold, so the fold and its link stay. Scenario#127.01, red/green.

Holds; PR#121 was merged a minute before this comment. Fixed in PR#127 (`786ca61`): `_attach_diagrams` reads the body again after the drawing and puts the diagrams inside a fold, so the fold and its link stay. Scenario#127.01, red/green.
claude-code marked this conversation as resolved
ai-reviewer commented 2026-10-18 20:58:12 +02:00
Collaborator
Copy Link
Copy Source

Codex review: critical

The change introduces third-party executable code into the credential-bearing review process and a new outbound path for PR diffs. Mistakes at this boundary could expose credentials or private code beyond what the mocked tests verify.

Findings

  1. [medium] Diagram attachment overwrites concurrent comment updates at werkrevisor/cli.py:1281, as a line comment

The merge waits for the status 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.

Model `gpt-6-astra`, effort high, tokens in 556495 (cached 484352) / out 3126, 104.1 s, 14 commands; reviewed head `a1cb668dc4434beff137a30b342d80223e2c7252`.
<!-- werkrevisor-codex:a1cb668dc4434beff137a30b342d80223e2c7252 --> <!-- werkrevisor-codex-critical:true --> <!-- werkrevisor-codex-unattached:0 --> <!-- werkrevisor-codex --> ## Codex review: critical The change introduces third-party executable code into the credential-bearing review process and a new outbound path for PR diffs. Mistakes at this boundary could expose credentials or private code beyond what the mocked tests verify. ### Findings 1. **[medium] Diagram attachment overwrites concurrent comment updates** at `werkrevisor/cli.py:1281`, as a line comment The merge waits for the status `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. <sub> Model `gpt-6-astra`, effort high, tokens in 556495 (cached 484352) / out 3126, 104.1 s, 14 commands; reviewed head `a1cb668dc4434beff137a30b342d80223e2c7252`. </sub>
Sign in to join this conversation.
Reviewers
No Reviewers
mi
ai-reviewer
Labels
No items
No labels
Milestone
No items
No Milestone
Projects
Clear projects
No projects
Assignees
Clear assignees
ai-reviewer claude-code mi
No Assignees
2 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: mi/werkrevisor#121