claude-code

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 23:29:43 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 11, fixed in eb62556:

werkrevisor/earlierreview.py:68 model-written headings (Codex) and werkrevisor/reviewreport.py:41 Prompt Injection (and the security concern):…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 23:29:43 +02:00

eb625565e7 Review round 11: single-line fields of the reports stay in one line

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 23:19:42 +02:00

d22415a769 Review round 10: the assessment keeps its subheadings; only the top marker folds a report

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 23:19:41 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 10, fixed in d22415a:

werkrevisor/earlierreview.py:73 subheadings in the assessment (Codex) and earlierreview.py:38 Truncated Assessment: the assessment, the last section,…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 23:11:45 +02:00

51e8aaa280 Review round 9: the earlier verdicts and values of Jev stay out of the note

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 23:11:44 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 9: werkrevisor/earlierreview.py:146 Jev Reuse, fixed in 51e8aaa; latest drops the verdict line of each earlier finding, e.g. "ORANGE; Jev: real 0.66, severity 2.2 of 3", so the…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 23:01:45 +02:00

7d90ff549d Review round 8: the earlier review is read by the order of its sections

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 23:01:44 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 8, fixed in 7d90ff5:

werkrevisor/earlierreview.py:73 quoted Codex text as assessment (Codex, twice) and earlierreview.py:70 Misparsed heading (werkrevisor): latest reads…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 22:50:52 +02:00

f71dbfaae1 Review round 7: Scenario#142.03 names the behaviour

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 22:50:51 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 7:

YELLOW Scenario#142.03 implementation_details: fixed in f71dbfa; the scenario names what the PR author writes and what the model gets, not the block lines and test details. -…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 22:31:56 +02:00

0c15d5413f Review round 6: no note with a possible secret, no folded report, Codex findings count

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 22:31:56 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 6:

werkrevisor/earlierreview.py:113 Secret Exposure (and the security concern): fixed in 0c15d54; a match of SECRET_PATTERNS in the earlier findings, Codex findings or…

claude-code pushed to claude-code/model-facts at mi/werkrevisor

2026-10-23 22:30:03 +02:00

6f047e17a4 Review round 5: the stack only informs the model; the verdict stays as on main

claude-code commented on pull request mi/werkrevisor#141

2026-10-23 22:30:02 +02:00

werkrevisor 0.72.0: the model learns the base branch and which changesets are new

Review round 5 (Codex), fixed in 6f047e1 by a change of design, chosen by the human:

werkrevisor/cli.py:427 outgoing PRs: the stack no longer changes the verdict. A changed or removed…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 21:45:19 +02:00

37633af64c Review round 5: a newer review without findings replaces the older one

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 21:45:17 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 5:

werkrevisor/earlierreview.py:65 newest review without findings (Codex): fixed in 37633af; the newest review counts, and without findings it gives no note. Red/green:…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 21:37:40 +02:00

2f7f240b28 Review round 4: the first findings heading counts; tests for the newest review and quoted headings

claude-code commented on pull request mi/werkrevisor#142

2026-10-23 21:37:38 +02:00

werkrevisor 0.69.0: the model judges the findings of the last posted review

Review round 4:

werkrevisor/earlierreview.py:62 Misparsed Findings (and the security concern): does not hold for quoted text. report.render_comment and the hs-review report indent every…

claude-code pushed to claude-code/previous-findings at mi/werkrevisor

2026-10-23 21:19:56 +02:00

1bda450071 Review round 3: the earlier review goes into the user prompt, not the system prompt

e9c3490165 Review round 2: no commit subjects in the note, only their number

31accd0388 Review round 1: commit subjects stay inside the data block of the earlier review

14479e6a21 Shorter test names, so that the PR document keeps its line length

81c19521c2 README, Werkbaum, version 0.69.0 and the solution of the earlier review

Compare 9 commits »

claude-code pushed to claude-code/model-facts at mi/werkrevisor

2026-10-23 21:19:05 +02:00

ed0a05c7d8 Review round 4: a protected branch with an open PR gives no stack

d69e44d1ef Review round 3: a branch with two targets gives no stack; duplicate ids are no released change

3fdb1770a3 Review round 2: only the head branch of an open PR makes a stack

95f40e2af2 Review round 1: a repeatable changeset missing on the default branch is not released either

c61d5d0213 README, Werkbaum, version 0.68.0 and the solution of the model facts

Compare 9 commits »