Repository navigation
fix(ce-noslop): keep claim certainty and rule out semicolons as dash substitutes - #1847
Conversation
…substitutes A rewrite now fails when it changes how certain a claim is, not only when it drops a qualifier. The em dash rule says a semicolon is not a substitute. Four eval scenarios cover the dash rewrite, hedged claims, and longer drafts where an opinion and real hedges must survive.
Codex Review SummaryThis comment shows the latest Codex review activity on this pull request.
ℹ️ About Codex in GitHubYour team has set up Codex to review pull requests in this repo. Reviews are triggered when you
Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings. |
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0cf3bea58b
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
The certainty rows passed when a rewrite dropped every hedge but kept the numbers, because must_include also reads the summary line. A new result_must_include grade reads only the RESULT block, and the rows now require the hedged wording itself. The semicolon row is post-only, since the baseline rule 29 did not prohibit semicolons.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 20ab2a41af
ℹ️ About Codex in GitHub
Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
1 participant
Summary
ce-noslopnow treats a rewrite that changes how certain a claim is as a failure, not only one that drops a qualifier, and its em dash rule no longer allows a semicolon as the replacement. Both came from reviewing phuryn/work-humanizer for ideas. A third idea from that review was tried and removed because evals showed it made the skill drop claims.What was tried and removed
A "What a fix may do" paragraph in
references/patterns.mdsaid that fixes must keep certainty, must cut a claim (edit mode) or name the missing fact (detect mode) when a specific is missing, and must not leave clipped sentences or add personality. On longer drafts it made things worse:The "cut the claim" instruction fired on hedged opinions as well as puffery. The clipped-sentence guard did not prevent the one fragment stack. Detect mode already named missing evidence without inventing it (4/4 before the change).
Validation
fixes-leave-full-sentences,edit-keeps-claim-certainty,long-release-note-keeps-hedged-opinion,long-status-update-keeps-hedges, baselined at 67035e9. All 8 post-change cells pass on Claude and Codex with the paragraph removed.result_must_includegrade, which reads only the RESULT block. That way a summary line that names a kept hedge can't satisfy the check. The semicolon row is post-only, since the baseline rule 29 did not prohibit semicolons.bun run test,bun run typecheck, andbun run release:validatepass.Security Disclosure
No security-relevant changes.
Agent Disclosure
Claude Code · claude-opus-5-5