fix(ce-noslop): keep the edit-mode change summary out of user-facing text - #1700
Conversation
…text Edit mode returned the text plus one line saying what changed. Skills run in the caller's context, so that line landed in the user's message or in artifacts like PR bodies. The summary is now produced only when the caller asks and stays separate from text that reaches a user or an artifact. Detect mode is unchanged. Records the general pattern as a skill-design learning and refines the Handoff seam glossary entry. Claude-Session: https://claude.ai/code/session_018A2yNqHr1YhjW4ZY5G2gUC
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: dc16d29782
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".
Review restated two conditions. The change line, when asked for, stays outside the rewritten text and out of artifacts rather than being kept from the requester. The non-English notice rides the same channel: it appears only inside detect findings or an asked-for change line. Claude-Session: https://claude.ai/code/session_018A2yNqHr1YhjW4ZY5G2gUC
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: a08bddba09
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".
…ing standard The rule now fires from portable-agent-skill-authoring.md, which ce-skill-work requires, with the learning kept as the worked case. The learning's guidance matches the revised skill: a requester who asked receives the line, separately from the text. Claude-Session: https://claude.ai/code/session_018A2yNqHr1YhjW4ZY5G2gUC
1 participant
Summary
Prose that a skill routes through
ce-noslopedit mode no longer arrives with a line of drafting narration in front of it. Edit mode returned the rewritten text plus one line saying what changed; skills run in the caller's context, so that line went wherever the caller printed next. A livelfgrun that routed a judgment throughce-povopened its message to the user with "Edit-mode pass applied: split two long sentences, ...". On a PR-description consumer the same line lands in the PR body.The fix is at the callee: edit mode now says what changed only when the caller asks for it, and keeps that line outside the rewritten text and out of any artifact, so a requester who asked still receives it. The non-English notice rides the same channel. Detect mode is unchanged. The general rule is recorded in the skill-authoring standard with a linked worked case, and the Handoff seam glossary entry gains the sentence that explains it.
Validation
Fresh-agent cells (
bun run test:skill-eval-cell) on Claude and Codex, pre-change (HEAD) and post-change:A detect-mode cell listed findings on both hosts in both arms. The full suite passes, and the SKILL.md body is 4,033 bytes CRLF-adjusted against the 4,096-byte kernel limit, which leaves little headroom for a later edit.
Review rounds
Codex's two findings on the skill lines were condition-level and were restated rather than patched: the change line stays outside the payload rather than away from the requester, and the non-English notice is scoped to detect findings or an asked-for change line. Its two findings on the learning doc moved the rule into
portable-agent-skill-authoring.mdand aligned the doc's wording with the revised skill.Security Disclosure
No security-relevant changes.
Agent Disclosure
Claude Code · claude-fable-5-1https://claude.ai/code/session_018A2yNqHr1YhjW4ZY5G2gUC