[kyle-sexton]

Refs: #6916

Summary

Delivers the additive plugins/planning part of the repo-mode /multi-agent:audit-defaults sweep: section anchors, as-of dates and recheck triggers for claims that had none, and judgment labels where no upstream page sets the value.

Fix

• plan and draft-goal-condition: pointers to agent-teams#avoid-file-conflicts, workflows#when-to-use-a-workflow and sub-agents#available-tools; the maxTurns aside now cites the record in agents/plan-reviewer.md, which gains the section anchor.
• interview/context/session-config.md: the advisor default is labeled this skill's judgment, and the baseline pointer uses the current #combine-models anchor.
• upload-reader: the model and effort pin is labeled this repository's choice.

No edit changes a recommendation or a step. These edits cut no rule, so no eval evidence applies.

Two restatements in draft-goal-condition stay: removing them would be a skill-body cut, which needs eval evidence this run cannot produce. They are tracked for #6916.

Verification

• Each claim was read against the raw upstream pages on 2026-10-31.
• scripts/check-affected.sh --tests: 14 passed, 2 skipped.
• Left out: claude plugin eval, because this run spawns none and no edit here is a cut.

Related

• Refs docs(multi-agent,session-flow): fix repo-mode audit pilot findings and schedule a full sweep #6916 — sweep issue; one PR per plugin.

Remaining

• /multi-agent:audit-defaults repo over every file in its target list, with the count comment on docs(multi-agent,session-flow): fix repo-mode audit pilot findings and schedule a full sweep #6916.
• ci-status green on each sweep PR.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-31T16:35:38.343241Z	f8ffe2d	Draft marked ready

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" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[github-actions]

claude-security-review reviewed f8ffe2d686144a899d4aaca5b68f79313c563396: 0 finding(s), 0 as inline comments.

No security findings.

critical-findings: 0

• Scope: I read all 6 changed files in the full diff (.claude-lane/pr.diff) and the PR context. They are one .changes entry and five Markdown files under plugins/planning: agents/plan-reviewer.md, agents/upload-reader.md, skills/plan/SKILL.md, skills/draft-goal-condition/SKILL.md and skills/interview/context/session-config.md.
• Trusted paths: none of the changed files matches a TRUSTED PATHS: glob. No workflow, action, script, CODEOWNERS, .gitmodules or pr-pipeline.yaml is touched, so this change widens no trust control.
• Plugin manifests: no plugin.json or marketplace.json changed, so there are no manifest <field> lines.
• Instruction-surface lens: I checked the skill and agent body edits for a deleted, narrowed or softened rule. I did not read docs/contracts/instruction-exception-register/README.md, so the protected-class list was not consulted and I judged each edit by consequence instead.
  • Most edits only add section anchors, as-of dates and recheck triggers, or relabel a default as this repository's judgment.
  • In plan/SKILL.md step 3, "older clients do not mark a maxTurns stop" became "whether a maxTurns stop is marked depends on the client", plus a pointer to the record. The rule that any return not ending in ### Summary counts, or the literal No plan gaps found., is incomplete and never a clean pass is unchanged.
  • In upload-reader.md, the opus/medium pin is reworded as this repository's choice. The Read-only cage and the framing that stops injected text being obeyed are untouched, and the tools restriction is not edited.
  • session-config.md only adds the "this skill's own judgment" label and swaps the baseline anchor to #combine-models.
• Spend: no change to turn limits, models, efforts or schedules, so there is nothing to note.

[github-actions]

claude-review reviewed f8ffe2d686144a899d4aaca5b68f79313c563396: 0 finding(s), 0 as inline comments.

No blocking findings. I read the whole diff (6 files, docs and prompt text only) and the PR context.

• Scope: every edit adds a section anchor, an as-of date, a recheck trigger or a judgment label. I found no change to a step, a recommendation or any executable code, and no manifest, userConfig or hook change.
• Changelog fragment: bump: patch fits a docs-only fix.
• Internal consistency: plan/SKILL.md Step 3 now points at "the maxTurns record in agents/plan-reviewer.md". That record exists in the file (the **Verification.** Claim: maxTurns… paragraph) and covers the v2.1.246 partial-marking requirement, so the reference resolves.
• Not verified: I did not check the new upstream anchors (#supported-frontmatter-fields, #available-tools, #when-to-use-a-workflow, #avoid-file-conflicts, #combine-models) against the live pages. The PR says they were read on 2026-10-31, which is the author's claim and not something I confirmed. The existing recheck triggers cover them if a section is renamed.
• Test results: I did not run scripts/check-affected.sh --tests. The "14 passed, 2 skipped" result is the author's claim.
• Spend: none. No turn limits, models, schedules or lane runs change.

Trusted-path matches: none of the changed files matches the listed globs.