chore(deps): update agent skills (plugins/agentic-engineering/skills/agent-improvement)

[botantler-1]

Automated update of the agent skill at plugins/agentic-engineering/skills/agent-improvement to its latest version.

Each changed skill gets its own pull request, so one blocked update does not hold back the others.

[devantler]

🤖 Generated by the Agentic Engineer

@coderabbitai full review

[coderabbitai]

⚠️ Action not completed

Review skipped.

[coderabbitai]

Important

Review skipped

This PR was authored by a bot without an assigned CodeRabbit review seat. Ask an organization administrator to assign a seat to the bot. See the seat-assignment guide. After the seat is assigned, request a review again.

⚙️ Run configuration

• Configuration used: Organization UI
• Review profile: ASSERTIVE
• Plan: Advanced
• Run ID: 760ac38b-8f99-404f-86f0-b02d1e482e54

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

• Autopilot · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}

[devantler]

🤖 Generated by the Agentic Engineer

Self-review (fallback — CodeRabbit, Codex and Cursor Bugbot unavailable)

Reviewed commit: d6d29010275594af22bea7a9507cd75ea9b0f97f

• CodeRabbit: declined the disclosed request at this head at 2026-10-05T06:12Z ("Review skipped" — the PR author is a bot without a review seat).
• Codex: usage limit since 2026-10-05T05:05Z (lane health read 06:05Z).
• Cursor Bugbot: usage limit since 2026-10-05T05:09Z (lane health read 06:05Z); no check-run at this head.

What was checked

• Provenance. The updater classifier returns the trusted "requires semantic review" state for this head, so this round is the review it asks for.
• The synced skill equals its upstream tag. agent-skills refs/tags/v1.21.29 resolves to the tree recorded in the skill's metadata (c5d8b763). Eight of the nine files under the skill are blob-identical to that tree. The ninth, SKILL.md, differs only in its front matter (the installer's provenance metadata and a re-flowed description); the body is the same text.
• The recorded digest matches. skillSha256 in the desired-state resource equals the SHA-256 of SKILL.md at this head (19eeac53…).
• The upstream range is small and reviewed. v1.21.27 → v1.21.29 is three verified commits; only one touches this skill (agent-skills#247, one line in scripts/measure-flow.jq).
• The logic change is sound. The uncertainty test now ignores the selected item and judges only the alternatives. The selected item was already excluded from the "oldest unstarted" result, so its unknown state could never change that result; it could only turn a known answer into UNKNOWN. An alternative with unknown state, and incomplete candidate evidence, still yield UNKNOWN, so the fail-closed behaviour for the inputs that matter is kept.
• Version bookkeeping is consistent. 6.1.10 → 6.1.11 in both marketplace files and both plugin manifests, with a matching changelog entry.

Verdict: no P0/P1 findings