refactor(skills): narrow discovery descriptions

[Wibias]

Summary

Keep discovery metadata focused on ownership, triggers, and boundaries while preserving all executable GitHub Delivery policy in the skill bodies.

• remove ship-gate execution procedure from the top-level description
• narrow babysit, babysit-pr, and review-security override descriptions to compatibility routing only
• keep the actual watch/security procedures unchanged in each skill body
• add a deterministic frontmatter regression test that rejects executable workflow leakage from these discovery descriptions

Why

The installed description is part of the model's skill-discovery context. Procedure-level instructions such as MUST run scripts/... or always load ... and run ... belong after selection, not in the always-visible routing surface.

This is intentionally a discovery-only change. No delivery policy, mutation boundary, ship gate, watch behavior, or security-review procedure is weakened or removed.

TDD evidence

Before the rewrites, the new discovery-frontmatter test failed 4/4 cases on the current descriptions. With the narrowed descriptions, the same test passes 4/4 on the exact branch head.

The full local Windows repository check also caught one PR-caused formatting regression in the existing ship-gate docs contract: the canonical Watch MUST run scripts/ship-gate.mjs every wake policy remained in the SKILL body but was wrapped across a newline, while the old frontmatter had previously supplied the regex-compatible one-line occurrence. The fix keeps the policy in the body and restores the existing assertion without putting executable procedure back into discovery metadata.

Two additional local Windows failures are outside the changed paths and are treated as baseline/environmental gaps, not as passing evidence: verify-pr-head resolves its test fixture /tmp/repo to missing D:\tmp\repo, and the Windows trace-bin package test fails in its npm pack/install probe under the local npm runtime.

Changed paths

• SKILL.md
• overrides/babysit/SKILL.md
• overrides/babysit-pr/SKILL.md
• overrides/review-security/SKILL.md
• tests/discovery-frontmatter.test.mjs

Remaining gates

• rerun the focused ship-gate docs contract and discovery-frontmatter tests on the exact repaired branch head
• confirm the full diff remains limited to the five declared paths
• run the Agent Skill qualification/review flow where applicable
• sync the accepted descriptions into the installed .agents copy after this upstream change is accepted

Draft until those gates have concrete evidence.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 7993a3c8-4801-4432-9571-ff18246a1179

---

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.}