FE-1533: Add a petrinaut-shipping skill to the repo skills

[kube]

🌟 What is the purpose of this PR?

Ships the Petrinaut shipping procedure as a repo skill so every agent session is offered it, following the skills layout from #9323. Stacked on #9391; the standing conventions the skill applies land in #9390.

🔗 Related links

• FE-1533 (internal)

🔍 What does this change?

• Adds the petrinaut-shipping skill under .agents/skills/: the per-package gate commands, the changeset step, docs-sync pointers, pre-PR hygiene checks, and the draft-to-ready CI loop. Standing conventions stay in the package AGENTS.md and are referenced, not restated.
• Adds the .claude/skills/petrinaut-shipping symlink and regenerates skill-rules.json with yarn agents:skill-management generate-skill-rules.

Review fixes:

• Keywords are down to petrinaut. changeset fuzzy-matched unrelated prompts through the hook's Levenshtein pass, so any monorepo changeset prompt pulled in this Petrinaut-only skill.
• Dropped the files: block: the activation hook reads promptTriggers and never fileTriggers, so those globs did nothing.

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

• does not modify any publishable blocks or libraries, or modifications do not need publishing

📜 Does this require a change to the docs?

The changes in this PR:

• are internal and do not require a docs change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

• do not affect the execution graph

🛡 What tests cover this?

• yarn agents:skill-management validate passes with the new skill.
• yarn workspace @local/claude-hooks test:unit passes against the regenerated rules.

❓ How to test this?

1. yarn agents:skill-management validate
2. echo '{"session_id":"t","transcript_path":"/dev/null","cwd":".","permission_mode":"default","prompt":"fix a petrinaut lint failure and open a PR"}' | yarn workspace @local/claude-hooks run:skill
3. Confirm petrinaut-shipping appears under the recommended skills.

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
hash	Ready	Preview	Sep 1, 2026 8:38am UTC
petrinaut	Ready	Preview	Sep 1, 2026 8:38am UTC
petrinaut-docs	Ready	Preview	Sep 1, 2026 8:38am UTC

1 Skipped Deployment

Project	Deployment	Actions	Updated
hashdotdesign-tokens	Ignored	Preview	Sep 1, 2026 8:38am UTC

[cursor]

PR Summary

Low Risk
Documentation and agent-hook configuration only; no runtime product or publishable library changes.

Overview
Adds a petrinaut-shipping agent skill so sessions that touch Petrinaut work get a suggested end-to-end ship checklist: per-package turbo/format/pytest gates, changeset and docs-sync pointers (conventions stay in package AGENTS.md), pre-PR hygiene, and draft-to-ready CI guidance including a gh pr checks JSON polling loop.

Wires it into the existing skills layout with .agents/skills/petrinaut-shipping/SKILL.md, a .claude/skills/petrinaut-shipping symlink, and a regenerated skill-rules.json entry (high-priority suggest, keyword petrinaut, intent patterns for implement/fix/PR/CI prompts; no file globs).

^{Reviewed by Cursor Bugbot for commit 9990f1a. Bugbot is set up for automated code reviews on this repo. Configure here.}

[lunelson]

Cool. The one thing that's odd here to me, although I know it's a pre-existing convention in this repo, is this non-standard frontmatter on the SKILL.md:

metadata:
  triggers:
    type: domain
    enforcement: suggest
    ...

The Agent Skills spec defines metadata as a string-to-string map, so a nested triggers object is non-conforming. A few concerns:

• It creates two routing systems: the description-based model routing and HASH’s keyword/regex hook.
• Several fields are still decorative anyway: the Claude hook reads prompt triggers and priority, but not fileTriggers, enforcement, blockMessage, or skipConditions.
• The broad keywords are already causing trouble: Cursor correctly noted that changeset fuzzy-matches unrelated prompts.
• Putting Claude-specific machinery in canonical .agents/skills somewhat undermines the supposed cross-agent portability.
• The generated JSON makes the same configuration appear in three places: description, trigger metadata, and generated rules.

Personally I would remove metadata.triggers and rely on a precise description. If Claude-only activation is genuinely required, we should maybe work out a way to keep that configuration in a Claude-specific sidecar; or else find a way to encode it as a namespaced string-valued metadata field so the skill remains spec-compliant.

[kube]

Sadly I cannot remove metadata.triggers, as it's required by our custom tooling. We need to check with Bilal and Tim for this.

For the record, yarn agents:skill-management validate (which CI runs as lint:skill) rejects a skill without it: metadata: Invalid input: expected object, received undefined. Frontmatter.metadata is required in libs/@local/repo-chores/node/scripts/skill-management/schemas.ts, and FrontmatterTriggers is a strictObject, so dropping the block is a schema change across all 14 skills rather than something this PR can do.

Your other points hold and I have acted on what is in scope here: the keywords are down to petrinaut (the changeset one was fuzzy-matching unrelated prompts, as Bugbot spotted), and the files: block is gone since the hook never reads fileTriggers.

[indietyp]

Note that this is a recent change, as evident by: https://github.com/agentskills/agentskills/pull/479/changes, we added these skills w/ metadata when it was first released, where it supported arbitrary key-value pairings, indeed these were all run through the validator during creation.

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 9990f1a. Configure here.}

[cursor]

CI wait loop exits too early

High Severity

The draft-PR wait loop treats a pending count of 0 as “CI finished”, but gh pr checks --json prints [] when no checks have been scheduled yet. jq then prints 0, so the loop breaks immediately after open. The comment that an absent result keeps waiting does not apply: that case yields 0, not an empty string.

 

^{Reviewed by Cursor Bugbot for commit 9990f1a. Configure here.}