CC - Shared Claude Code tooling: skills, agents, hooks, and rules (6288)

[stracker-phil]

Changelog: (none - internal Claude tooling)

Description

🎯 The goal

Start a shared, version-controlled set of Claude Code configuration so AI assistance is consistent across the team instead of living in each developer's local setup.

This first batch establishes the structure and the initial tooling. The full inventory and how the pieces fit together lives in .claude/docs/README.md; in short:

• Skills - slash-commands you invoke: research an area (/understand), gate your branch before a PR (/pr-readiness), draft this description (/describe-pr), review a teammate's PR (/code-review), and run a PHP cleanliness pass (/php-review).
• Agents - focused workers the skills and hooks dispatch: PHP and JS test writers, a PHP cleanliness reviewer, a test/lint runner, a UI-copy editor, and a reader/writer for local research notes.
• Hooks - auto-run the PHP reviewer on the code you touch, then again once your turn settles. Opt out locally via .claude/hooks/config.local.json.
• Rules - path-scoped conventions that route PHP and JS test writing to the right agent.
• Shared standard - .claude/docs/code-quality.md, the single quality lens the reviewers apply.

🔍 What to review

• The .gitignore change carries the most weight. .claude/ is now tracked, while *.local.* (e.g. settings.local.json, CLAUDE.local.md) stays ignored so personal overrides are never committed. Confirm this split matches how the team wants to share config versus keep it local.
• The hooks execute locally. .claude/settings.json registers two shell scripts (php-review-track.sh, php-review-stop.sh) that run automatically. Teammates approve them once on first use (workspace trust), and they no-op without jq. Worth a sanity read since they run on your machine.
• Everything else is prompt and documentation content, with no effect on the plugin runtime, build, or shipped assets.