Migrate Claude config from subagents to skills

[plx]

This PR modernizes the Claude Code setup by making CLAUDE.md a lightweight include of AGENTS.md, with shared global guidance moved there. It factors persistent instructions into modular .claude/rules files, including a path-scoped content-collections rule to reduce unnecessary always-on context. It migrates legacy .claude/agents task prompts into .claude/skills/*/SKILL.md for accessibility auditing, accessibility implementation, lint fixing, and Playwright QA workflows, and removes the obsolete agent files.

[claude]

Code Review: PR #39 — Migrate Claude config from subagents to skills

Overall: Clean, well-structured modernization. Approve with minor observations.

---

What this PR does

Replaces a monolithic CLAUDE.md with a lightweight indirection chain:

• CLAUDE.md → @AGENTS.md (single-line redirect)
• AGENTS.md holds shared project summary + @ imports of modular .claude/rules/ files
• Four .claude/agents/*.md task prompts converted to .claude/skills/*/SKILL.md format

Net result: −334 / +219 lines, substantially simpler.

---

Strengths

• Modular rules files (00-project-basics, 10-commands-and-validation, etc.) are easier to maintain independently than a single large file.
• Path-scoped loading in 20-content-collections.md (the paths: frontmatter block) is smart — it avoids injecting content-collection rules into every conversation, only loading them when relevant files are in scope.
• Skills are tighter than agents: the old agent files had substantial "you are an expert who..." persona prose. The new skills skip the preamble and go straight to checklists and output format — more signal, less noise.
• disable-model-invocation: true on lint-fixer and web-qa-playwright is an appropriate efficiency flag for deterministic tool-use workflows.
• argument-hint metadata on skills is good UX.

---

Observations / Minor Concerns

1. @ imports in AGENTS.md for non-Claude tools: The .claude/rules/ files are included via @ imports inside AGENTS.md. Claude Code will expand these; tools that read AGENTS.md natively without @-import support (e.g., Codex) will see the directives literally. The "For tools that support @ imports:" header acknowledges this, and the AGENTS.md summary is sufficient for non-Claude tools, so this is a deliberate design choice — worth a brief inline note if this is expected to be a multi-tool repo.

2. tests/ directory reference in 00-project-basics.md: The new rules file lists tests/ as a Playwright test suite location, but the previous CLAUDE.md didn't mention this directory. If tests/ doesn't exist yet, this is forward-looking documentation rather than current state — fine, but worth confirming it matches reality.

3. CLAUDE.md legibility on GitHub: The file now contains only @AGENTS.md. Visitors browsing the repo on GitHub will see a one-liner with no rendered content, since GitHub doesn't expand @ imports. Low priority, but a one-sentence HTML comment explaining the redirect would help human readers without affecting Claude Code behavior.

---

No issues found in

• Security: no user-facing code changed.
• Performance: static config files only.
• Test coverage: N/A (tooling config, not application code).
• Lint/style: rule files are concise and consistently formatted.

---

Verdict: The migration is correct, the skills are well-scoped, and the modular structure is an improvement. The observations above are all minor — none block merging.

[github-actions]

PR Validation ✅

All checks passed! Ready for review.

Checks Performed:

✓ Linting
✓ Spell check (source)
✓ Type checking & Build
✓ Spell check (HTML)
✓ Internal link validation
✓ Artifact upload

This is a complete dry-run of the deployment process, ensuring your changes will deploy successfully when merged.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 334ee33387

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".