AI Guidelines & Skills (Repo-Specific)
This repo is package-boost, so it has a dual structure that downstream consumers don't:
.ai/skills/and.ai/guidelines/— internal. Drive the per-agent files for this repo only.resources/boost/skills/andresources/boost/guidelines/— shipped. Distributed to every package that installs package-boost.
Anything authored under resources/boost/ lands in arbitrary
downstream repos; anything under .ai/ stays here.
For general skill-authoring mechanics (frontmatter, naming,
namespacing across host / vendor / package-boost defaults, choosing
between .ai/skills/ and resources/boost/skills/), use the shipped
skill-authoring skill — it covers the full picture and is the
same guidance consumers see.
Never Edit Generated Files
CLAUDE.md, AGENTS.md, GEMINI.md, and the per-agent skill
directories — .claude/skills/, .cursor/skills/,
.agents/skills/, .github/skills/, .junie/skills/,
.kiro/skills/ — are auto-generated by package-boost:sync
from .ai/ + resources/boost/. Direct edits get overwritten on
the next sync.
If you're about to edit one of those, stop and find the source under
.ai/ (or resources/boost/ for shipped skills) instead.
Picking the Source Directory in This Repo
| Content | Path | Audience |
|---|---|---|
| Skill used only when working on this repo | .ai/skills/{name}/SKILL.md |
Internal contributors |
| Skill shipped to every package-boost consumer | resources/boost/skills/{name}/SKILL.md |
Downstream packages |
| Guideline used only here | .ai/guidelines/*.md or *.blade.php |
Internal contributors |
| Guideline shipped to every consumer | resources/boost/guidelines/*.md |
Downstream packages |
Most internal-only tooling skills (code-review, evaluate,
pre-release, etc.) belong in .ai/skills/. Reach for
resources/boost/skills/ only when consumers should receive the
content.
Authoring Guidelines
Edit files in .ai/guidelines/:
- Markdown files (
*.md): standard markdown. - Blade files (
*.blade.php): wrap code examples in@verbatimblocks so PHP / Blade syntax doesn't render.
Filename controls ordering (alphabetical across the source dir).
Sync Workflow
Always run package-boost:sync after editing any file under
.ai/ or resources/boost/. That regenerates the per-agent
files locally so you're working against current content.
vendor/bin/testbench package-boost:sync
Verify cleanly:
vendor/bin/testbench package-boost:sync --check
Exits non-zero when generated files diverge from sources.
Generated Files Are NOT Committed in This Repo
Important repo-specific exception, set by
.ai/guidelines/release-automation.md (always loaded into
CLAUDE.md): the generated outputs (CLAUDE.md, AGENTS.md,
GEMINI.md, every per-agent skill dir) are gitignored because
the sync-command tests under tests/ exercise the same filesystem
paths and would clobber any committed copies on every Pest run. So:
git statusshould show only the.ai/orresources/boost/source change. Generated dirs not appearing is correct.- Don't fight the gitignore — re-sync is run locally on demand.
- Downstream packages that consume package-boost work the opposite
way (they commit the generated files); the shipped
skill-authoringskill describes that flow for them.
Checklist
- Edit only files under
.ai/orresources/boost/— never the generated outputs. - If creating or renaming a skill, follow the shipped
skill-authoringskill (frontmatter, naming, namespacing). - Run
vendor/bin/testbench package-boost:sync. -
git statusshows the source change only — no generated dirs should appear (they're gitignored).
Source: SanderMuller/package-boost — distributed by TomeVault.