FAF Skills
Claude starts every session knowing your project. These skills help you build a project.faf: one small file with your project's goal, stack and the who, what and why, scored 0–100% so you can see exactly what's missing. Built on the IANA-registered .faf format (application/vnd.faf+yaml). The set also includes skills for risk-tiered test plans and repo health checks.
⭐ Star it on GitHub: bookmarks it for you, helps other devs find it too.
# Plugin marketplace
/plugin marketplace add Wolfe-Jam/faf-skills
/plugin install faf@faf-skills
# Or manual
git clone https://github.com/Wolfe-Jam/faf-skills.git && cp -r faf-skills/skills/* ~/.claude/skills/
The skills
| Skill | What it does |
|---|
faf-context — /faf-context | The quickstart: a sharp goal plus the who, what, why, where, when and how, to 100% ✪ fast |
faf-go — /faf-go | A guided interview (AskUserQuestion) for the slots only you can fill |
faf-wizard — /faf-wizard | Done-for-you project.faf generator |
faf-expert — /faf-expert | The reference: how scoring works, MCP setup, sync, and the 33 slots every .faf is scored on |
wjttc-builder — /wjttc-builder | Plans and generates WJTTC test suites, tiered by risk (Brake · Engine · Aero · Tyre · Pit) |
wjttc-tester — /wjttc-tester | Runs a test plan, reproduces bugs, and files a tiered report |
repo-maintainer — /repo-maintainer | Multi-phase repository health audit |
Works with CLAUDE.md
Claude Code reads CLAUDE.md for your project instructions. project.faf doesn't replace it. It's the structured source behind it:
- Scored. A 0–100% score shows exactly which facts about your project are missing.
- Kept in step.
faf sync writes CLAUDE.md from it, so the two stay in step.
- Portable. The same file writes
AGENTS.md, .cursorrules, GEMINI.md and Copilot instructions (faf export), so every AI tool gets the same facts.
- Yours. Plain YAML in your repo: read it, diff it, commit it.
Write your instructions in CLAUDE.md. Keep the facts about your project in project.faf.
Try it
Ask Claude in any project:
- "Create a project.faf for this repo and score it." →
faf-context / faf-wizard
- "Interview me for the slots you can't fill, until we hit 100%." →
faf-go
- "Plan a tiered test suite for this repo." →
wjttc-builder
Receipt: agents-md-facts went from 52% → 100% ✪ (PR #1) by following exactly the model faf-context teaches: faf auto detected and seeded the facts and marked slotignored the 11 slots a single-package CLI doesn't use; the one gap it couldn't source was filled from the repo (bun.lock → pkg_manager: bun). Re-scored: 100%. The score is deterministic: run npx faf-cli score project.faf in that repo and you get the same number.
Every command, flag and tool a skill names is checked against the live faf-cli and claude-faf-mcp before it ships.
Tier System
| Score | Tier | Symbol |
|---|
| 100% | Trophy | ✪ |
| 99% | Gold | ★ |
| 95% | Silver | ◆ |
| 85% | Bronze | ◇ |
| 70% | Green | ● |
| 55% | Yellow | ● |
| 1% | Red | ○ |
| 0% | White | ♡ |
✪ Trophy = 100%: AI is optimized to code. Every definition slot for your app type is filled, so AI has the truth about your stack and your intent.
Prerequisites
npm install -g faf-cli # required (or: brew install wolfe-jam/faf/faf-cli)
claude mcp add faf -- npx -y [email protected] # optional: MCP server
What this plugin runs
The plugin ships skills only: no hooks, no MCP servers, nothing runs on install. The skills ask Claude to run local faf commands, which read and write files in your project (project.faf, CLAUDE.md, AGENTS.md). Network use: installing faf-cli or running npx [email protected] downloads from npm; faf git <url> shallow-clones a public GitHub repo to a temp folder; faf bench --submit (opt-in) posts a score receipt to a public ledger. The plugin itself collects no data.
Support: GitHub issues · [email protected]
skills.json is generated from the SKILL.md frontmatters and verified in CI — never edit it by hand (node scripts/build-skills-json.mjs).
Resources
★ bookmark repo: github.com/Wolfe-Jam/faf-skills
License
MIT License
By @Wolfe-Jam · curated, not collected · proof over promises.
