FAF-CLI for Bun Developers

August 1, 2026 · View on GitHub

One command. Every agent in your session gets the same accurate, versioned context.

FAF-CLI (v7.1, "The AGENTS.md Edition") is the canonical tool for the .faf format: persistent, versioned, AI-readable project context. It runs straight through Bun's package runner — bunx faf — detects your Bun stack, and emits the context files AI coding tools read. It's bundled with bun build, so it's built the way you build. Part of the FAF ecosystem — over 100k downloads; see faf.one/downloads for latest stats.

Quick start (zero-install)

bunx faf                     # auto-detect the stack, write project.faf, score it
bunx faf export --agents     # generate AGENTS.md agents actually read
bunx faf score               # AI-readiness score (target: Trophy 100%)
bunx faf git owner/repo      # instant, scored context for any remote repo — no clone

No bun add -g needed — bunx runs the latest faf-cli on demand. Add bunx faf export --agents to a package.json script or a pre-commit hook and context files stay fresh.

Why this fits Bun projects

FAF is Bun-aware end to end — it reads your project the way Bun sees it:

  1. Runs through bunx — zero-install, always the latest release, no global state to manage.
  2. Detects your Bun runtime — faf recognizes bunfig.toml and bun.lock / bun.lockb and reports Bun as the runtime and package manager, not a generic Node guess.
  3. Emits bun-based commands — when faf sees a Bun lockfile it prefers bun as the runner, so the authored AGENTS.md carries bun install / bun test / bun run — the commands your project actually uses.
  4. Same toolchain — faf-cli is bundled with bun build; you're running a tool built on the runtime you ship on.

Why context matters on a fast stack

Bun makes the run loop fast. FAF makes the agent's loop accurate — so the speed isn't spent re-discovering the same project every session:

  • CI / bootstrap gate — bunx faf export --agents on every run keeps AGENTS.md current, so an agent never starts from stale context.
  • Scoring as a readiness signal — bunx faf score answers "is this project ready for serious agent work?" before you point an agent at it. Target the Trophy (100%).
  • Non-destructive — FAF maintains a labeled block and preserves everything you wrote by hand.
  • Git-native — the .faf versions with your code, so context travels with the branch.

For general AI coding sessions

  • Static + live — the emitted Markdown is the baseline; wire a FAF MCP server so agents can call score, validate, and context ops, not just read files.
  • One source → many surfaces — change project.faf, and every emitted file (AGENTS.md, .cursorrules, GEMINI.md, CLAUDE.md via faf sync) stays consistent.
  • External work — bunx faf git owner/repo gives an agent instant, scored context on any repo or dependency without a full clone.

Why it's a natural fit

You already reach for bunx to run a tool without installing it. bunx faf gives your AI agents the one thing they usually lack — accurate, versioned, stack-aware project context — in a single command, with zero install, correctly detecting the Bun runtime you actually use. Fast stack, accurate agents.

Links: faf.one · .faf format · faf-cli on npm · IANA application/vnd.faf+yaml