SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI codingfaf-cli

faf-cli

faf-cli is a context tool for AI coding agents, based on the IANA-registered .faf format. It detects from a code repository's real tech stack and generates the context files read by agents, such as AGENTS.md, CLAUDE.md, GEMINI.md, and .cursorrules, ensuring that content is versioned with the code, does not drift, and requires no repeated explanation. Through a single project.faf file, it provides a consistent foundation of project understanding for multiple AI coding tools. It supports installation via npm, PyPI, and crates.io, and offers scoring, export, Git-native version management, and the portable .fafm agent memory format.

Open website

About this tool

faf-cli is a command-line tool built around the .faf (Foundational AI Format) context format, with the goal of letting AI coding agents obtain accurate, versioned project context from a repository's real tech stack. The official description emphasizes: the context read by every AI coding agent is generated from your repository, not guessed.

Purpose

  • Provide project-level context for AI coding agents (such as Claude Code, Grok, Cursor, GitHub Copilot, etc.).
  • Use a single project.faf file to automatically generate agent instruction files such as AGENTS.md, CLAUDE.md, GEMINI.md, and .cursorrules, avoiding content drift and repeated explanations.
  • Let context follow Git version management, so that every clone, fork, or checkout obtains complete AI context.

Main capabilities

  • Automatic detection: running faf auto detects and populates project information; running faf without arguments displays the project score.
  • Multi-format generation: faf export can generate AGENTS.md (--agents), Copilot-readable files (--copilot), and more.
  • Scoring mechanism: since v8.0 it uses the always-33 scoring kernel (Rust WASM), scoring across all 33 Mk4 slots, with the application type determining which slots are active; the remaining slots are marked slotignored but remain visible. The recommended target is Trophy 100%.
  • Custom rules: custom instructions can be written in ai_instructions.warnings in project.faf, and will appear verbatim at the top of every AGENTS.md written by faf.
  • Git-native: since v7.0 it supports faf diff, faf log, and faf hooks, with context versioned alongside the code.
  • Cards and identity: since v8.1, faf card init can generate agent.fafa, an AI Catalog entry, and an ARD entry from seven questions; .fafa is the portable identity format for the Agent Layer.
  • Memory format: .fafm is the portable agent memory format introduced in v7.2.0 and is interoperable with claude-fafm-sdk 1.0.
  • Rendering and display: faf show can render .faf as a browsable page; faf loop can drive a repository to 100% or until it hits an artificial bottleneck.
  • Compiled artifacts: since v6.0, Bun single-file compilation is used, producing standalone binaries suitable for CI/CD, Docker, or isolated environments.

Usage (based on official text)

  • After installation, run faf with no arguments to view the project score; faf auto automatically detects and fills it in.
  • Run faf --help to view the complete command set and options.
  • Distributed via npm, PyPI, and crates.io.
  • Custom rules are written into ai_instructions.warnings in project.faf.
  • Generate AGENTS.md with faf export --agents; generate Copilot files with faf export --copilot.

Applicable scenarios

  • Projects where multiple AI programming tools coexist and need a unified source of context.
  • Teams that want context versioned alongside code and to avoid repeatedly explaining project background to AI.
  • Scenarios that need a single-file binary for CI/CD, Docker, or isolated environments.
  • Scenarios that provide portable identity (.fafa) and memory (.fafm) for agents.

Limitations and notes

  • The official text does not provide specific pricing, fees, or commercial licensing information; the project uses the MIT license and is free and open source.
  • The official text does not describe supported operating system details, only mentioning that compiled artifacts cover four platforms and that tests run on Windows, Ubuntu, and macOS.
  • The scoring target is 100%; tiers below 100% still exist in the scoring ladder, but the official target is no longer 85%.
  • The official text mentions that when upgrading from 7.x to 8.x, the .faf score missing 12 enterprise markers may drop (21 filled = 64%), and running faf auto to write the markers can restore the score.
  • The official text does not provide specific installation commands, only stating distribution via npm, PyPI, and crates.io; if installation commands are needed, refer to each platform's package manager.

Related tools

Browse category