Your AI agent is flying blind. Fix it in one command.

gitcontext

Generate CLAUDE.md / AGENTS.md files for any codebase in seconds. Static analysis detects your stack, architecture, commands, and conventions. --deep mode uses Gemini or Claude to produce a rich context doc from actual source code.

The Problem

AI coding agents (Claude Code, Cursor, Copilot) work better with context files. Writing them manually is tedious and they go stale. Most repos have nothing, so your agent guesses wrong about your build system, test commands, and architecture.

Before: Agent hallucinates your project structure, runs wrong commands, misses conventions.

After: One command generates a context doc. Your agent knows your stack instantly.

$ gitcontext . --output CLAUDE.md

Quick Start

Run on your repo in 10 seconds:

pip install gitcontext
gitcontext . --output CLAUDE.md

That's it. Commit the file and your AI agent instantly knows your stack, commands, and architecture.

Install

pip install gitcontext

For AI-enhanced deep analysis:

pip install 'gitcontext[all]'  # Gemini + Claude support

Usage

# Analyze current directory
gitcontext .

# Analyze any local repo
gitcontext /path/to/repo

# Analyze a GitHub repo directly (shallow clones it)
gitcontext https://github.com/huggingface/lerobot

# Write output to file
gitcontext . --output CLAUDE.md

# AGENTS.md format (slightly different structure)
gitcontext . --format agents

# Deep mode: LLM reads your source code for richer output
export GEMINI_API_KEY=...   # or ANTHROPIC_API_KEY
gitcontext . --deep --output CLAUDE.md

# Choose model for deep analysis
gitcontext . --deep --model gemini-2.5-pro

Example Output (Static)

Running gitcontext . on huggingface/lerobot:

## Project Overview

lerobot — State-of-the-art Machine Learning for Real-World Robotics in Pytorch

## Tech Stack

Python · PyTorch · Hugging Face Transformers · Hugging Face Datasets · Gymnasium · uv

## Development Setup

uv sync --locked

## Key Commands

uv run pytest tests
pre-commit run --all-files

## Architecture

- benchmarks/ — Performance benchmarks
- docker/ — Docker configuration
- docs/ — Documentation
- examples/ — Example scripts and tutorials
- src/ — Source code
- tests/ — Test suite

## Entry Points

- lerobot-calibrate = lerobot.scripts.lerobot_calibrate:main
- lerobot-record = lerobot.scripts.lerobot_record:main
- lerobot-train = lerobot.scripts.lerobot_train:main
- ... and 14 more (see pyproject.toml)

## CI/CD

CI: GitHub Actions (benchmark_tests, docker_publish, documentation, fast_tests, full_tests)

Generated in ~2 seconds. No API key needed.

Deep Mode (--deep)

Deep mode sends your key source files (configs, entry points, core modules) to an LLM and generates a comprehensive context doc with:

• Architecture explanations (the why, not just file names)
• Conventions and patterns (naming, config systems, inheritance)
• Common gotchas and non-obvious behaviors
• Exact setup and contribution workflows

export GEMINI_API_KEY=your-key    # Free tier works fine
gitcontext . --deep --output CLAUDE.md

Auto-detects provider from environment variables. Set GEMINI_API_KEY or GOOGLE_API_KEY for Gemini, ANTHROPIC_API_KEY for Claude.

See examples/output-deep-lerobot.md for sample deep output.

Supported Languages & Frameworks

Category	Detected
Languages	Python, TypeScript, JavaScript, Rust, Go, Java, C/C++, Ruby, Kotlin
Python Frameworks	FastAPI, Flask, Django, PyTorch, TensorFlow, Click, Pydantic
JS/TS Frameworks	React, Next.js, Express, Vite, Svelte
Build Systems	pip, uv, poetry, npm, yarn, pnpm, cargo, go mod, cmake, make
CI/CD	GitHub Actions, GitLab CI, CircleCI, Jenkins
Test Frameworks	pytest, unittest, jest, vitest, cargo test, go test

How It Works

1. Walk the repo file tree (respects .gitignore patterns, skips binaries)
2. Detect languages by extension ratio, frameworks by dependency files, build/test/CI by config files
3. Generate structured markdown from detections

With --deep:

4. Select the ~15-20 most important files (configs, entry points, largest source modules)
5. Send to Gemini/Claude with a prompt tuned for generating contributor-facing docs
6. Output the LLM-generated CLAUDE.md

Total context sent to the LLM is capped at ~80K characters to stay within free tier limits.

Comparison

	gitcontext	gitingest	Manual	Nothing
Auto-detects stack	Yes	No (dumps raw files)	N/A	N/A
Structured output	Yes	No	Yes	No
Works on any repo	Yes	Yes	No (per-repo effort)	N/A
AI-enhanced mode	Yes	No	No	No
Speed	~2s static, ~15s deep	Varies	Hours	0s
Output quality	High (targeted)	Raw dump	Highest	None

gitingest gives you a raw file dump to paste into an LLM. gitcontext gives you a finished, structured context doc ready to commit.

GitHub Action

Keep your context doc up to date automatically. Add this to .github/workflows/update-context.yml:

name: Update CLAUDE.md
on:
  push:
    branches: [main]
    paths:
      - 'src/**'
      - 'pyproject.toml'
      - 'package.json'
      - 'Cargo.toml'

jobs:
  update-context:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: jashshah999/gitcontext@main
        with:
          deep: false
          output: CLAUDE.md
          commit: true

For deep mode, add your API key as a repository secret and set deep: true:

      - uses: jashshah999/gitcontext@main
        with:
          deep: true
          output: CLAUDE.md
          commit: true
        env:
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}

See examples/workflow.yml for the full example.

Contributing

git clone https://github.com/jashshah999/gitcontext.git
cd gitcontext
pip install -e '.[all]'
pytest tests

PRs welcome. Keep it simple — the goal is a tool that works on any repo without configuration.

License

MIT