AI Context Generator

Scan any repository and automatically generate AI context files — AGENTS.md, .cursorrules, CLAUDE.md, .windsurfrules, .clinerules, and more. The universal governance layer that makes every AI coding assistant understand your project's architecture, standards, and conventions.

Works with Cursor, Windsurf, Claude Code, GitHub Copilot, Cline, Amazon Q, Continue.dev, Zed AI, Aider, Antigravity, and any LLM-based tool that reads project context files.

---

📚 Documentation

Doc	Description
Getting Started	Install, run, and secure your API key
Configuration Reference	Every .ai_context.toml option
CI/CD Integration	GitHub Actions, pre-commit, Azure DevOps, GitLab CI
Architecture	How the scanner → analyzer → prompt → LLM pipeline works
Contributing	Build, test, lint, type-check and submit PRs

---

The Problem

Every AI coding assistant reads a different context file:

Tool	File it reads
Cursor	.cursorrules
Windsurf (Codeium)	.windsurfrules
Claude Code	CLAUDE.md
GitHub Copilot	.github/copilot-instructions.md
Cline	.clinerules
Amazon Q Developer	.amazonq/rules/project-rules.md
Continue.dev	.continue/rules.md
Zed AI	.rules
Aider	CONVENTIONS.md
Antigravity / AI PR Reviewer	AGENTS.md

Maintaining all of them by hand is redundant, error-prone, and quickly becomes outdated as your project evolves.

---

The Solution

ai-context-generator scans your codebase once, generates a single AGENTS.md as the source of truth, and automatically creates lightweight pointer files for every AI tool your team uses — with zero duplication.

Your Codebase
     │
     ▼
ai-context-generator (scan + LLM analysis)
     │
     ▼
AGENTS.md ─── Single Source of Truth
     │
     ├──► .cursorrules          (Cursor)
     ├──► .windsurfrules        (Windsurf)
     ├──► CLAUDE.md             (Claude Code)
     ├──► .github/copilot-instructions.md  (GitHub Copilot)
     ├──► .clinerules           (Cline)
     ├──► .amazonq/rules/       (Amazon Q)
     ├──► .continue/rules.md    (Continue.dev)
     ├──► .rules                (Zed AI)
     └──► CONVENTIONS.md        (Aider)

Opt-in bridge model: When you have a [output] section in .ai_context.toml, only the bridges you explicitly list are generated — keeping your repo clean. With no config at all, the tool generates all bridges on first run (maximum compatibility). This means your project root stays minimal: if you only use Cursor and Claude, only .cursorrules and CLAUDE.md are created.

Smart updates (cost-saving): The tool stores a lightweight .ai-context.sig signature file in your repo. When the detected project profile is unchanged, the paid LLM call is skipped entirely and the existing AGENTS.md is kept. When the profile does change, the generated content is compared with the current file — if the architecture hasn't changed significantly (less than 10% diff), no files are written, keeping your git history clean. Commit .ai-context.sig alongside AGENTS.md to benefit in CI.

Grounded in your actual repository. The generator does not just send a list of detected frameworks to the LLM. It also sends the real directory layout, entry points, root config files, direct dependencies and the build/test commands it verified in your manifests — and the prompt forbids referencing anything that is not in that evidence. Fewer generic rules, more rules a reviewer can actually check.

Safe against untrusted repositories. An existing AGENTS.md is fed back to the model fenced as untrusted data, with injected fence markers stripped, so a hostile file in a scanned repository cannot hijack the generator.

---

Ecosystem

ai-context-generator  ──generates──►  AGENTS.md  ──consumed by──►  ai-pr-reviewer
                                           │
                       All AI IDEs and assistants read the same source of truth

Used together with ai-pr-reviewer, every Pull Request is reviewed by an AI that already understands your project's architecture, security requirements, and coding standards.

---

Quick Start

⚡ Zero-Install CLI (Recommended for fast local testing)

No installation required using uvx or pipx:

# Run instantly with uvx
uvx ai-context-generator generate --workspace . --api-key $AI_API_KEY

# Or with pipx
pipx run ai-context-generator generate --workspace . --api-key $AI_API_KEY

# Dry-run preview without writing files
uvx ai-context-generator generate --dry-run

📦 Standard Pip Install

pip install ai-context-generator

# Run in your project root
export AI_API_KEY="sk-..."
ai-context-generator generate --workspace .

🔐 Never pass your key on the command line in production. Prefer the AI_API_KEY (or OPENAI_API_KEY) environment variable, or a .env file in your workspace root (loaded automatically, and already in .gitignore). A key passed via --api-key can end up in your shell history.

🪝 Pre-Commit Hook Integration

Add ai-context-generator to your .pre-commit-config.yaml to keep context files updated before every commit:

Note: the hook runs on every commit (always_run) and requires an API key via AI_API_KEY (or OPENAI_API_KEY) in the environment. It is cheap in practice — when the repository profile hasn't changed, the LLM call is skipped automatically.

repos:
  - repo: https://github.com/edsoncarlosdevops/ai-context-generator
    rev: v2.0.0
    hooks:
      - id: ai-context-generator

🤖 GitHub Actions

Add to .github/workflows/ai-context.yml:

name: Generate AI Context
on:
  workflow_dispatch:
  schedule:
    - cron: '0 9 * * 1'  # Every Monday

jobs:
  generate:
    runs-on: ubuntu-latest
    permissions:
      contents: write
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
      - uses: edsoncarlosdevops/ai-context-generator@v1
        with:
          ai_api_key: ${{ secrets.DEEPSEEK_API_KEY }}
          model: deepseek-chat
          create_pr: 'true'

---

🏷️ Add Badge to Your Project README

Show that your project maintains universal AI governance by adding this badge to your README:

[![AI Context: AGENTS.md](https://img.shields.io/badge/AI%20Context-AGENTS.md-6C3FB5?style=flat-square&logo=cpu)](https://github.com/edsoncarlosdevops/ai-context-generator)

---

Configuration

All options live in a .ai_context.toml file at your repository root — the file is entirely optional (sensible defaults are built in). The two most important knobs:

[generator]
model = "deepseek-chat"    # BYOM: any OpenAI-compatible endpoint
language = "english"       # english, portuguese, spanish, french, german
max_lines = 150
timeout = 120.0            # seconds per LLM call
max_retries = 3            # attempts for rate-limit / 5xx / connection errors

[scan]
# Merged on top of the built-in defaults — never replaces them, so you can
# never accidentally start scanning node_modules by adding one entry here.
exclude_dirs = ["my_generated_dir"]
max_file_size_kb = 100
max_files = 100000         # hard cap for huge monorepos
use_gitignore = true       # also skip plain directory names from .gitignore

Sensible directory exclusions ship out of the box (node_modules, venv, target, vendor, .next, dist, build, coverage, .terraform, …), and plain directory names from your .gitignore are skipped too.

Scriptable output. --json prints a machine-readable summary, with or without --dry-run:

ai-context-generator generate --dry-run --json | jq '.domain, .frameworks'

Bridge files stay clean. The tool automatically creates a .gitattributes file marking all generated bridges as linguist-generated, so they collapse in GitHub PR diffs and don't count toward your language statistics.

📖 Full reference — including every bridge flag, CLI override and env var — in docs/configuration-reference.md.

Pipeline Integration

Add path triggers to update context files automatically when your project structure changes. Full GitHub Actions, pre-commit, Azure DevOps and GitLab CI recipes:

📖 docs/ci-cd-integration.md

---

BYOM — Bring Your Own Model

Provider	Model	Base URL
DeepSeek	deepseek-chat	https://api.deepseek.com
OpenAI	gpt-4o, gpt-4-turbo	https://api.openai.com/v1
Anthropic	claude-3-5-sonnet-20241022	https://api.anthropic.com/v1
Ollama (local)	llama3, mistral, codestral	http://localhost:11434/v1
Any OpenAI-compatible	any	custom base_url

# Prefer environment variables — no key on the command line
export AI_API_KEY="$MY_KEY"
ai-context-generator generate --model gpt-4o --workspace .

For Ollama locally, use a dummy key and a custom base_url:

ai-context-generator generate --workspace . --api-key dummy \
  --base-url http://localhost:11434/v1 --model llama3

Model, base URL and language can also come from the environment (AI_CONTEXT_MODEL, AI_CONTEXT_BASE_URL, AI_CONTEXT_LANGUAGE), which is handy in CI. Precedence is CLI flags > environment > .ai_context.toml > defaults.

---

Upgrading to 2.0

The CLI, the .ai_context.toml format and the GitHub Action are unchanged — ai-context-generator generate works exactly as before.

One breaking change affects anyone importing the package in Python: the top-level module was renamed from the generic core to ai_context_generator, so that installing this package no longer squats the core name in your environment.

- from core.scanner import CodebaseScanner
+ from ai_context_generator.scanner import CodebaseScanner

If you invoked the module directly, use python -m ai_context_generator.cli (or just the ai-context-generator console script).

---

License

MIT — edsoncarlosdevops