Braxis

Auto-generate AI agent context files. Keep them in sync with your code.

Your AI agents (Claude Code, Cursor, Copilot) read from AGENTS.md to understand your project. When your code changes, that file gets stale. Agents miss patterns, violate conventions, hallucinate.

Braxis solves this: one command generates context files that stay in sync with your codebase—with deep pattern detection, repository gotchas, ownership mapping, and dual-format guidance.

What's New in v2.2: Framework detection with three tiers (fast/balanced/deep), validated complexity formula for onboarding estimation, and security scanning benchmarks showing competitive accuracy (F1: 0.857) against industry tools. All changes backward compatible.

Before Braxis:

Day 1: Agent reads stale AGENTS.md from 2 weeks ago Sees old directory structure Doesn't know about new error handling pattern Makes bad suggestions based on outdated info

After Braxis:

Every push: GitHub Actions runs Braxis Analyzes current codebase Regenerates AGENTS.md, CLAUDE.md, .cursorrules, .agentic-config.json Creates PR with updates Your agents always see current reality

• ✅ One Command - Generate all context files with braxis generate
• ✅ Zero Config - Works out of the box, no setup needed
• ✅ Auto-Score - Measure your project's AI agent readiness (0-100)
• ✅ Score History - Track improvements over time with trends & analytics
• ✅ LLM Recommendations - AI-powered suggestions using Claude API (optional)
• ✅ Multi-Language - Supports Python, JavaScript, TypeScript, Go, Rust, Java, and more
• ✅ CI/CD Ready - GitHub Actions workflow included
• ✅ Pre-commit Hooks - Validate before every commit
• ✅ Safe & Reliable - Input validation, atomic writes, comprehensive error handling
• ✅ Well-Tested - 30+ unit tests with 100% pass rate
• ✅ No Dependencies - Pure Python, zero external packages (LLM features optional)
• ✅ Production-Grade - Used in real projects, actively maintained

• ✅ Three-Tier Detection - Fast (pattern-based), Balanced (AST), and Deep (wrapper analysis) modes
• ✅ Deep Detection by Default - Automatically detects frameworks even when imported indirectly through wrappers
• ✅ Framework Discovery - Automatically identifies project frameworks and dependencies with versions
• ✅ Complexity Scoring - Estimates onboarding effort based on codebase complexity
• ✅ Experience Adjustments - Calibrated for junior, mid-level, and senior developers
• ✅ Production Validated - Tested on 20 diverse projects with 25% average variance

• ✅ Category A: Operations Manual - Automatically extracts AI policy, procedures, and workarounds from CONTRIBUTING.md
• ✅ Category B: Context Guide - Comprehensive architectural context and agent-understanding
• ✅ Smart Extraction - Filters markdown formatting, identifies policy constraints, procedures, and gotchas
• ✅ Graceful Handling - Works with or without CONTRIBUTING.md, never creates empty sections
• ✅ Production Tested - Validated on projects from 4 files to 22,679 files (PyTorch, Keras)
• ✅ Clear Section Separation - 🚨 AI Policy & Operations vs 🏗️ Architecture & Context Guide

• ✅ Repository Gotchas Detection - Automatically identifies common pitfalls (I/O in transactions, missing error handling, etc.)
• ✅ Subsystem Ownership Mapping - Visualizes which subsystem owns what, prevents scope creep and conflicts
• ✅ Cross-Subsystem Contracts - Documents dependencies and data flow between subsystems
• ✅ Common Mistakes Identification - Detects anti-patterns with before/after code examples
• ✅ Enhanced Testing Patterns - Comprehensive analysis of test frameworks, structure, and coverage capabilities
• ✅ 100% Backward Compatible - All v1.1 features preserved; new patterns seamlessly integrate

• ✅ Hierarchical AGENTS.md - Auto-generates scoped guidance files for monorepo subsystems (pnpm, uv, yarn, npm, lerna)
• ✅ MCP Detection - Automatically detects and documents MCP servers (Claude Desktop config, Python entry points)
• ✅ Project Scale Analysis - Categorizes projects by size with growth-phase-appropriate contribution boundaries

# Basic installation (core features)
pip install braxis

# With LLM support (for AI recommendations)
pip install braxis[llm]

For LLM-powered recommendations:

export ANTHROPIC_API_KEY='sk-ant-...'

Get your API key: https://console.anthropic.com

cd /path/to/your/project

braxis generate

That's it. Braxis creates:

your-project/ ├── AGENTS.md (Universal agent instructions) ├── CLAUDE.md (Claude Code optimized) ├── .cursorrules (Cursor IDE rules) ├── .agentic-config.json (Machine-readable metadata) └── (your existing files)

cat AGENTS.md

git add AGENTS.md CLAUDE.md .cursorrules .agentic-config.json
git commit -m "chore: add AI agent context files"
git push

In Claude Code: Automatically reads CLAUDE.md In Cursor: Copy .cursorrules into Cursor Settings → Rules In any agent: Reads AGENTS.md (universal format)

For monorepo projects, Braxis v1.1 automatically generates scoped guidance files:

braxis generate

Output for pnpm/uv/yarn monorepo:

Detected monorepo! Generating hierarchical AGENTS.md...
* AGENTS.md (root - monorepo patterns)
* api/AGENTS.md (backend-specific)
* web/AGENTS.md (frontend-specific)
* cli/AGENTS.md (CLI conventions)
* packages/AGENTS.md (library patterns)

✓ Monorepo detected: PNPM with 4 subsystems

Key Benefits:

• Follows Nearest File Pattern - Developers read the most relevant scoped AGENTS.md
• Scales Infinitely - Works equally well from 10 to 10,000 files
• Maintains Consistency - Root AGENTS.md keeps monorepo-wide patterns
• Allows Customization - Each subsystem can have local rules

Braxis v1.1 automatically detects and documents MCP servers:

braxis inspect

Output for MCP-enabled project:

✓ MCP servers detected: 1 server(s)
  - llama-index-tools-mcp (python_entry_point)

Features:

• Detects from .claude/claude_desktop_config.json
• Detects from pyproject.toml entry points
• Detects custom MCP configs
• Generates MCP usage documentation in AGENTS.md

Braxis automatically analyzes project scale and suggests contribution boundaries:

💡 Project Scale (large): This large project should define clear contribution 
boundaries. Consider LlamaIndex model: 'no new X' policy.

Scale Categories:

• Micro (<50 files): Accept most contributions
• Small (50-200 files): Consider boundaries as you grow
• Medium (200-500 files): Define clear scope
• Large (>500 files): Define "no new X" policies

Braxis v1.4 generates AGENTS.md files that combine two complementary categories:

Category A: 🚨 AI Policy & Operations

• Extracted automatically from your CONTRIBUTING.md
• Contains: AI policies, requirements, procedures, workarounds
• Example: "AI agents must not commit directly to main branch"
• Helps agents follow operational constraints and best practices

Category B: 🏗️ Architecture & Context Guide

• Auto-generated from your codebase
• Contains: Project structure, design principles, testing strategy, scoring
• Example: Directory maps, design patterns, AI readiness dimensions
• Helps agents understand your codebase architecture

# AGENTS.md

**Dual Format**: This file combines Category A (Operations Manual) 
and Category B (Context Guide) for comprehensive agent guidance.

## 🚨 AI Policy & Operations

### AI Policy
- AI agents must not commit changes directly to main branch
- All changes require human review before merging

### Key Requirements
- Python 3.10+ required (from setup.py)
- All tests must pass before submitting PRs

### Development Procedures
- Run `pip install -e .` to set up your environment
- Run `pytest -v` before submitting code
- Follow black/mypy code standards

### Known Workarounds & Caveats
- CI sometimes fails due to network issues - retry if needed
- Docker builds require a VPN in some networks

## 🏗️ Architecture & Context Guide

### Project Structure
- `src/` - Main source code
- `tests/` - Test suite
- ...

### Design Principles
1. Modularity - Clear separation of concerns
2. Testability - Comprehensive test coverage
...

Tested on 4 different project sizes:

Key Benefits:

• ✅ Single comprehensive AGENTS.md (no multiple files to manage)
• ✅ Operational guidance extracted automatically (saves time)
• ✅ Architecture context always generated (fills the gap)
• ✅ Works at any scale (tested from 27 to 22,679 files)

Automatically regenerate context files on every push using GitHub Actions.

Braxis includes a ready-to-use workflow. Copy it to your repo:

mkdir -p .github/workflows
cp /path/to/braxis/.github/workflows/braxis-score.yml .github/workflows/
git add .github/workflows/braxis-score.yml
git commit -m "chore: add braxis auto-update workflow"
git push

Or manually create .github/workflows/braxis-score.yml:

name: Braxis Score Check

on:
  push:
    branches: [ main, develop ]
    paths:
      - '**.py'
      - 'package.json'
      - 'pyproject.toml'
      - 'setup.py'

jobs:
  score:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v4
      - uses: actions/setup-python@v4
        with:
          python-version: '3.11'
      - run: pip install braxis
      - run: braxis score
      - run: braxis generate
      - name: Create Pull Request for updates
        uses: peter-evans/create-pull-request@v5
        with:
          commit-message: 'chore: regenerate braxis context files'
          title: 'chore: update agent context files'
          branch: braxis/auto-update

Result: Every push automatically regenerates context files and creates a PR if needed. ✨

Validate context files before every commit using pre-commit.

pip install pre-commit
pre-commit install

The hooks will run automatically on git commit.

Copy the example config to your project:

cp /path/to/braxis/.pre-commit-config.example.yaml .pre-commit-config.yaml

Then install:

pip install pre-commit
pre-commit install

Now braxis will validate your project before each commit! 🔐

Track your project's AI agent readiness score over time.

# View all historical scores
braxis history

# View scores with trends and direction indicators
braxis history --trends

Example output:

============================================================
Score History for myproject
============================================================

1. 2026-10-01 - 65/100 (AI-Native)
2. 2026-10-05 - 72/100 (AI-Native)
3. 2026-10-10 - 78/100 (AI-Native-Plus)

Trend: 📈 +13 points

============================================================

Features:

• Automatic score persistence on every braxis score run
• Project-specific tracking (stored in ~/.braxis/history/)
• Trend indicators: 📈 (improving) 📉 (declining) ➡️ (stable)
• Configurable history limits
• Timestamps and tier information included

Get intelligent, actionable recommendations from Claude AI.

Setup:

# Install with LLM support
pip install braxis[llm]

# Set your API key
export ANTHROPIC_API_KEY='sk-ant-...'

Usage:

braxis recommendations

Example output:

============================================================
LLM-Powered Recommendations for myproject
============================================================

1. Add Comprehensive Test Suite
   Why it matters: Testing is the foundation of reliable code.
   Current state: Only 7% testing coverage
   How to implement:
   - Start with pytest fixtures for common patterns
   - Aim for 80%+ coverage on core modules
   - Run: pytest --cov to measure progress

2. Implement Input Validation Framework
   Why it matters: Validation prevents bugs and security issues
   Current state: No systematic validation detected
   How to implement:
   - Use Pydantic for request validation
   - Add schema validation to all API endpoints
   - Example: from pydantic import BaseModel

[... more recommendations ...]

Features:

• Uses Claude Opus 5.5 for high-quality analysis
• 5-7 actionable recommendations per run
• Concrete implementation steps for each suggestion
• Focuses on improving Agent Readiness Score
• Gracefully handles missing API keys
• Optional dependency (braxis works without it)

Braxis welcomes contributions! Read CONTRIBUTING.md for:

• Setup instructions
• Development workflow
• Testing guidelines
• Code style guide
• PR process

Quick start:

git clone https://github.com/YOUR_USERNAME/braxis.git
cd braxis
python3 -m venv venv
source venv/bin/activate
pip install -e .
python3 -m unittest test_braxis -v

Create .agentic-config.json:

{
  "name": "MyApp",
  "description": "A production API service",
  "exclude_patterns": [
    "node_modules/**",
    ".venv/**",
    "build/**"
  ],
  "custom_conventions": {
    "error_handling": "Always use try/except and log",
    "async_patterns": "All I/O must be async",
    "validation": "Use Pydantic models for inputs"
  },
  "critical_files": [
    "src/main.py",
    "src/api/routes.py",
    "README.md"
  ]
}

AGENTS.md - Universal format read by any AI agent

CLAUDE.md - Optimized for Claude Code

.cursorrules - Rules for Cursor IDE

.agentic-config.json - Machine-readable metadata

Check how ready your codebase is for AI agents:

braxis score

Example output:

============================================================
Agent Readiness Score: 71/100
============================================================

Breakdown:

 Architecture          10/100 [██░░░░░░░░░░░░░░░░░░]
 Testing                7/100 [█░░░░░░░░░░░░░░░░░░]
 Dependencies          12/100 [██░░░░░░░░░░░░░░░░░]
 Conventions           10/100 [██░░░░░░░░░░░░░░░░░░]
 Entry Points           4/100 [░░░░░░░░░░░░░░░░░░░]
 Security              10/100 [██░░░░░░░░░░░░░░░░░░]
 Build                 10/100 [██░░░░░░░░░░░░░░░░░░]
 Documentation          8/100 [█░░░░░░░░░░░░░░░░░░]

Tier: AI-Native

Detected:
 Languages: python
 Build System: Python (pip/setuptools)
 Test Frameworks: pytest, unittest
 Test Files: 1
 Critical Files: 0

Recommendations:
 * Increase test coverage
 * Add input validation and security checks

• Architecture - Critical files, entry points, project structure
• Testing - Test coverage and test framework detection
• Dependencies - Build system and dependency management
• Conventions - Code patterns, error handling, type hints
• Entry Points - Main functions and executable files
• Security - Input validation, security checks, config management
• Build - Build files and dependency tracking
• Documentation - README and project documentation

python3 braxis.py score --path /path/to/project

# Check version
braxis --version

# Score your project's agent readiness
braxis score
braxis score --path /path/to/project  # Score a specific project

# Generate context files
braxis generate
braxis generate --path /path/to/project

# Inspect project analysis
braxis inspect
braxis inspect --path /path/to/project

# Validate context files exist
braxis validate
braxis validate --path /path/to/project

# View score history
braxis history
braxis history --path /path/to/project
braxis history --trends                 # Show trends with emoji indicators
braxis history --path /path/to/project --trends

# Get LLM-powered recommendations (requires: export ANTHROPIC_API_KEY='sk-ant-...')
braxis recommendations
braxis recommendations --path /path/to/project

• Python 3.8+
• Zero external dependencies
• Works on macOS, Linux, Windows

• Report Issues - GitHub Issues
• Discussions - GitHub Discussions
• Contributing - See CONTRIBUTING.md

# New developer clones repo
$ braxis score
Agent Readiness: 78/100 (AI-Native-Plus)

# They immediately understand the project structure, conventions, and quality baseline
# Opens AGENTS.md in Claude Code - instant context

# Initial state
$ braxis score
Score: 45/100 (Agent-Aware)

# After 2 weeks of improvements
$ braxis history --trends
📈 +28 points

# Team celebrates progress with visual trend data

# Every commit triggers workflow
$ braxis score
$ braxis generate
# PR created with updated context files
# Agents always have latest project info

$ braxis recommendations
# Get AI-powered guidance on what to improve
# Prioritize high-impact changes
# Measure progress with `braxis history`

MIT - Free to use in personal and commercial projects

See LICENSE for details.

See ROADMAP.md for detailed version history and upcoming features.

Braxis — Keep your agents aligned. Keep your code context current.

Continuously analyze. Automatically improve. Always sync. ✨

Made with ❤️ for AI-native development by developers, for developers.