Jaykrishna316 has released Braxis, a Python-based tool designed to eliminate the drift between AI coding agents and actual project codebases. By auto-generating context files like AGENTS.md and CLAUDE.md, Braxis ensures that tools such as Claude Code and Cursor operate on current reality rather than stale documentation. The project addresses a critical pain point in AI-native development: when agents hallucinate or violate conventions because their instruction files haven't been updated in weeks.

Automated Context Generation and Scoring

The core functionality revolves around the braxis generate command, which analyzes the codebase to produce four specific files: AGENTS.md for universal agent instructions, CLAUDE.md optimized for Anthropic's tool, .cursorrules for the IDE, and .agentic-config.json for machine-readable metadata. Beyond generation, Braxis assigns an Agent Readiness Score from 0 to 100, categorized into five tiers ranging from 'Not Ready' to 'Agent-Optimized'. This score is calculated based on architecture, testing coverage, dependencies, conventions, entry points, security, build systems, and documentation, providing a quantitative baseline for how well an AI agent can understand a project.

CI/CD Integration and Historical Tracking

Braxis includes a ready-to-use GitHub Actions workflow that triggers on pushes to main or develop branches, specifically monitoring Python and JavaScript project files. When changes are detected, the workflow runs braxis score and braxis generate, automatically creating a pull request if the context files need updating. Additionally, the tool maintains a local history of scores in ~/.braxis/history/, allowing developers to track trends over time with visual indicators. This historical data helps teams measure the impact of refactoring efforts on their AI agent compatibility, offering a clear metric for 'AI-Native' maturity.

LLM-Powered Recommendations and Zero-Dependency Core

For teams seeking deeper insights, Braxis offers optional LLM-powered recommendations using the Claude API, specifically leveraging Claude Opus 5.5 for high-quality analysis. These suggestions provide actionable steps to improve the readiness score, such as implementing input validation frameworks or increasing test coverage. The core tool remains pure Python with zero external dependencies, ensuring it runs smoothly on macOS, Linux, and Windows without bloating the development environment. With over 30 unit tests and a 100% pass rate, the project emphasizes reliability and atomic file operations to prevent partial writes during generation.

Key Takeaways

  • Braxis auto-generates four context file formats (AGENTS.md, CLAUDE.md, .cursorrules, .agentic-config.json) to keep AI agents in sync with code changes.
  • The tool assigns a 0-100 Agent Readiness Score across eight categories, tracking historical trends via a local history database.
  • Optional integration with Claude Opus 5.5 provides AI-driven recommendations for improving codebase structure and test coverage.
  • CI/CD workflows for GitHub Actions automatically regenerate context files and create PRs on push, ensuring agents always see current code.
  • Core functionality requires zero external dependencies and supports Python, JavaScript, TypeScript, Go, Rust, and Java.

The Bottom Line

Braxis turns AI agent alignment from a manual chore into a measurable engineering discipline. By treating context files as versioned artifacts with a quantified readiness score, it provides the necessary rigor for teams to trust their AI copilots in production environments.