v1.5.0-core: Support for Claude Code, OpenHands и Monorepos

https://preview.redd.it/0u3m2unn892h1.png?width=1200&format=png&auto=webp&s=1e850832bd21ba4f62722ec4a6dc3841738ef955

Release Notes v1.5.0-core

We are excited to introduce v1.5.0-core of the My AI Factory ecosystem! This release focuses on improving the stability of AI agent skill management under our Zero-Dependency philosophy, integrating modern execution environments (surfaces), and fixing critical issues in projecting and cleaning up installed skills.

🚀 Key Features

1. Support for Modern AI Surfaces

Added out-of-the-box integration and automated prompt deployment for cutting-edge coding agent platforms:

• Claude Code (the new powerful CLI developer tool by Anthropic)
• OpenHands (popular open-source software engineering agent platform)

All installed skills are now seamlessly projected and registered directly into these environments.

2. Advanced vendor_as & subpath Catalogs

Introduced support for complex monorepo configurations in catalog.yaml:

• Git subfolder filtering using the subpath directive (e.g., pulling only the necessary skill subfolder from a monorepo).
• Local cache sharing via the vendor_as directive, enabling multiple distinct skills to reuse a single repository clone, drastically saving time and disk footprint.

🛠 Bug Fixes & Stability Improvements

1. Critical Skill Removal Cleanup Fix (cmd_remove)

• Problem: Previously, when deleting a skill installed via subpath (like Anthropic's canvas-design), the installer looked for a directory matching the safe catalog name (anthropics-skills-canvas-design), whereas the actual folder on disk was named canvas-design. This left orphaned plugin files cluttering the coding surfaces.
• Solution: The cmd_remove engine in factory.py has been redesigned to resolve folder structures using Path(subpath).name alongside default slug options.
• Outcome: Safely cleanses all 16 projection files and folders across every supported surface (Cursor, Codex, OpenHands, OpenCode, Claude Code) without affecting other components.

2. Streamlined Catalog (catalog.yaml)

• Completely removed redundant office-related skills (docx, pdf, pptx, xlsx) and the experimental skill-creator generator tool.
• Retained and optimized robust developer integrations: Anthropic MCP Builder (mcp-builder), Canvas UI design guidelines (canvas-design), and the full suite of Vercel best practices.

🔒 Zero-Dependency Commitment

The factory.py core script remains strictly zero-dependency, relying exclusively on Python 3 standard library modules. YAML parsing, environment wiring, path routing, and configuration backups run instantly out-of-the-box in any environment.

🧪 Verification & Testing

• Successfully validated through end-to-end integration smoke tests using tests/smoke_subpath.sh.
• Verified robust projection cleanups via deep uninstallation verification scripts, guaranteeing stable execution.

https://preview.redd.it/hkwah5t6jv0h1.png?width=1674&format=png&auto=webp&s=9ea748891c3d93da705da81228df055cba81bd82

We’ve fundamentally rebuilt the core. factory.py is no longer just an installer - it’s a full-fledged Cognitive Architecture Deployment Engine (think npm or apt, but for AI personas, skills, and workflows).

🔥 What’s new in v1.2.0:
* True Package Manager CLI: Added commands like list, add, remove, init, and doctor. Total control over your AI environment.
* catalog.yaml & Lockfiles: Say goodbye to hardcoded repos. Skills are now managed via a declarative YAML catalog, ensuring 100% reproducible environments through factory-lock.json.
* Smart Recursive Routing: Downloads any external agent repo and intelligently routes agents/, workflows/, and tools/ into a unified .agent/nervous system.
* Managed vs User Space Architecture: We adopted the Unix philosophy (/etc vs /usr/local/etc). The engine natively prioritizes your custom/ agents and automatically backs up files before updates. No more overwritten local tweaks!
* Auto-generated Orchestrators: Instantly builds a pipeline.md workflow, seamlessly linking Analyst, Architect, Developer, and QA roles for autonomous
execution in Gemini CLI or Claude Code.

Build systems, not just code. 
Check it out on GitHub!

Welcome to the fold. I've put together my own build using several repositories. You can try it out, and I'd appreciate any thanks and feedback.