Operational capabilities for high-leverage AI coding agents.
A modular, .md-first library of 23 foundational skills designed to make AI agents reliable, observable, deterministic, and self-improving.
Born from real-world friction in a complex production project, this collection prioritizes safety gates, state visibility, impasse detection, adversarial review, and structured learning from failure — before layering on higher-order capabilities.
It provides capabilities, not full autonomy, policy enforcement, or agent orchestration.
In the rush to build powerful agents, most skill collections jump straight to flashy features. This library does the opposite: it enforces a proper foundation first — exactly as rigorous process evaluation (Lean, Six Sigma, fault-tree analysis) demands in traditional engineering.
Every skill lives in its own self-contained folder with a clear SKILL.md (natural-language instructions and triggers) plus supporting scripts. This design makes the library transparent, auditable, and easy to extend or port to other agent frameworks via the MCP connector.
- Create or navigate to the skills directory:
mkdir -p ~/.gemini/antigravity/skills cd ~/.gemini/antigravity/skills
Skill: SKILL-016: MCP Connector
server.py - Exposes this entire library as an MCP Server.