shipped project
Agent Skill Registry
Agent Skill Registry is a production-grade environment for managing the lifecycle of agentic capabilities. It abstracts complex LLM integrations into portable SKILL.md packages that can be built, governed, and shared. The platform features an embedded 'Librarian' AI archivist that intelligently routes natural language queries, cites internal skill logic, and stages safe command executions directly into a native terminal console. With multi-provider config routing (OpenAI, Gemini, Local LLMs), strict T1-T4 trust-tier governance gates, and an ultra-premium dark-mode UI, it represents a complete paradigm shift in how AI tools are managed.
Problem
What it solves.
Currently, AI capabilities are tightly coupled within monolithic agent codebases, making them nearly impossible to version, audit, or safely share across engineering teams. Agent Skill Registry solves this by isolating capabilities into portable, markdown-based skill packages. Combined with a secure execution console and strict trust-tier governance, it provides a unified source of truth where teams can build, govern, and interact with agent tools safely.
What I built
The specific parts I implemented.
What I owned
My responsibility surface.
Product direction
Full-stack architecture
AI workflow (Librarian)
UI/UX (omo.dev style)
Architecture
How data moves through the system.
Data model
What entities the system tracks.
- id - name - description - trust_tier - body_path
- id - provider - model - routing_role
Verification
How I tested or evaluated it.
- ✓Run local stack: npm run dev:backend + npm run dev.
- ✓Verify Librarian provides accurate citations sourced from local SKILL.md files.
- ✓Test staging a ```console block from the Librarian panel directly into the console input.
Next improvements
Known gaps & next moves.
- →Persist Librarian chat history server-side.
- →Add more SKILL.md bodies for registry skills.
- →Expose 'teach Librarian' flow from the Builder.