-
skill-vault: One MCP Endpoint for Every Skill Your Agent Will Ever Need
Every skill in an agent's context costs ~50 tokens of pure metadata. At 10,000 skills that blows past most context windows. I built a self-hostable semantic skill registry over a single MCP endpoint — search, retrieve, and verify on demand.
-
mcp-apps-render — Your MCP Apps UI, Right in the Terminal
A CLI that renders MCP Apps interactive UI payloads (dashboards, forms, visualizations) as terminal-safe ASCII — proving you don't need an iframe, a browser, or a GUI to actually see what your MCP tools return.
-
doc-inject-guard: Detecting Prompt Injection in Documents Before They Reach Your AI
The Context Collapse AI worm proved malicious instructions in Word docs can alter Copilot's financial data. I built a CLI that scans DOCX, PDF, MD, and HTML for injection payloads before they hit your agent pipeline.
-
agent-knowledge-graph: Persistent Graph Memory for AI Agents
I built a local-first memory system that stores agent sessions in Neo4j, extracts structured knowledge via LLM, and enables semantic recall — all running on your own machine with zero API costs for embeddings.
-
eu-act-check — EU AI Act Compliance Scanner, With 3 Days Until Article 50 Takes Effect
A CLI that scans files for C2PA provenance manifests, EXIF AI-generation metadata, and synthetic text markers — and tells you exactly what's missing before the August 2 deadline.