I build local-first tools for AI coding agents: reusable skills, durable context systems, and evidence-backed engineering workflows. My work favors explicit boundaries, deterministic checks, and artifacts that another developer or agent can audit.
| Project | What it demonstrates |
|---|---|
| Codex Custom Skills | 14 reusable agent skills with trigger contracts, evals, validators, and portable workflows. |
| GitHub Pattern Knowledge | A TypeScript pipeline that turns commit-pinned repository evidence into reusable engineering patterns. |
| Codex Memory System | A sanitized, executable blueprint for layered agent memory, retrieval, write gates, and governance. |
- Treat tests, schemas, and source traceability as product features.
- Separate canonical source, generated projections, runtime evidence, and private data.
- Keep agent workflows explicit about triggers, stopping conditions, and verification.
- Prefer local-first systems that remain inspectable and portable.
TypeScript, Python, React, Next.js, FastAPI, GitHub Actions, Markdown knowledge systems, and AI agent tooling.