Agent systems, developer tools, and open-source infrastructure
I build agent runtimes and tools for developers. Most of my work is about making long-running agents easier to inspect, test, and recover when something goes wrong.
Right now I am working on CodeRook, a local-first coding agent with durable sessions, explicit permissions, reviewable changes, and evidence-backed run results.
| Project | What it does |
|---|---|
| CodeRook | Local-first coding agent with TUI and web clients backed by one persistent daemon |
| CoursePilot | Course learning agent that grounds Q&A and study paths in teacher-reviewed material and knowledge graphs |
| work-hunter | Local job-search assistant that ranks roles and keeps a human approval step before outreach |
| Plinth | Local runtime for inspectable agent workflows, durable run evidence, and explicit effect authorization |
- Apache Maka #3644: fixed Windows cleanup failures in non-blocking shell runtime tests. Merged.
- OpenHands software-agent-sdk #4717: added structured launch-time skill overlays for profile-based ACP agents.
- Model Context Protocol TypeScript SDK #2699: added bounded validator caching for schemas without usable IDs.
- Microsoft Agent Framework #7839: fixed message input for declarative workflows in DevUI.
- Strands Agent SOP #78: added deterministic recursive discovery for SOPs in nested directories.
I work mainly in Python and TypeScript, with Rust, Go, and C++ where they fit the problem.