I design developer tools that make AI-assisted engineering repeatable, inspectable, and resumable across Claude Code, Codex, and agentic runtimes. My work focuses on file-backed workflows, fail-closed architecture controls, and evidence a human can actually review.
| Project | What it does |
|---|---|
| repo-harness | Turns coding-agent sessions into a repeatable repo-local workflow with plans, checks, handoffs, and review evidence. |
| Fable-agents | Routes orchestration, execution, deep reasoning, and ship gates across specialized Claude models and Codex. |
| ArchContext | Embeds an architecture control loop into agentic coding runtimes, with local-first and fail-closed boundaries. |
| Agent Auditor | Produces evidence-backed audits for AI agents, Claude Code setups, and Claude Agent SDK applications. |
- Reliable coordination between coding agents and human review
- Local-first architecture enforcement with explicit failure boundaries
- Agent evaluation grounded in repositories, traces, and reproducible checks
Start with repo-harness · Explore all repositories