Staff Software Engineer in Austin, TX. 18 years in software. I build AI developer tooling, including coding-agent workflows, automated code review, and agent tools that work in more than one harness. I also own DevSecOps and compliance for a network-security product.
- How my coding-agent kit fits together: the devkit, the shared knowledge base, and the agent tools, with diagrams.
- Keep Going: an anti-stall design for coding agents: harness instructions that keep coding agents working between steps instead of waiting for confirmation.
- The devkit's ticket-to-PR agent pipeline: ticket breakdown, Temporal workflows for developer, tester, and PR-creator agents, and prompts built from the knowledge base.
- AI pull-request reviewer: separate review passes, review history stored in PR comments, and blocking on high and critical findings only.
- PR comment responder: a verdict for every review thread, corrected inline suggestions, and a summary generated from those verdicts.
- Agent tool manifests and the browser handoff: a tool manifest catalog, and a containerized browser that hands login pages to me over noVNC.
- Tracking parallel coding-agent sessions: per-agent status files, a census, and rules for resuming and splitting work.
- agent-tools: a typed tool-manifest catalog and a containerized browser runtime with a sign-in handoff.
Contact: sendtobenny@icloud.com · Docs index: bennytaipei.github.io