Full-stack engineer building software end-to-end with AI agents in the loop. Long-form work at C'mon Labs.
I treat AI agents as collaborators in the loop, not just autocomplete.
- Design before code. Every non-trivial feature starts with a brainstorm and a short written spec.
- Parallel agents for disjoint work. I split plans into independent tasks, dispatch them in parallel, then review and stitch the results.
- Custom skills as muscle memory. Repeatable workflows become one-command agent recipes I can re-run.
- Extended tool surface via MCP and n8n. I wire agents to the systems they need — databases, observability, automation — instead of feeding them guesswork.
- thought-shower — Claude
Code plugin: 6-stage pipeline (branch setup → brainstorming → finishing →
Codex review → CodeRabbit review → merge handoff).
- Install:
/plugin marketplace add yuntian3008/yuntian3008then/plugin install thought-shower@yuntian3008
- Install:
- BetterPrompt — production AI product I'm building at C'mon Labs.
- yuntian.dev — personal site and case studies.
- bun-i2c — zero-dependency I2C
wrapper for Bun via
bun:ffi. Linux syscalls, no native addons. - blackat — end-to-end encrypted messaging system.
TypeScript end-to-end. Node.js · Next.js · React Native · PostgreSQL · Drizzle · Redis · BullMQ · Bun · Docker · GitHub Actions · AWS. For AI: any coding agent (Claude Code, Codex, …) — and I write the skills, plugins, and workflows that extend them.