AI Infra Lead @ Yuanfudao (猿辅导) · Beijing
From observability engineer to AI engineer — building infrastructure for the AI-native workplace.
I'm working on the full stack that makes AI-native software development possible.
Platforms
- Rush — AI code generation platform. Users describe what they want, get a full web app with live preview. 2400+ projects created in 4 months. Latest: 0.8.0 native plugin format — one install, works across Claude Code / Cursor / Codex.
- OpenRush — Open-source managed-agents platform for Claude Code, self-hosted on your own infrastructure. Sandboxed execution, Agent / Skill / MCP registry, dual-layer credential Vault, streaming event contract.
- Pilot — Cloud AI Agent platform. AI agents (Claude Code / Codex) run in sandboxed containers, autonomously executing coding tasks end-to-end. SDK (
@kanyun-ai-infra/pilot) and CLI (@kanyun-ai-infra/pilot-cli) shipping with v0.1.0.
Tools
- reskill — Skills package manager for AI agents. Think
npmfor AI skills — declarative config, multi-agent support, version management.npx reskill installworks across Cursor, Claude Code, Codex, Windsurf, and more. - Sparring — Multi-backend peer review plugin for Claude Code. Cursor Agent / Codex / GLM reviewers with primary/fallback and background jobs — better signal than single-model self-review.
- cowork-mdm — Claude Desktop enterprise deployment toolkit. MDM config + plugin marketplace management for Mac/Windows IT admins.
Loop
- TentaClaw — Closing the loop beyond coding: Issue → Code → Deploy → Observe → Feedback. One Repo Agent per repo, handling the full software lifecycle.
Before AI engineering, I spent 4 years building Octopus — Yuanfudao's company-wide observability platform (RUM, Logs, Trace, Alerts, LLM Observability). That experience is why TentaClaw's "Observe" loop isn't theoretical — it comes from real production pain.
I don't just write code — I define products, design architectures, and ship. Recent pace: 4.75 PRs/day on Rush; TentaClaw Phase 1 (11 packages, 300+ tests) built in a single day with 4 parallel AI agents.
I also practice what I preach: Rush, Pilot, and OpenRush are built using AI agents (Claude Code + Cursor) with Sparring peer review as the quality gate.
TypeScript · React · Next.js · Node.js · Hono · PostgreSQL · Redis · Docker · K8s
Claude Agent SDK · AI SDK · Gemini · Codex · MCP · OpenTelemetry
- Email — adwerrd@hotmail.com
- X — @kris_wpf
- 小红书 — @kriiswu






