Projects

Open-source Claude Design alternative for turning prompts into polished prototypes, slide decks, and marketing assets locally. As project lead, I shaped the product direction and led the core agentic design architecture: a design-specific tool loop, starter artifact workflow, syntax and visual verification harness, local snapshot history, and multi-model BYOK support. The project reached 2k stars within 3 days of launch.

A local-first Agent workspace built for real work, with a shared runtime across the desktop app, TUI/CLI, and Headless runner. It combines controlled tool permissions, recoverable event logs, durable tasks, artifacts, and multi-model connections. As a team member, I help explore and build better and more efficient Work Agents, taking responsibility for the development and implementation of automated tasks and Computer Use capabilities.

Open-source Claude Cowork desktop app for Windows & macOS. Contributed to feature development, product iteration, and community collaboration. The project was reposted by 机器之心.

A mobile AI coding app supporting Claude Code, Codex, Gemini, OpenCode, and more. Invited by weishu to join as a project collaborator, contributing to core feature development.

A research-backed tool that diagnoses and optimizes AI Agent SKILL.md files using real session transcripts and static analysis. Outputs prioritized P0/P1/P2 fix reports across 6 scoring dimensions — trigger rate, user reaction, workflow completion, static quality, undertrigger, and token economics. Compatible with Claude Code, Codex, Cursor, OpenCode, Gemini CLI, and any Agent Skills-standard agent.

220+ Chinese PDF lecture notes for AI/LLM courses, covering Stanford CS336, CS224R, CS25, CS231N, Berkeley LLM Agents, and more. Built with a Whisper transcription + LaTeX auto-generation pipeline.