I love coding and designing elegant systems.
- 🤖 Exploring: Long-Horizon Harness & Long-term Memory & Agent Infra(Sandbox、Computer Use)
- 🎓 Education: IoT @ HHU | Applying for an MS in CS @ HKU / CUHK / HKUST
- 🧑💻 Internship: prev @GIAN
- 🌐 Links: visit my blog here or reach me on
- YuBlog
- 黑白极简 Astro 开源博客主题
- RepoMirror
- 将 GitHub 仓库或仓库子目录镜像到本地文件夹树的 macOS 桌面应用
- Trajex
- Past Claude Code, Codex, and Pi sessions — queryable by your agent, browsable by you
- XingClaw
- miniopenclaw
- Replica - Memory Layer for AI, built with FastAPI, React, PostgreSQL, and pgvector
- PageIndexRag-Web - A RAG Web application based on PageIndex
- skills - 我平时使用的 Skill 集合,涵盖开发、设计和研究
- understand-project - Agent Skill for Better Understanding Projects
- pr-lifecycle - Treat a PR as a converging contract — SPEC, design, implementation, review, and test ablation
- pi-deep-dive
- In-depth analysis and source-code exploration of pi-coding-agent
- pi-extensions - 我平时使用和推荐的 pi 扩展集合
- pi-repomap - A Tree-sitter-powered codebase structure awareness extension for Pi
- pi-dynamic-workflows - A Claude Code-like Dynamic Workflows extension for Pi
- pi-tool-offloading
- Pi的上下文卸载扩展,将超大工具结果移至会话侧边文件,按需检索
- pi-zen-mode
- pi 的极简模式扩展
pi
· 6 PRs
· Extensible AI coding agent and development toolkit
memU
· 6 PRs
· Memory infrastructure for AI agents
OpenViking
· 9 PRs
· Context database and infrastructure for AI agents
deer-flow
· 12 PRs
· Open-source super agent harness for research and coding
deer-workflow
· 1 PR
· AI-powered workflow automation for software development
| PR | Commit |
|---|---|
| #7 | feat: support Pi Coding Agent 0.84.1 |
maka
· 5 PRs
· Agent that keeps a complete record of everything it did
powercontext
· 8 PRs
· Context for work that humans and agents hand off and continue
letta-code
· 1 PR
· Memory-first coding agent built on Letta
| PR | Commit |
|---|---|
| #4205 | fix(agent): keep memoryScope on provider-not-supported subagent retry |
llm-space
· 1 PR
· Infrastructure and runtime components for LLM applications
| PR | Commit |
|---|---|
| #148 | fix(core): confine LocalFileSystem paths |
open-connector
· 3 PRs
· Open-source connector infrastructure for AI applications
| PR | Commit |
|---|---|
| #435 | fix(oauth): prune expired authorization states |
| #436 | fix(runtime): handle JSON media types case-insensitively |
| #439 | fix(providers): allow private network credential validation |
OpenKB
· 2 PRs
· Open-source knowledge base for AI agents and RAG
| PR | Commit |
|---|---|
| #59 | auto-propagate LLM_API_KEY to provider-specific env vars via config-driven detection |
| #79 | Eliminate duplicate wiki-tool implementations in skill/tools.py |
Unmerged PRs
| Project | PR | Title |
|---|---|---|
| memU | #690 | feat(retrieve): fuse BM25 with cosine via RRF |
| OpenViking | #4658 | feat(session): extract session skills even when no training cases are produced |
| #4660 | feat(bot): deliver send:// generated images on Telegram, Discord, and Slack | |
| #4730 | docs(retrieval): clarify score threshold and hotness semantics | |
| #4736 | feat(bot): process independent sessions concurrently | |
| pi | #8744 | feat(tui): add opt-in overlay selection exclusion |
| maka | #5051 | feat(computer-use): move overlay cursor for set_value and press_key |
| #5052 | feat(computer-use): keep overlay cursor elevated when the aim point is covered | |
| #5050 | feat(computer-use): show snapshot selected text in model observations | |
| #5054 | feat(runtime): add Gemini Google Search grounding | |
| AstrBot | #10039 | feat: support PPTX uploads in knowledge base |





