I build tools for everyday work with AI coding agents.

I'm a software engineer based in China. My projects help you choose skills for a project, see what an agent is doing, and take its results with you. My background is in quantum measurement and control, from chip topology to calibration workflows.

Otterpack brings together the tools, themes and workflows I use day to day. It includes my own tools, community adaptations and third-party projects, with their sources and licenses documented.

Find a starting point · Browse the tools · See the workflows

• dynamic-skills — Keep a shared skill library; choose and pin the skills each project needs. Built for Codex, Claude Code, Kimi Code and Pi. Install · PyPI, Alpha.
• conv-docs — Read selected workspace documents on your phone, with token authentication, temporary previews and exclusion rules. Deployment guide.
• pi-hud — See your model, context snapshots, tool activity and usage in Pi's terminal. Full-session accounting and provider quota queries are optional. Local install.
• cd-connect — Hand a scoped task to DeepSeek Harness Desktop, then inspect its execution and results before accepting the work. Quick start.
• QArray — Represent quantum-chip qubits and couplers as arrays; work with rows, columns and regions of a chip. Examples.

Also building Qingniao, a local model gateway for choosing routes independently across Claude Code instances. The core is runnable; there is no public registry release yet.

_{Pi 1.3.0 with Otterpack's theme and plugins, and conv-docs on mobile. English interfaces with demo content.}

Selected merged fixes:

• cc-connect — made skill discovery follow the selected workspace, use Codex's enabled catalog, and find Pi's project skills.
• herdr-remote — added Codex conversation-history reading so sessions can be reviewed from the remote client.
• NextAI Translator — restored the missing Quit action in the Linux tray menu.

• OpenAI Agents SDK Learning Lab — a hands-on path from a first agent to a bounded, observable terminal application.
• Learn QEC — learning quantum error correction, surface codes and lattice surgery in public.

I start with reproducible problems, keep changes reviewable, and check behavior at real agent-host and provider boundaries. Tests, recovery behavior and documentation are part of the work.

Python · Rust · TypeScript

For questions or feedback, open an issue in the relevant repository. I'm happy to talk about agent tooling, quantum chip topology and QEC learning.