cyq

I work on AI and agent systems, with a focus on developer tools, computer-use infrastructure, and reliable workflows for coding agents.

I care about small, reviewable changes: clear problem statements, narrow diffs, focused tests, and careful follow-up after maintainer feedback.

Current Focus

• Computer-use agents: desktop automation APIs, accessibility trees, capture, coordinate systems, and multi-OS driver behavior.
• Agent infrastructure: runtime state, tool execution, session records, approval flows, and verification loops.
• Developer tools: command-line workflows, local-first project context, and repeatable handoffs between human operators and coding agents.

Open Source

Recently I have been contributing to agent and developer-infrastructure projects including:

• Cua: computer-use agent infrastructure, desktop drivers, and structured observation APIs.
• CodeWhale: coding-agent runtime, TUI reliability, configuration, and CI/debuggability fixes.
• OpenHands CLI: terminal UI and coding-agent interaction reliability.
• Daytona: workspace and sandbox CLI behavior for agent-friendly development environments.

I try to keep public contributions practical and maintainer-friendly: a small fix, a test that proves the behavior, and a short PR description.

Projects

Conductor

A CLI for coordinating multiple AI coding agents with structured handoff, project records, and review-oriented workflows.

project-scout

A local-first CLI and skill source for prior-art discovery before building, adopting, or reworking tools.

WenYuan

An academic productivity system for humanities research workflows, built around Obsidian and coding-agent assisted knowledge work.

Workspace Agent Harness

A private harness for workspace-level agent tasks, training-ready trajectories, typed tools, sandbox guardrails, replay, and repeatable evaluation workflows.

awesome-agent-rules

Notes and examples for making AI coding agents more predictable through project-local instructions and operating rules.

awesome-context-md

Context templates for helping agents understand a repository before changing it.

Stack

Python, TypeScript, Rust, FastAPI, React, Docker, GitHub Actions, macOS/Linux, LLM APIs, local automation, and agent workflow design.

How I Work

• Read the issue and existing code before proposing a fix.
• Check whether a slice is already occupied before opening a PR.
• Prefer local, focused verification over broad claims.
• Keep learning notes separate from public pull requests.
• Write public comments that are brief, specific, and factual.