Building practical local-first tools, AI-agent workflows, and small products that can be tested in the real world.

Projects · Stars · AiTeam · MirrorLife · Nestful

I try to build in a way that is honest about the current state of the work: useful first, polished later, with clear boundaries around what is proven and what is still an experiment.

My GitHub bio says it simply: be real, be brave, be positive.

• Reliable local systems: tools that can run on my machine, keep data ownership visible, and avoid unnecessary cloud coupling.
• AI agents with accountability: AI can help with execution, but the product judgment, scope control, and final responsibility should stay explicit.
• Small product loops: I like MVPs with a real feedback path instead of broad architecture that has not met users yet.
• Transparent learning: I am comfortable showing unfinished work when the direction, limits, and next questions are clear.
• Human-scale software: family tools, personal data tools, narrative games, and workflows that serve ordinary life instead of only demos.

A local dashboard and operating loop for independent developers working with AI coding agents. It collects local Codex / Claude-style sessions, groups them by project, and turns execution traces into project health, role quality, audit findings, and rule feedback.

Why it matters to me: AI coding is powerful, but without a feedback system it is easy to confuse activity with progress.

A personal narrative social-simulation game. The core loop is: real-life fragment -> avatar action -> social response -> city echo -> archive. It is intentionally framed as an experimental game, not as therapy or a psychological promise.

Why it matters to me: I am interested in whether simulation and narrative can make reflection feel alive without pretending to solve someone's life.

A WeChat ecosystem MVP for family collaboration: reminders, activities, memories, health routines, files, and household coordination in one private family space.

Why it matters to me: family software should feel calm, useful, and trustworthy. The hard part is not feature count; it is making shared routines easier without becoming noisy.

Over the last year, my stars have clustered around:

• AI agents, coding agents, multi-agent systems, agent memory, and agent evaluation;
• Codex / Claude Code / Cursor-style workflows, MCP, code intelligence, and knowledge graphs;
• local-first and privacy-aware tools for chat history, WeChat data, personal archives, and automation;
• generative agents, social simulation, game agents, and narrative systems;
• builder infrastructure such as self-hosted backends, API gateways, newsletters, social publishing, and automation tools.

The strongest language signal from those repos is Python and TypeScript, with regular interest in Go, Rust, C/C++, and local desktop or CLI tooling.

• I prefer a concrete first loop over a complete imagined platform.
• I write down scope and non-goals because trust usually comes from clear limits.
• I like local-first defaults for private or personal data.
• I use AI agents heavily, but I review their output as an operator, not a spectator.
• I try to turn each project into a system that can learn from its own usage.

This profile is generated from public GitHub evidence: my public repositories, recent public activity, and recent starred repositories. It is not meant to overstate contribution level or ownership. When I mention a fork or an experiment, I label it that way.

For a short evidence log behind this page, see PROFILE_SOURCES.md.

Last refreshed from public GitHub data: 2026-06-29.

_{Real work, clear limits, steady improvement.}