临书 (tmpbook)

Python developer working at the intersection of AI agents and practical, reliable tooling. I care about code that survives real-world edge cases — race conditions, lazy installs, and the bugs that only show up in production.

What I'm working on

• Contributing to hermes-agent — an open-source personal AI agent that runs the same core across a CLI, a messaging gateway (Telegram, Discord, Slack, and ~20 other platforms), a TUI, and a desktop app.
  • Recent fix: rebind TypeHandler after lazy dependency install — fixed a poisoned module-state bug that silently broke Telegram reconnects for ~50 hours.
• AI agent infrastructure — lazy dependency management, gateway platform adapters, and making agent tooling robust enough to run unattended.

Tech stack

• Languages: Python, TypeScript, Bash
• AI & agent tooling: OpenAI, LangChain, Anthropic
• Tools & platforms: Git, GitHub, Docker, Linux

Quick facts

• I enjoy hunting down the "should never happen" bugs — the ones that only manifest after a restart races with a dependency reinstall.
• I believe good fixes cover the whole bug class, not just the one site a reporter happened to hit.
• Currently exploring: making agent systems more observable and self-healing.

Stats

• 526 stars earned
• 146 forks
• 112 followers
• 53 public repos

Open to interesting conversations about AI agents, Python, and open source.