agent-teacherlisted

# teacher When someone wants to **understand** a concept, definitions don't stick — running code does. This skill turns "what is X" into a small, scannable lesson: an intuition, a tiny piece of code that demonstrates X, a walkthrough of that code, the trap that makes X necessary, and a few directions to go next. ## When this triggers The user wants to **build a mental model**, not solve a task. Cues: - Direct asks: `解释一下 闭包` / `什么是 CAP` / `教我 monad` / `explain GIL` / `what is a futex` - Confusion cues without "explain": `我不太懂 generator` / `这个 actor model 是啥` / `i never got how async works` - Comparison asks: `process 和 thread 区别` / `mutex vs semaphore` - "How does X work" — the user wants the mechanism, not a fix If the user is asking you to *fix code that uses X*, that's a debugging task, not a teaching task — answer the task directly and skip this skill. If the user has already shown they know X and is asking a follow-up detail (e.g. "in Python, does the GIL also block during I/O?"), answer the detail directly. This skill is for building a fresh mental model, not for incremental questions. ## Output language Match the user's input language. Chinese in → Chinese lesson. English in → English lesson. If the user mixes both, follow the dominant language of the latest message. The code itself stays in its native syntax regardless of explanation language. ## Output medium Default: **in the conversation**, as a structured markdown response. Concept learning is a back-and