skill-evolution-in-uselisted
Install: claude install-skill Zane456/skill-evolution-in-use
# skill-evolution-in-use — let a skill learn from being used
Give one chosen skill (the *target*) a memory of how it is really used. Every
conversation that invokes it is copied into the target's own folder, its
changes are stored as numbered versions, and any version can be restored.
Real use shows what test prompts cannot: what the user corrects.
`scripts/evo.py` does all of it (its logic lives in `scripts/evo_core.py`).
Run it instead of redoing its steps by hand: it keeps state, versions and
changelog consistent, and a hand edit under `evolve/` breaks rollback.
```bash
python3 "<this-skill-dir>/scripts/evo.py" <command> [<skill>]
```
`<skill>` is a folder name under the skills root, the skill's `name`, or a path.
Exit 0 = done. Exit 2 = refused or failed, also for bad arguments, with the
reason printed: relay it, don't work around it. Exit 3: see Enabling. If the
skill was never enabled (exit 2 says so), offer `enable`; never enable on your
own initiative.
## What the user wants → command
| User intent | Command |
|---|---|
| turn it on ("给 X 开启进化", "let X evolve in use") | `enable <skill>` |
| how is it going, what was recorded | `status [<skill>]` |
| which versions exist | `versions <skill>` |
| go back to a version ("X 退回 v2") | `rollback <skill> v<N>` |
| undo the last change ("退回上一版") | `rollback <skill> prev` |
| keep a checkpoint now | `snapshot <skill> --note "<why>"` |
| freeze / unfreeze the skill | `lock <skill>` / `unlock <skill>` |
| stop recording | `