agent-sync-create-skilllisted

# Create a skill in the library, not in this agent The user's skills live in one canonical library that agent-sync deploys into every agent on every one of their machines. A skill written directly into this agent's own skills directory exists in exactly one place, and agent-sync will later report it as unmanaged drift. **Write it in the library instead — then it is born synced.** ## The flow 1. **Scaffold in the library.** ``` agent-sync new skill <id> --description "<one line>" --json ``` The id is lowercase, digits, hyphens and underscores only. The JSON gives you `path` (the scaffolded `SKILL.md`) and `directory` (where any reference files go). That directory is where you write. 2. **Write the skill there.** Author `SKILL.md` at the path step 1 printed, plus any reference files it needs beside it. Do not create anything in `~/.claude/skills`, `~/.codex/skills`, `~/.cursor/skills`, or a project's agent directories. 3. **Ship it in one transaction.** ``` agent-sync save -m "add skill/<id>" --json ``` `save` takes no artifact argument — it validates the library, deploys everything that is out of date to every routed agent, commits, and pushes, as one transaction. If the push fails because the machine is offline, the local work is still applied and committed — say so and move on; the next sync retries. 4. **Report what happened.** Read `save`'s JSON and tell the user which agents received it, and that their other device