awb-using-skillslisted

# Using skills — the routing protocol > **Announce on activation:** "Using awb-using-skills to route this to the right skill." This is the one always-on skill. It does not implement anything; it decides **which** skill should run. The actual trigger → skill map is **not** here — it lives in [`skill-registry.md`](../skill-registry.md), the single source of truth, and the SessionStart hook injects a compact copy at the top of every session. This file is the *method* for reading it. ## The mandate **If there is even a ~1% chance a skill applies, invoke it.** Routing is cheap; a missed guard is not. A one-second skill invocation is more reliable than your memory of what the skill says. After invoking: announce `"Using <skill> to <purpose>."` If the skill is a checklist, track each item (e.g. a TODO per step) and finish them in order. ## Priority — who wins when guidance conflicts ``` 1. The user's direct instruction (this chat, CLAUDE.md / AGENTS.md) ← highest 2. A matching skill (the registry + this protocol) 3. Claude Code default behavior ← lowest ``` ## When several skills match — resolve in this order 1. **Tier precedence: Workflow > Guard > Feature > Audit.** A workflow skill orchestrates guards *inside* it, so if the user wants the full process, the workflow wins — and it may run a guard within itself (both active is normal, not a conflict). 2. **Match on the OBJECT, not the verb.** Verbs ("fix", "optimize", "review",