skill-auditlisted

# Skill audit This is `/tool-audit` with the scope narrowed to skills, and with one thing it does not do: **it runs the picks** instead of only recommending them. **Read `/tool-audit` first** (`~/.claude/skills/tool-audit/SKILL.md`) and follow it, with the changes below. Everything there still binds: the three rules, the one ledger, the caps, Proven first, memory notes, the report, the debrief. Nothing is duplicated here, because two copies of a rule is one copy that goes stale. **It does not run on its own.** Only when the user asks. There is no session-start hook and no first-prompt trigger; both were removed on 2026-09-18 because the per-session recommendation was not worth the delay on every session. Do not reinstate either. ## What changes **Scope.** Skills only, live and parked. Cap of 5 stands. Ignore the agent, MCP and plugin caps: if the answer to the goal is an MCP server or a plugin, say so in one line and hand over to `/tool-audit`, which can write a launcher. This skill never writes one. **Candidates come from a whole-library sweep**, the same one `/tool-audit` uses, filtered to skills (live, parked, plugin and Open Design) plus memory-note tools that behave like skills: ```bash python3 ~/toolbox/bin/library.py index # if anything was added since the last run python3 ~/toolbox/bin/library.py find "<goal + synonyms>" -n 60 --named "<skills the user named>" ``` Read its `COVERAGE searched N of N` line first. For anything beyond a quick looku