verify-skilllisted

# Verify Skill Skill-authoring guidance changes faster than any skill that copies it. So this skill copies none of it. Every rule applied here is fetched from the upstream docs on each run, and every finding quotes the fetched line it came from. Four axes: - **A. Conformance** to the current spec and authoring guidance. - **B. Currency** of the Claude Code features the skill uses. - **C. Freshness** of the information the skill states. - **D. Upstream drift** of an imported skill against the repository it came from. ## Step 1: Resolve the target | Trigger | Scope | |---|---| | A path to a SKILL.md or to a skill directory | That skill | | A skill or plugin name | Find it, then verify it | | "all skills in \<dir\>" or "all my skills" | Every SKILL.md below that root, one subagent and one report each, dispatched in one message. Claude Code runs up to 20 subagents at once (`CLAUDE_CODE_MAX_CONCURRENT_SUBAGENTS`): with more skills than that, dispatch 20 and start the next one each time one finishes | | Pasted skill content, no file | Verify the text, skip the file-layout checks | | No target given | Ask which skill, with `AskUserQuestion` | For a fan-out, run Step 2's three fetches once yourself into a fresh `mktemp -d` directory. Give each subagent that directory and the URL behind each file in it, and have it run Steps 1 and 3 to 7, fetching into the same directory only the conditional docs its skill needs. Step 8 stays with you. To find a skill by name, Glob these roots