The body is a dense operational command reference with no concept explanations Claude already knows and no padding; commands, flags, dry-run/apply patterns, and impact notes each earn their tokens.

Every section gives fully executable 'bun run admin -- …' commands with concrete flags, real examples, and confirmation-token flows that are copy-paste ready rather than pseudocode.

Safety Rules sequence the operation (target → reason → confirm → write) with a dry-run-to-apply feedback loop, and a dedicated Verification section provides explicit post-write validation per resource for these destructive/batch operations.

No external references are needed and none are used; the single self-contained command map is clearly sectioned with headers (Safety Rules, Command Map, per-resource subsections, Verification, Impact Notes) for easy navigation, with no nested-reference depth.

The description enumerates eight concrete target objects (skills, users, org publishers, plugin packages, trusted publishers, official publishers, guarded staff email) and names the tool, but the only action verb is the generic 'moderation actions' rather than distinct verbs like ban/delete/transfer.

It explicitly states what the skill does (repo-local ClawHub admin tool acting on enumerated targets) and provides an explicit 'Use for…' trigger clause answering when to use it, so the cap-at-2 rule for missing trigger guidance does not apply.

'Use for ClawHub staff moderation actions' is a natural trigger and terms like skills/users/plugin packages match staff phrasing, but 'trusted publishers', 'official publishers', and 'guarded staff email' are specialized jargon lacking common natural variations.

The product-specific (ClawHub), staff-only, tool-scoped niche with distinct moderation triggers makes it very unlikely to fire for an unrelated skill.