Conciseness

Every line carries non-obvious project knowledge — the groupAllowFrom mechanics, ignored-flag semantics, auto-registered Matrix accounts, and the no-container/MinIO/Higress exception — with zero padding or explanation of concepts Claude already knows. This matches the 'lean and efficient; every token earns its place' anchor exactly.

Actionability

The Quick Create block is a complete, copy-paste-ready bash invocation with realistic values, and the Operation Reference gives concrete executable commands ('agt get humans', 'agt delete human <name>'). This matches the 'fully executable, copy-paste ready, covers common cases' anchor.

Workflow Clarity

The body delegates the create workflow to the reference and offers no sequence for the destructive operations it surfaces ('agt delete human', level changes that recalculate all groupAllowFrom). Per the rubric, a skill involving destructive operations without validation/verification steps is capped at 3 regardless of simplicity — checkpoints are absent, not merely implicit, so it cannot score 4.

Progressive Disclosure

The body is a clear overview that splits detail correctly: the full workflow lives one level deep in references/create-human.md (verified to exist and to not nest further), and the Operation Reference table maps each admin intent to the right file or command. This matches the 'clear overview with well-signaled one-level-deep references' anchor.

Specificity

Lists four concrete actions ("adding a human user", "changing human permissions", "removing human access", "managing human accounts"), matching the 'several specific actions with minor gaps' anchor. It falls short of 5 because 'managing human accounts' is generic and largely redundant with the other listed actions, and the actions are framed as trigger conditions rather than stated capabilities.

Completeness

The 'when' is explicit and detailed (four concrete trigger phrases), but the description contains no 'what' capability statement — it never says what the skill actually does. This sits between anchor 2 ('only when present') and anchor 3; the rich trigger list weakly implies the what, so 3 rather than 2, but it cannot reach 4 without a stated capability.

Trigger Term Quality

Includes natural admin phrasings users would actually say — "adding a human user", "changing human permissions", "removing human access" — matching the 'good keyword coverage, a few natural terms missing' anchor. Not 5 because common synonyms like 'onboard', 'offboard', 'revoke', or 'delete user' are absent.

Distinctiveness Conflict Risk

The 'human' qualifier consistently applied across all trigger phrases distinguishes it from generic agent or user-management skills, matching 'mostly distinct; minor overlap risk'. Not 5 because phrases like 'changing permissions' or 'managing accounts' could still overlap with a generic user-administration skill.