Conciseness

The body is maximally lean: every line carries operational content (exit codes, severity names, gate conditions, hard rules) and nothing explains concepts Claude already knows. It matches the "every token earns its place" anchor; the closing provenance paragraph is licensing/design justification rather than padding, so it is not a 4.

Actionability

The sh block gives copy-paste-ready commands with flags and exit semantics ("python3 $S/human_gate.py open plan.md --launch", "close plan.md # exit 2 = NOT done"), plus a concrete sidecar worked example and "human_gate.py --sample" to exercise the whole loop. It is not a 4 because the common cases (open/status/collect/close/waive) are all covered executably.

Workflow Clarity

The loop is clearly sequenced (open → status → collect → close) with explicit validation checkpoints: exit-2 refusal, G1–G7 gate rules, round-cap escalation, and the G7 integrity check for mistyped severities. Feedback loops are explicit (collect applies every item; close names what is open), matching the score-5 anchor rather than 4.

Progressive Disclosure

The body is a well-sectioned overview with one-level-deep references, each signaled with its purpose: "Reasoning lives in references/ — human-in-the-loop canon, feedback batching, loop discipline", "Worked example and JSON contract are in assets/", and conditional "Read human-gate-context.md first if it exists". Note: no bundle files are present in this evaluation copy, so this is scored on the body's structure and signaling, which match the score-5 anchor; it is not a 4 because every reference is clearly named and motivated.

Specificity

The description lists multiple concrete actions — "Builds a single-file HTML review page", "collects batched feedback as a structured artifact", "runs a gate that refuses to close while a BLOCKER is open, the reviewer is unnamed, or nobody has reviewed at all" — with named conditions, matching the comprehensive-coverage anchor. It stays in third person ("Runs", "Builds") with no voice penalty, and is not a 4 because there are no coverage gaps in the action list.

Completeness

Both what and when are explicit and concrete: the what is the three named actions, and the when is "Use when a plan, spec, RFC, report, landing page, migration, or any irreversible action needs human sign-off before shipping" with quoted trigger phrases. This matches the score-5 anchor exactly; it is not a 4 because the when-clause needs no added specificity.

Trigger Term Quality

It quotes natural user phrasings verbatim — "'get sign-off', 'have someone check this", "'hold until reviewed', 'needs approval first'" — alongside "needs human sign-off before shipping", covering synonyms of the core trigger. It is not a 4 because no common natural variant is missing.

Distinctiveness Conflict Risk

It carves a clear niche (human sign-off lane vs machine verification) and explicitly disambiguates adjacent skills: "NOT for making AI text sound human (use content-humanizer or behuman). NOT for reviewing code diffs (use md-review or code-reviewer)". Minimal conflict risk; not a 4 because the negative triggers remove even the minor overlap.