Conciseness

The ~16-line body is lean and efficient with no padding and no explanation of concepts Claude already knows; every line prescribes behavior ('Ground truth lives in truth/, which the verifier never reads', 'Never tune on the set you report'). It matches the 5 anchor.

Actionability

It names concrete metrics ('false-done (target 0), false-reject, missed honesty flag, false honesty flag, errors') and control conditions, but provides no commands or mechanics for executing them: 'Run every scenario', 'Implement, spot-check' and the scoring step all assume an unstated harness, and the scenario file layout is unspecified beyond 'truth/'. This sits at anchor 3 (concrete but incomplete, key execution details missing) rather than 4, which requires mostly executable guidance with only minor gaps.

Workflow Clarity

The four-step round is clearly sequenced with an explicit feedback loop (for each miss: root cause, improvement, target scenario, then implement and retest) plus exit criteria, matching anchor 4. It falls short of 5 because 'spot-check' is undefined and there is no explicit validation step or pass threshold beyond 'target 0' for false-done.

Progressive Disclosure

The skill is under 50 lines, needs no external references, and organizes its content into three well-labeled sections (Benchmark, Round, Exit criteria), meeting the simple-skill exception for a 5. No bundle files exist and the only path mentioned ('truth/') is scenario data, not a skill reference.

Specificity

The description lists multiple concrete actions with comprehensive coverage: 'Measure and improve how often the done-gate is fooled', 'seeded false-done benchmark with hidden ground truth, genuine-fix controls, honest-partial controls, and fresh blind report sets', and 'rounds of test, improvement plan, implement, retest'. It matches the 5 anchor; the 4 anchor would require minor coverage gaps, and none are present.

Completeness

It explicitly answers both what (measure and improve via a seeded benchmark with controls, run in rounds) and when ('Use after changing any agent, prompt, test, or verifier, and on a weekly schedule'). The 'when' is fully explicit with concrete triggers, matching the 5 anchor rather than the 4 anchor's weaker 'when'.

Trigger Term Quality

'Use after changing any agent, prompt, test, or verifier' plus 'done-gate', 'benchmark', and 'verifier' give good keyword coverage a user of this skill would naturally say, but common phrasings like 'claims done', 'false completion', or 'verify the agent finished' are missing. It sits above anchor 3 (relevant keywords present) but short of anchor 5's synonym-level coverage.

Distinctiveness Conflict Risk

Done-gate calibration with hidden ground truth, blind report sets, and honesty-flag metrics is a clear niche with distinct triggers and minimal conflict risk with other skills. The description matches the 5 anchor.