Conciseness

The body is dense and assumes competence — no basic concepts are explained, and every step carries operational detail. It is not a 5 because the opening rationale paragraph and a few aphoristic asides ('a wrong invariant is cheaper to correct now than after the hunt') could be trimmed without losing guidance.

Actionability

Concrete commands appear throughout ('git rev-parse', 'git log --oneline <fixed-point>..HEAD', 'git status --short', 'git diff plus new files via git add -N') and the Brief is a complete, copy-paste-ready subagent dispatch template whose angle-bracket placeholders are explicitly justified ('Replace the angle-bracket fields. Keep the rest.'). It fully matches the 5 anchor.

Workflow Clarity

Six clearly sequenced steps with an explicit evidence gate (step 4's five-part finding requirement), a feedback loop ('Send anything less back with "reproduce or drop"'), verification requirements ('confirm the focused suites pass with them and that exactly your new tests fail without them'), and a done-when checklist. This matches the 5 anchor's validation steps, feedback loops, and checklist.

Progressive Disclosure

Sections are well organized (Steps, Brief) with clearly signaled one-level cross-skill references (../code-review/SKILL.md, ../code-review/rules.md, ../pr/SKILL.md). It is not a 5 because the ~40-line brief template inlined in a >50-line skill could plausibly live in a bundle reference file, leaving SKILL.md a leaner overview.

Specificity

The description lists multiple concrete actions — 'writes inputs that try to break each property the design promises', 'a finding counts only with a test that fails on this head', 'changes since a fixed point' — comprehensively covering the skill's capability in third person. It exceeds the 4 anchor because there are no notable gaps in coverage of what the skill does.

Completeness

Both 'what' (adversarial invariant falsification with test-gated findings) and 'when' are explicitly answered with concrete trigger phrases ('Use when the user asks for an invariants hunt, to break or stress a change...'). This matches the 5 anchor exactly.

Trigger Term Quality

Natural trigger phrases are present — 'invariants hunt', 'break or stress a change', 'prove that a design holds' — but a few natural synonyms users might say (e.g. 'verify', 'holds up', 'falsify', 'property-based') are missing. Good coverage rather than comprehensive, fitting the 4 anchor.

Distinctiveness Conflict Risk

The skill occupies a clear niche — invariant falsification via adversarial inputs — with triggers distinct from ordinary code review or testing skills. It is not the 4 anchor since overlap risk with closely related skills is minimal given the explicit 'Do NOT review the diff' framing carried by the trigger set.