Conciseness

The body is token-lean with no explanation of concepts Claude already knows, but carries repeated statements: the deterministic-only/'never CLEAN' scope appears in the intro blockquote, Step 1's expected outcome, and Step 2's policy; 'NO REVIEWER KNOBS' is stated twice. Not 5: these redundancies and the over-nested Step 2 policy sentence ('exit 0 ⟺ produced at the current pin ∧ gate exists ∧ paper unchanged since ∧ gate matches the current ledger ∧ decision pass-capable') could be trimmed without loss; not 3: the padding is minor — most content is dense, non-generic, and earns its place.

Actionability

Step 0 is fully copy-paste-ready bash with error handling, and Step 2's gate command is concrete. Not 5: two variables are unresolvable from this bundle — $GATE_HELPER is resolved 'via the canonical helper chain (shared-references/integration-contract.md §2, Policy A)', a file that does not exist in this skill's bundle, and $PAPER_DIR is never derived from $ARGUMENTS; Step 1 delegates to an upstream SKILL.md path rather than giving commands. Not 3: the majority of guidance is genuinely executable, and the gaps are specific and minor rather than pseudocode.

Workflow Clarity

A clearly sequenced Step 0→3 workflow with explicit validation checkpoints: pristine-tree verification ('FATAL: tree not pristine'), the upstream eval gate ('refusing an unvalidated pin'), the out-of-tree marker, FATAL-exit feedback loops on every failure, the `fresh` preflight with defined exit-0 semantics, and tamper protections ('Any ledger mutation deletes the standing gate.json', 'refuses a report older than any paper file'). This matches the anchor-5 pattern of sequence + validation + error-recovery loops, and the destructive git reset/clean steps are validated.

Progressive Disclosure

The bundle has no references/, scripts/, or assets/ directories, yet the body cites 'shared-references/integration-contract.md §2, Policy A' — a dangling, unverifiable reference — and inlines dense policy material (the multi-clause Step 2 gate policy and Step 3 obligations discipline) that belongs in a reference file. Not 4: organization gaps are more than minor because a key navigation target is missing from the bundle entirely; not 2: the body does have clear section structure (Constants, Steps 0–3) and the upstream SKILL.md delegation is one level deep and clearly signaled.

Specificity

The description lists multiple concrete actions — 'Run the ... DETERMINISTIC slice (numeric core + rules-only reporter) against a paper via a SHA-pinned thin launcher', 'convert the verdict into a typed policy gate (BLOCK/WARN/NO_NEW_BLOCKER)', 'an append-only obligations ledger' — with comprehensive coverage of the skill's scope, including the explicit boundary 'it can flag, it can never say CLEAN'. Not 4: coverage is comprehensive across run/gate/ledger phases, not leaving minor gaps.

Completeness

It explicitly answers both: what ('Run ... deterministic slice ... then convert the verdict into a typed policy gate ... and an append-only obligations ledger') and when ('Use when user says "integrity forensics", "forensic audit this paper", "投稿前自查诚信"') with concrete trigger phrases. This matches the anchor-5 example structure exactly; the 'when' clause is explicit, so the ≤3 cap for a missing trigger clause does not apply.

Trigger Term Quality

Explicit triggers are given ('integrity forensics', 'forensic audit this paper', '投稿前自查诚信') — good, natural, and even multilingual — but common synonyms such as 'research integrity check', 'audit my paper', 'statcheck', or file/argument triggers (pdf, arxiv-id) are missing. Not 5: anchor 5 requires comprehensive synonym and file-extension coverage, which this lacks; not 3: the included terms are natural phrases users would actually say, not generic keywords.

Distinctiveness Conflict Risk

A clear niche (paper research-integrity forensics against a named upstream repo, Anti-Autoresearch) with distinct trigger phrases and minimal overlap risk; it even disambiguates itself from the sibling 'mainline' Claude-Code pack by scoping to 'deterministic-only mode'. Not 4: there is no meaningful overlap with other skills — the niche is uniquely identifiable.