Conciseness

The body is lean and assumes competence — no tutorial padding or explanation of concepts Claude already knows; even the distinction "Research novelty is NOT the same as patent novelty" is domain-specific value. The one notable trim candidate is the ~10-line reviewer-routing blockquote ("resolve invocation → `.aris/reviewer.json` → Claude Code default (Codex MCP) with `reviewer_route.py` through the canonical helper resolver..."), dense orchestration jargon that largely belongs in the referenced reviewer-backends.md. This matches "Efficient; minor instances of over-explanation that could be trimmed" rather than 3, since it is operational instruction rather than explanation of known concepts, and rather than 5 because that block does cost meaningful tokens.

Actionability

Concrete, executable guidance throughout: a claim-element/prior-art matrix template with per-cell evidence, explicit verdict rules ("ANTICIPATED: One reference discloses every element → claim is not novel"), an obviousness combination matrix with enumerated motivation-to-combine tests, a copy-paste-ready MCP examiner prompt with model and config specified ("model: gpt-6-astra, config: {\"model_reasoning_effort\": \"xhigh\"}"), and a complete output file template. This matches "Mostly executable guidance; concrete code or commands with minor gaps" — gaps being the undefined count behind "the top references" and unresolved placeholders like "[USPTO/CNIPA/EPO]". Not 5 because of those minor gaps; not 3 because nothing is pseudocode or abstract.

Workflow Clarity

Six clearly sequenced steps with an explicit conditional gate ("If the invention is novel (passes Step 2), test for obviousness"), per-reference verdicts as checkpoints, cross-model examiner verification as an independent check, and an error-handling path ("If the selected reviewer is unavailable, record `REVIEW_UNAVAILABLE`; continue only as an explicitly unreviewed draft"). This matches "Clear sequence with most checkpoints present; minor validation gaps" — the gap being no guidance for a missing `patent/PRIOR_ART_REPORT.md` input. Not 5 because there is no validate-and-retry loop after the examiner review; not 3 because checkpoints (per-reference verdicts, novelty gate) are explicit, not implicit.

Progressive Disclosure

The body is well-sectioned (Constants, Inputs, Shared References, Workflow, Key Rules) with clearly signaled, one-level-deep references ("Load `../shared-references/patent-writing-principles.md` for novelty/non-obviousness standards"). This matches "Good structure; most content is appropriately placed; references mostly clear; minor organization gaps" — the reviewer-routing blockquote duplicates content that belongs in reviewer-backends.md, and Step 5's inline jurisdiction frameworks (35 USC 102/103, Article 22 CN, Article 54/56 EPC) partially duplicate the loaded patent-format-us.md. Not 5 because of that duplication and because the referenced shared files are not verifiable within this bundle; not 3 because references are clearly signaled and nothing requiring a separate file is fully inlined.

Specificity

"Assess patent novelty and non-obviousness against prior art" names the domain and two concrete actions (novelty assessment, non-obviousness assessment), matching the anchor "Names domain and 1-2 concrete actions, but not comprehensive". It is not 2 because the actions are concrete and domain-specific rather than minimal or generic, and not 4 because it does not list several specific actions (e.g., claim-element drafting, jurisdiction-specific analysis, cross-model examiner verification) that the skill actually performs.

Completeness

It clearly answers 'what' ("Assess patent novelty and non-obviousness against prior art") and 'when' with concrete quoted trigger phrases ("Use when user says '专利查新', 'patent novelty', '可专利性评估', 'patentability check', or wants to evaluate if an invention is patentable"), matching the anchor for explicitly answering both what AND when with concrete triggers. Not 4 because the 'when' clause is already fully explicit and specific, leaving nothing merely implied.

Trigger Term Quality

The trigger set — "专利查新", "patent novelty", "可专利性评估", "patentability check", "wants to evaluate if an invention is patentable" — has good natural keyword coverage with bilingual synonyms, matching "Good keyword coverage; a few natural terms missing". It is not 5 because common variations such as "novelty search", "inventive step check", or "prior art" phrasings are absent.

Distinctiveness Conflict Risk

The description occupies a clear niche (patent novelty/non-obviousness assessment) with distinct, specific triggers unlikely to fire for unrelated skills, matching "Mostly distinct; minor overlap risk with closely related skills". It is not 5 because the 'what' clause mentions "prior art", creating minor overlap risk with an adjacent prior-art-search skill; it is not 3 because the triggers are specific rather than generic.