Conciseness

The ~190-line body is dense with non-obvious operational knowledge and contains no filler explaining concepts Claude already knows, but several rules are restated across sections: null-dollars-are-not-zero appears in Phase 2 ("Sum dollars only when at least one amount is reported... never $0"), Analysis Rules ("Treat award_amount = null as not reported, never zero"), and Quality Check; RCDC-inspection and SBIR phase-conversion warnings similarly repeat 2-4 times. This matches anchor 3, "mostly efficient but... could be tightened" — the reinforcement is deliberate but the body could shed a quarter of its length by consolidating repeated rules into one place.

Actionability

Guidance is fully concrete and executable for an instruction-only skill: exact endpoint names ("Call OpenNIH_source_status before a broad report"), exact argument behavior ("ic='ALL' returns one combined NIH scope"), exact numeric limits ("require meta.total <= 100050; limit is 50 and offset is capped at 100000"), concrete fallback sequences ("retry Last, First and surname-only"), and precise rendering rules ("render 0.528174 as 52.82%", "render 3.9084 as 3.91%, not 390.84%"). Every instruction names the exact tool, field, threshold, or transformation needed, matching the anchor-5 standard of copy-paste-ready specificity.

Workflow Clarity

Phases 0-3 are clearly sequenced (scope/source state -> resolve -> validate -> synthesize), with an entire validation phase, explicit error-recovery loops ("If a call times out or returns 429/5xx, retry once. If it still fails, report the endpoint as unavailable"), rerun guidance for ambiguous queries, and a literal closing checklist ("## Quality Check: Before returning, verify that...") covering every hazard introduced earlier. This matches anchor 5: clear sequence, explicit validation steps, feedback loops, and a checklist for a complex process.

Progressive Disclosure

The body opens with well-signaled, purpose-stated, one-level-deep references ("Always read references/tool-reference.md before choosing tools... Read references/verified-cases.md when testing... Read references/public-value-cases.md for a... public-facing request"), and all three referenced files exist with no nested second-level references. It falls short of anchor 5 because the body itself is a 190-line rulebook retaining substantial endpoint-specific detail (e.g. the FY1985-1998 historical ranking guard, per-field rendering rules) that overlaps tool-reference.md's scope, and the three scripts/ files (verify_live_cases.py, verify_output_links.py) are never mentioned from SKILL.md, leaving part of the bundle unnavigable.

Specificity

Quotes like "Analyze NIH grant portfolios and funding history using the OpenNIH FY1985-present corpus, then connect grants to investigators, institutions, publications, clinical trials, patents, targets, or drugs" name the domain plus several concrete actions with a rich object list. It stops short of anchor 5 because the verbs themselves are broad ("analyze", "connect") rather than distinct concrete operations like the anchor's "fill forms, merge documents, convert pages", though coverage of downstream outputs is comprehensive.

Completeness

The what is explicit ("Analyze NIH grant portfolios and funding history... then connect grants to...") and the when is explicit with concrete trigger phrases ("Use for NIH grant discovery, topic or Institute/Center trends, PI and institution profiles, ... or public-facing NIH explainers for patients, advocates, ... journalists, trainees, applicants, institutions, entrepreneurs, and taxpayers"). This directly matches the anchor-5 example structure of capabilities followed by an explicit Use-when clause with enumerated triggers; voice is third-person throughout.

Trigger Term Quality

The "Use for..." clause covers many natural phrases users would actually say: "NIH grant discovery", "topic or Institute/Center trends", "PI and institution profiles", "funding growth and concentration", "SBIR/STTR landscapes", "expert discovery", and audience terms like "grant-writing landscape research". A few natural synonyms are missing (e.g. "R01", "grant awards", "federal/biomedical research funding"), which keeps it at anchor 4 rather than 5.

Distinctiveness Conflict Risk

The description is anchored to a clear niche (NIH grant data via the named "OpenNIH FY1985-present corpus" and "ToolUniverse") with distinct trigger vocabulary such as "SBIR/STTR", "activity-code or mechanism analysis", and "Institute/Center trends". It would not plausibly fire for unrelated skills, matching the anchor-5 "clear niche with distinct triggers; minimal conflict risk".