Conciseness

The body is lean throughout: a 14-row handoff table instead of prose, three compact code blocks with inline comments (e.g. "# registry key, HF id, or local path"), and terse rule bullets ("Do not add cloud calls to PHI workflows"). It assumes Claude's competence — no concept is explained that Claude already knows — so it sits on the 5-anchor rather than the 4-anchor ("minor instances of over-explanation"), as even the extras paragraph adds new information beyond the pip block.

Actionability

All guidance is copy-paste ready: exact pip extras ("pip install \"openmed[hf]\""), three fully-parameterized Python calls (openmed.analyze_text, openmed.deidentify, openmed.extract_pii with model_name, output_format, method, policy), CLI equivalents ("openmed analyze --text … --format json"), and runtime discovery calls. These cover the common cases; not a 4 because no key execution detail is missing.

Workflow Clarity

The "typical pipeline" section lays out a clear five-stage sequence (ingest → de-identify → extract → ground → assemble FHIR → evaluate) and a validation gate is explicitly present ("De-identification is verified, not assumed. Gate on leakage with openmed.eval, not on F1 alone"), so the destructive-operation cap does not apply. Not a 5: the error-recovery feedback loop (what to do when leakage is detected) is deferred to the companion skill rather than stated inline, matching the 4-anchor ("most checkpoints present; minor validation gaps") rather than the 5-anchor ("feedback loops for error recovery").

Progressive Disclosure

This is a router skill executed exactly as the 5-anchor describes: a concise overview whose bulk is a well-signaled, one-level-deep handoff table ("## When to use this skill" → per-task skill/API pairs) with no content inlined that belongs in a deeper file. No bundle files exist to verify (references/, scripts/, assets/ are absent), but structure alone — table-driven navigation, short themed sections — places it clearly above the 4-anchor ("minor organization gaps").

Specificity

The description lists multiple concrete actions — "named-entity recognition, PHI de-identification, FHIR export, and evaluation" plus "Orient and bootstrap any project" — giving comprehensive coverage of the library's capability surface. Not a 4: there are no minor gaps; each listed capability is a specific, verifiable action rather than a generic verb.

Completeness

Both halves are explicit: the 'what' ("Orient and bootstrap any project that uses OpenMed… for named-entity recognition, PHI de-identification, FHIR export, and evaluation") and a concrete 'when' clause with four specific trigger conditions. Not a 4: the 'when' is not merely present but enumerates concrete trigger phrases, matching the 5-anchor exactly.

Trigger Term Quality

Natural trigger phrases cover the ways a user would actually ask: "mentions OpenMed", "wants to install it", "asks which OpenMed capability or model fits a task", and "build a clinical/medical text pipeline". Synonym coverage is present ("clinical/medical"), and no natural variation relevant to a named library (no file extension applies) is missing.

Distinctiveness Conflict Risk

The description is anchored to a uniquely named library ("OpenMed, the on-device clinical and biomedical NLP library") and positions itself as the entry point ("needs the right entry point"), which distinguishes it from the sibling focused OpenMed skills it routes to. Conflict risk with unrelated skills is minimal.