Content
78%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured skill body: lean overview, correct use of a single one-level reference file, a clear sequenced ETL workflow with explicit validation, and strong edge-case coverage (unmapped concepts, domain routing, PHI handling). The main gaps are a not-quite-executable quick-start snippet and the absence of a complete example OMOP row or load code.
Suggestions
Make the Quick start snippet executable: define or comment out the undefined fact_start/fact_end variables (e.g. resolve them from result["entities"][0]["start"]) so the example runs as written.
Add one complete example OMOP row (e.g. a full condition_occurrence INSERT or dict) built from the Quick start fact, so the concrete end-to-end output shape is shown, not just described.
Close the workflow's feedback loop: specify what to do when Achilles/DataQualityDashboard checks fail (which failures block analytics vs. can be logged), mirroring the 'needs review' staging already defined for undated facts.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — the Quick start, field table, and gotchas are dense and non-padded — but the intro explains background ("The OMOP Common Data Model (CDM) is the OHDSI standard for observational health data") and the no-license-vocabulary rule is stated twice (Quick start paragraph and Edge cases). Matches the 4 anchor (minor trimming possible), not the 3 anchor since padding is incidental, not pervasive. | 4 / 5 |
Actionability | Concrete guidance dominates: a realistic fact dict with real concept ids (45533010, 201826), a domain-to-table field matrix, and a per-field reference file. However the Quick start snippet is not fully executable ("char_span": (fact_start, fact_end) references undefined variables) and no complete example OMOP row or load statement is shown. Sits between the 3 anchor (pseudocode, missing key details) and 5 (copy-paste ready); the concrete ids and full field reference keep it at 4. | 4 / 5 |
Workflow Clarity | The 6-step workflow is clearly sequenced and includes an explicit validation step ("Run OHDSI Achilles/DataQualityDashboard on the loaded CDM before analytics") plus a staging route for undated facts, so the batch-operation cap at 3 does not apply. Missing the explicit validate-fix-retry feedback loop of the 5 anchor (e.g. what to do when DQ checks fail). | 4 / 5 |
Progressive Disclosure | The body is a well-organized overview and correctly offloads the bulk per-table field lists to a single, clearly signaled, one-level-deep reference ("See references/omop_cdm_v5_4_fields.md for the full per-table field list" — the file exists and delivers exactly that). No nested references; navigation is easy. | 5 / 5 |
Total | 17 / 20 Passed |