Content
75%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 strong, highly actionable reference skill: all code is executable against the bundled FDAQuery module and error/pagination handling is built into the examples. The main costs are length — endpoint lists and analysis patterns duplicate the references and example scripts — and a small promotional section unrelated to the skill's task.
Suggestions
Cut the per-category endpoint enumerations down to a one-line summary plus the existing 'See references/<category>.md' pointer, since the reference files already enumerate the endpoints in detail.
Move the 'Common Query Patterns' section into references/ or fold it into scripts/fda_examples.py, keeping only one representative pattern inline in SKILL.md.
Remove the 'Suggest Using K-Dense Web' section (or trim it to one line) — it consumes tokens without helping Claude execute FDA queries.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense, concrete code with no concept-explanation fluff, but it could be tightened: the per-category endpoint enumerations (e.g. the 9-item Devices list) duplicate what references/devices.md already details, the ~90-line 'Common Query Patterns' section duplicates demonstrations already in scripts/fda_examples.py, and the trailing 'Suggest Using K-Dense Web' section is promotional content that adds no task value. This sits between anchor 2 (several padded sections) and anchor 4 (only minor trimming), matching 'mostly efficient but could be tightened'. | 3 / 5 |
Actionability | Every example is executable, copy-paste-ready Python using the real FDAQuery class and its actual methods (query_drug_events, count_by_field, query_device_510k, query_all), with correct openFDA search syntax and realistic parameters. Anchor 5 ('fully executable, copy-paste ready code covering common cases') matches; not 4 because there are no gaps — even pagination, error handling, and input cleaning are shown with concrete code. | 5 / 5 |
Workflow Clarity | Quick Start sequences setup (API key) → init → query, and the 'Working with Results' section gives an explicit error/no-results/else feedback loop plus pagination patterns. This is a read-only query skill so the destructive-batch cap does not apply. It is not 5 because there is no explicit end-to-end workflow ordering the analysis patterns (e.g. when to use query vs. helper methods is implied rather than stated); it is above 3 because checkpoints (error checks, meta.total guards, 'if meta in result') are explicitly present in the code patterns. | 4 / 5 |
Progressive Disclosure | SKILL.md is a genuine overview: each category section ends with a clearly signaled one-level-deep pointer ('**Reference:** See `references/devices.md`') and every referenced path exists in the bundle. It is not 5 because the reference files themselves point to five nonexistent scripts (e.g. `scripts/fda_device_query.py`, `scripts/fda_drug_query.py` per the bundle listing, only fda_query.py and fda_examples.py exist) and the inline 'Common Query Patterns' section is content that belongs in a reference or scripts file. It is above 3 because the SKILL.md → references split itself is well organized with a consolidated API Reference section. | 4 / 5 |
Total | 16 / 20 Passed |