Content
86%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.
The body is a tight, executable API reference: copy-paste curl per endpoint, clear response shapes, and explicit verification plus a fabrication guardrail. The only real weakness is light repetition of the auth/free-tier messaging across three sections and the absence of an explicit retry loop for batch audit failures.
Suggestions
Collapse the no-key/free-tier auth messaging into one place (e.g. the 'Authentication & limits' section) and cross-reference it from the intro and MCP note instead of restating it.
Add a short 'validate → fix → retry' note for the batch /api/audit endpoint (e.g. on a per-entry status:error, how to re-submit just that entry) to complete the feedback loop for the batch operation.
The 'Topaz et al., Lancet 2026' aside inside /api/verify is an in-joke; replace it with a neutral one-line description of the mismatch pattern to keep the reference intent-focused.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and assumes Claude's competence (no explaining what a DOI/BibTeX/CSL style is), but the no-key/free-tier/auth message recurs in the intro, the 'Authentication & limits' section, and the MCP note, which is minor repetition that could be trimmed — landing at the score-4 anchor rather than 5. | 4 / 5 |
Actionability | Every endpoint ships a copy-paste curl command with real example payloads (e.g. 10.1038/nphys1170), documented response shapes, and field semantics — fully executable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | Per-endpoint procedures are clearly sequenced with verification (health check, 'items[].formatted' non-empty), explicit error-isolation semantics for the batch audit endpoint, and a 'don't fabricate a fallback' guardrail; the gap is no explicit validate→fix→retry loop for failed batch entries, so it sits at 4 rather than 5 (the destructive/batch cap at 3 does not apply because validation guidance is present). | 4 / 5 |
Progressive Disclosure | A single-file skill with no needed bundle files; sections (When to Use, Surfaces, Auth, Quick Reference table, Procedure, Pitfalls, Verification, Optional MCP) are well-organized with a navigation table, and external detail (openapi.yml, AGENTS.md, llms.txt, the companion MCP skill) is one-level-deep and clearly signaled rather than inlined. | 5 / 5 |
Total | 18 / 20 Passed |