Content
50%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 skill has a solid core — a clear four-domain QUADAS-2 workflow, a mandatory output template, quality rules, and a working PDF extraction script — but it is buried under roughly half a document of auto-generated template boilerplate, a broken example path, a dangling cross-reference ('See ## Workflow above' when Workflow appears below), and placeholder bundle files. The description is strong; the body needs a significant pruning pass and integration of its own bundle.
Suggestions
Delete the auto-generated boilerplate sections ('When to Use' bullets that restate the description, 'Key Features', 'Implementation Details', 'Dependencies' filler) — they consume tokens without adding guidance Claude doesn't already have.
Fix broken navigation: correct 'See ## Workflow above' (Workflow appears below), replace the nonexistent 'cd "20260316/scientific-skills/..."' example with a bundle-relative path, and replace 'copy and send to me' with a direct instruction to extract text and proceed to assessment.
Reconcile the bundle: either wire scripts/quadas_assessment.py into the workflow or remove it, and replace or delete the placeholder files (references/api_reference.md, assets/example_asset.txt) so the only signaled reference is the real criteria document.
Add an explicit completeness check at the end of the workflow (e.g., verify all signaling questions for all four domains are answered with quotes before producing the final report) to convert the implicit validation into a real checkpoint.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Large portions of the body are auto-generated boilerplate that adds no guidance: 'When to Use' restates the description almost verbatim across six padded bullets, 'Key Features' is pure meta-fill ('Packaged executable path(s)... plus 1 additional script(s)', 'Reusable packaged asset(s)'), and 'Implementation Details' offers generic filler ('validate the request, choose the packaged workflow, and produce a bounded deliverable'). This matches anchor 2 — several unnecessary padded sections — rather than 3, because the padding occupies roughly half the document and repeats the description rather than explaining concepts Claude doesn't know. | 2 / 5 |
Actionability | The core guidance is executable: numbered per-domain assessment steps, an exact mandatory output template, and runnable commands ('pip install PyPDF2', 'python scripts/pdf_extractor.py "paper.pdf" 5 15') pointing at a real script. Minor gaps keep it below 5: the Example Usage 'cd "20260316/scientific-skills/..."' path doesn't exist in the bundle, scripts/quadas_assessment.py is never used, and 'which you can then copy and send to me' is leftover conversational phrasing. | 4 / 5 |
Workflow Clarity | The assessment sequence itself is listed (four domain steps with a pointer to references/quadas_2_criteria.md) and the Output Format's 'Repeat for all signaling questions' acts as a partial checklist, but checkpoints are otherwise implicit — there is no verification that extraction succeeded or that every signaling question is answered, and the boilerplate 'Example run plan'/'Implementation Details' sections present competing pseudo-workflows that muddy the real path. This sits at anchor 3, not 4, because the validation that exists is implied by formatting rules rather than explicit steps. | 3 / 5 |
Progressive Disclosure | The one substantive reference (references/quadas_2_criteria.md) is clearly signaled and one level deep, but the rest of the bundle is placeholder noise — references/api_reference.md, assets/example_asset.txt, and scripts/quadas_assessment.py are placeholder files referenced only in meaningless boilerplate or not at all, so navigation to real detail is only partially organized. This fits anchor 3 ('could be better organized; references present but not clearly signaled') better than 4. | 3 / 5 |
Total | 12 / 20 Passed |