Content
53%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 core operational content is solid — an executable script invocation, a clear output format, and a well-signaled reference file holding the detailed signaling questions. However, roughly half the body is templated boilerplate (auto-generated "Key Features"/"Implementation Details" sections, a duplicated When-to-Use list, a date-stamped directory path) that inflates token cost without adding actionable guidance, and the workflow is split between two overlapping sections with only implicit validation checkpoints.
Suggestions
Delete the boilerplate sections ("When to Use" bullet list, "Key Features", "Dependencies", "Implementation Details") and keep only the substantive "When to Use This Skill" conditions; this would cut the body roughly in half.
Remove the date-stamped "20260316/scientific-skills/..." cd path and the py_compile/--help steps; show a real end-to-end run (e.g., extract_pdf.py paper.pdf then quadas_c.py --file extracted.txt) instead.
Fix the "See ## Usage above" reference (Usage is below), add a validation checkpoint after the run (e.g., confirm the JSON contains all four domains with valid Low/High/Unclear values), and merge the two overlapping workflow sections into one.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably padded: generic boilerplate sections ("Key Features", "Dependencies", "Implementation Details") add no skill-specific information, the "When to Use" section repeats the frontmatter description nearly verbatim as filler bullets, and a date-stamped path "20260316/scientific-skills/Data Analytics/..." injects time-sensitive cruft. This matches anchor 2 (several unnecessary/padded sections); it is not anchor 1 because the core sections (Usage, Execution, Output Format) are not explaining concepts Claude already knows, and not anchor 3 because the padding spans roughly half the document. | 2 / 5 |
Actionability | Provides an executable command ("python scripts/quadas_c.py --file 'path/to/paper.txt'"), a concrete JSON output schema (P/I/R/FT keys with Low/High/Unclear), a documented PDF extraction helper, and a real references/prompts.md file. Minor gaps keep it below anchor 5: the bash example runs py_compile and --help instead of an actual assessment, the "Example run plan" includes vague conditionals ("if the script uses fixed settings"), and direct command-line text input is mentioned in prose but never shown. | 4 / 5 |
Workflow Clarity | A sequence is present (Extraction → Assessment → Signaling Questions → Risk of Bias → Reporting) plus a run plan with input confirmation and output review, but checkpoints are implicit ("Review the generated output" with no criteria), the steps describe what the script does internally rather than what the agent executes, and the workflow is duplicated across two competing sections ("Example run plan" vs "Usage"). This matches anchor 3 (sequence present, checkpoints missing/implicit) rather than anchor 4's mostly-explicit checkpoints. | 3 / 5 |
Progressive Disclosure | The body appropriately points to a real, one-level-deep reference ("See references/prompts.md for the specific signaling questions and risk of bias criteria") and both referenced scripts exist in the bundle. Minor organization gaps (duplicate "When to Use" / "When to Use This Skill" sections, and a wrong forward reference "See ## Usage above" for a section that appears later) keep it below anchor 5. | 4 / 5 |
Total | 13 / 20 Passed |