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.
The body delivers a disciplined, actionable 8-step review workflow with concrete templates, logging formats, and a citation-verification gate. It falls short of top marks on conciseness (some textbook concept restatement), actionability (no concrete search-execution commands), and workflow clarity (no search-revision feedback loop).
Suggestions
Trim concept restatement: compress the Review Types definitions and PICO bullets to one-line references since Claude already knows these, keeping only the decision guidance on which rigor level to pick.
Add concrete search-execution examples per database (e.g., PubMed E-utilities URL and arXiv API query) so Step 2-3 are mechanically executable rather than only well-planned.
Add a feedback loop after screening: if the yield is too low or overwhelmingly off-topic, revise the search strings or broaden criteria and log the re-run, mirroring the validate-fix-retry pattern.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and procedural with compact tables and checklists, but the Review Types definitions and the PICO bullet breakdown restate concepts Claude already knows, fitting the minor-over-explanation anchor rather than the every-token-earns-its-place one. | 4 / 5 |
Actionability | Concrete, copy-paste-ready guidance throughout — ordered dedup rules, staged screening with explicit exclusion reasons, markdown templates, and a real PubMed query string in the search-log example — but no concrete mechanism for actually executing searches per database, keeping it below the fully-executable anchor. | 4 / 5 |
Workflow Clarity | Eight clearly numbered, well-sequenced steps with an explicit pre-finalization validation step (Step 8: verify citations) and checklists, but there is no feedback loop for error recovery such as revising the search when screening yields too few or too many results. | 4 / 5 |
Progressive Disclosure | A single self-contained file with clear, well-organized sections and no nested references, which is appropriate; however, at ~185 lines with no bundle, the output template and example tables are borderline candidates for a separate reference file, so structure is good rather than optimal. | 4 / 5 |
Total | 16 / 20 Passed |