Content
73%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 clear, actionable orchestration workflow with strong error handling and an explicit output contract. Its main weaknesses are redundancy — the verdict taxonomy appears three times and two boilerplate sections exist only for a conformance test — and content that could be split into a reference file.
Suggestions
State the five-verdict taxonomy once (in the Standards section, which has the definitions) and reference it from Step 4 and the output template instead of repeating the full list three times.
Remove or collapse the 'Contract' and 'Output Format' sections that exist only for the conformance test, folding their one-line guarantees into the existing body sections.
Provide the exact citation-mode prompt text to send to perplexity-research (or a short worked example of a claim → verdict page) so Step 3 is copy-paste ready rather than outlined.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient domain-specific content, but the five-verdict taxonomy is stated three times (Step 4, the output template, and the Standards section), the 'What this is' section repeats the description's trace chain, and the trailing 'Contract' and 'Output Format' sections exist solely 'for the conformance test' — pure padding. | 3 / 5 |
Actionability | Gives the exact `gbrain query "<paper title> OR <author name> OR <claim keywords>"` command, a copy-paste brain-page template, database URL patterns, and precise failure handling; the minor gap is that the perplexity-research prompt is outlined rather than supplied verbatim, and no worked example of a claim → verdict is shown. | 4 / 5 |
Workflow Clarity | A clear five-step sequence (scope → brain-first → perplexity citation-mode → format verdict → cross-link) with explicit validation criteria (the verdict standards and anti-patterns) and concrete error-recovery feedback loops: retry brain lookup with author/journal/year, retry once after the stated delay on rate_limited, and poll `gbrain write-request <request_id>` instead of double-writing. | 5 / 5 |
Progressive Disclosure | A single-file skill with well-organized sections and clearly signaled one-level references to conventions and perplexity-research; the inlined database table and full page template could live in a reference file at this length (~215 lines), and the test-oriented trailing sections weaken the structure. | 4 / 5 |
Total | 16 / 20 Passed |