Content
67%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 well-structured, largely actionable guide with a numbered workflow, an error-handling section, and concrete API payloads. Its main weaknesses are token overhead from test-scaffolding sections ("Contract", "Output Format") and routing info duplicated from the frontmatter, plus a few hand-wavy steps (put_page, monitoring cron) that keep it from fully copy-paste readiness.
Suggestions
Remove or relocate the 'Contract' and 'Output Format' sections — content that exists 'for the conformance test' adds tokens without operational value for the agent; fold any real guarantees into the relevant sections.
Make step 4 executable: replace 'gbrain put research/<slug> # via the put_page operation' with the actual put_page command/payload, and give the deal-monitoring cron pattern a concrete runnable loop instead of a comment-only bash block.
Add a validation checkpoint between the Perplexity call and put_page (e.g., confirm the response contains citations and a non-empty delta before writing the research page), and trim the duplicated routing table and 'The key insight' paragraph to tighten token budget.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Core sections (presets table, invocation, output template) are tight, but the trailing "Contract" and "Output Format" sections explicitly exist "for the conformance test" and add no operational value, the "When to use this vs other tools" table duplicates frontmatter routing, and "The key insight" paragraph over-explains. Not 2 because the bulk of the body is efficient and free of concept explanations Claude already knows. | 3 / 5 |
Actionability | Provides an executable curl call to POST /v1/agent with exact JSON, preset values, the recency-filter JSON shape, and a full output page template. Not 5 because step 2 is a comment template rather than a runnable command, "gbrain put research/<slug> # via the put_page operation" is vague, and the deal-monitoring pattern contains no actual command. | 4 / 5 |
Workflow Clarity | The Invocation section gives a clear 5-step sequence and "When it fails" supplies error-recovery paths (missing key, rate_limited retry, degraded brain context). Not 5 because there is no explicit validation checkpoint on the result (e.g., confirming citations/delta exist) before writing the page via put_page. | 4 / 5 |
Progressive Disclosure | Well-sectioned overview with clearly signaled, one-level-deep references (conventions/quality.md, brain-first.md, related skills). Not 5 because the referenced convention files live outside the skill directory (not verifiable as part of the bundle), and the full page template plus API-migration detail is inlined where a reference file could carry the bulk. | 4 / 5 |
Total | 15 / 20 Passed |