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.
A well-structured, actionable workflow with concrete filter syntax, per-source mappings, and ready-to-use response templates for the common cases. Its weaknesses are redundancy — the Notes section and query-parsing guidance repeat or re-teach what the numbered steps and Claude's own competence already cover — and a missing explicit verification checkpoint between ranking and presenting results.
Suggestions
Cut the 'Notes' section or reduce it to only the two points not already stated in steps 1-7 (e.g., the single-source and person-mention handling), removing the restated parallel-search, synthesis, attribution, and filter bullets.
Trim the step 2 query-parsing bullets ('Intent', 'Entities', 'Time constraints') to just the filter syntax definitions, trusting Claude to parse a query without the taxonomy.
Add an explicit checkpoint between steps 5 and 6: verify the top-ranked results actually answer the parsed intent (and that dedupe merged true duplicates) before formatting the unified response.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient but includes unnecessary explanation and repetition: query parsing guidance ('Intent', 'Entities', 'Time constraints') teaches Claude what it already knows, the 'Notes' section restates instructions from earlier sections ('Always search multiple sources in parallel — never sequentially' duplicates step 4; 'Synthesize results...' duplicates step 6), and 'Use natural language queries for semantic search when appropriate' is inferable filler. Not 4 because these redundant/padded sections are more than minor; not 2 because most sections carry real, skill-specific guidance. | 3 / 5 |
Actionability | For an instruction-only skill the guidance is fully concrete: explicit filter syntax (from:, in:, after:, before:, type:) with per-source translation mappings, and copy-paste-ready response templates covering factual, exploratory, find, no-results, partial-results, and ambiguous-query cases. The absence of literal tool names is justified flexibility since MCP tools vary; not 4 because the templates and filter grammar are specific enough to execute directly across the common cases. | 5 / 5 |
Workflow Clarity | A clear 7-step sequence (check sources → parse query → decompose → parallel execute → rank/dedupe → present → edge cases) with failure handling ('do not let one failure block others', partial-results template with retry suggestions). Not 5 because there is no explicit checkpoint verifying results actually answer the query intent before presenting, and the dedupe/rank step lacks a concrete validation loop; not 3 because failure recovery and error-tolerant execution are explicitly addressed throughout. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ absent), and the single well-sectioned SKILL.md with a clearly signaled one-level-deep pointer to CONNECTORS.md is good structure with minor organization gaps. Not 5 because at ~170 lines the response templates and per-source translation guides are borderline inline content that could be split into reference files; not 3 because what is present is clearly sectioned, navigable, and references are not buried. | 4 / 5 |
Total | 16 / 20 Passed |