Content
52%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 encodes a genuinely thoughtful, domain-aware policy (venue tiering, source ordering, layer-aware synthesis) with a clear workflow and good degradation handling. Its central structural flaw is that it is a monolith: the well-designed reference bundle is never referenced, so ~200 lines of its content duplicate the separate files, hurting both token efficiency and navigation. Adding one or two concrete executable examples would close the remaining actionability gap.
Suggestions
Link the existing bundle instead of inlining it: replace the full 'Venue Priority' tier lists, 'Source Selection'/'External Search Policy' details, and 'Output' table specification with one-line pointers to references/venue-tiering.md, references/source-policy.md, references/output-template.md, and references/domain-taxonomy.md — these files exist but are never mentioned in the body.
Add concrete executable examples: a sample Zotero MCP search invocation, one or two example IEEE Xplore/ScienceDirect query strings, and a filled example row of the literature table so the expected output shape is unambiguous.
De-duplicate within the body: the source values are listed twice in 'Source Selection', the database ladder appears in both 'Retrieval Order' and 'External Search Policy', and 'Key Rules' restates rules already stated in 'Graceful degradation rules', 'Usage rules', and 'Publication policy'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient prose with no padding of concepts Claude already knows, but it is ~295 lines and could be tightened substantially: valid source values are listed twice, the database ladder appears in both 'Retrieval Order' and 'External Search Policy', venue-tier rules are stated in full and then restated in 'Step 1' and 'Key Rules', and ~200 lines duplicate content that already exists in the unreferenced references/ files. Not 4 because the duplication is more than minor trimming; not 2 because every line carries skill-specific policy rather than generic explanation. | 3 / 5 |
Actionability | Concrete guidance exists (file globs like 'papers/**/*.pdf', extraction field list, exact table columns, ordered synthesis structure), but key executable details are missing for an instruction-only skill: no example Zotero/Obsidian MCP invocation, no sample search query strings for IEEE Xplore/ScienceDirect, and no filled example table row showing the expected output. Matches 'some concrete guidance but incomplete / missing key details'; not 4 because these gaps go beyond minor, not 2 because the steps and formats given are specific and followable. | 3 / 5 |
Workflow Clarity | The sequence is clear and ordered (Step 0a Zotero → 0b Obsidian → 0c local scan → 1 external ladder → 2 extraction → synthesis → output), and graceful-degradation rules ('If an unrequested source is unavailable... skip it silently'; 'Never fail because Zotero or Obsidian MCP is missing') act as explicit checkpoints. Not 5 because there are no validation/verification loops on the output itself (e.g., checking the table is populated or evidence was actually found before synthesizing); not 3 because checkpoints for the riskiest step (source unavailability) are explicit rather than implicit. | 4 / 5 |
Progressive Disclosure | The bundle contains four reference files (references/venue-tiering.md, source-policy.md, output-template.md, domain-taxonomy.md) whose content substantially duplicates body sections, yet the body never links or mentions any of them — the references are completely orphaned rather than merely 'not clearly signaled'. Content that clearly belongs in separate files (the full Tier A/B/C venue lists, source policy, output template) is inlined in SKILL.md. Not 3 because there is zero navigation to the existing bundle files, which is the worse of the two failure modes the anchors describe. | 2 / 5 |
Total | 12 / 20 Passed |