Content
68%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 admirably token-efficient and gives a concrete review checklist and output contract, but the workflow sequencing and validation checkpoints are implicit rather than explicit, and progressive disclosure is broken: all three referenced files point outside the skill bundle and do not exist. Fixing the reference paths (or inlining the essential policy) and numbering the workflow steps would lift the two weakest dimensions substantially.
Suggestions
Move the policy content into a references/ directory inside the skill bundle and fix the links (../../skill/references/review-policy.md, ../../skill/references/evidence-grounded-review.md, ../EDITION-MANAGEMENT.md all point outside the bundle and resolve to nothing), or inline the essential policy rules if the files are not shipped.
Number the workflow steps explicitly (identify version/scope → read policies → trace claims → check checklist items → report prioritized findings; revision: retain prior version and findings → correct sources → recheck) and specify how the 'recheck the resulting artifacts' validation is performed and what passing looks like.
Add a brief example of the expected findings format (severity, manuscript/figure location, supporting evidence, actionable correction) and how rendered pages should be inspected, to make the output guidance fully executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~150-word body is lean and dense: every sentence instructs ("Identify the exact manuscript version", "Follow the claims from abstract and main figures to source results", "Check contribution, controls, units, uncertainty..."), with zero explanation of concepts Claude already knows. It fully assumes competence; no padding to trim. | 5 / 5 |
Actionability | Guidance is mostly concrete for an instruction-only skill: a specific review checklist ("contribution, controls, units, uncertainty, interpretation, references and readability of figures or tables"), a tracing procedure ("follow the claims from abstract and main figures to source results and actual citation passages"), and defined output expectations ("prioritized findings with manuscript/figure locations, supporting evidence and actionable corrections"). Minor gaps keep it from anchor 5: no example findings format and no method for "inspect rendered pages". | 4 / 5 |
Workflow Clarity | A rough sequence exists (identify version/scope → read policies → trace claims → inspect pages → check items → report findings; revision path: retain earlier version and findings → correct affected sources → recheck artifacts), but it is implicit and interleaved rather than clearly ordered, and the one validation checkpoint ("recheck the resulting artifacts") is unspecified — recheck against what, and how. This matches anchor 3 (sequence present, checkpoints missing or implicit) and falls short of anchor 4's clear sequence with most checkpoints present. | 3 / 5 |
Progressive Disclosure | The body delegates core policy to three referenced files — ../../skill/references/review-policy.md, ../../skill/references/evidence-grounded-review.md, and ../EDITION-MANAGEMENT.md — but no references/, scripts/, or assets/ directories exist and none of these paths resolve; they escape the skill bundle two levels up, so the instructed reading is unreachable. Combined with a total absence of section structure (three unheaded paragraphs), this matches anchor 2's minimal/broken structure; it avoids anchor 1 only because the links are clearly signaled and not nested. | 2 / 5 |
Total | 14 / 20 Passed |