Content
60%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 instruction skill with a clear sequenced workflow, a mandatory output template, and genuinely useful one-level-deep reference files. Its main liability is heavy redundancy — at least four body sections restate content that already exists elsewhere in the body or in the references — which roughly doubles the token cost without adding guidance.
Suggestions
Delete or merge the redundant sections: "Hard Rules" vs references/hard-rules.md, "Core Function" vs "Task", and "What This Skill Should Not Do" vs "Scope Boundary" — keep one authoritative copy of each rule set.
Trim "Sample Triggers" and "Important Distinctions" or move them into the relevant reference files to cut the body's token footprint substantially.
Add one short worked example of a consistency finding rendered in the mandatory A–H output structure to make the template fully concrete.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably padded through structural duplication: "Hard Rules" (10 rules) repeats almost verbatim in references/hard-rules.md and again in "What This Skill Should Not Do"; "Core Function" restates the earlier "Task" list; "Input Validation" and "Sample Triggers" duplicate Step 1 of Execution; and "Important Distinctions" overlaps the severity/drift reference rules. It avoids score 1 because it never explains domain concepts Claude already knows — the padding is redundancy, not basic-concept tutorials — but several sections could be deleted outright. | 2 / 5 |
Actionability | For an instruction-only skill, the guidance is mostly concrete: an 8-step execution sequence, a mandatory output template with exact section headers (A–H), a named detection taxonomy ("sample-size inconsistencies, shifting endpoint wording, population-definition drift, figure/table numbering mismatch"), and explicit severity classes. It misses score 5 because no worked example of a finding in the required output format is provided, and the actual comparison criteria are deferred to the reference files. | 4 / 5 |
Workflow Clarity | Steps 1–8 are clearly sequenced with an explicit input-sufficiency checkpoint ("If the user provides only a vague request... do not immediately produce a full review") reinforced by output section A ("Input Match Check") and section H (missing-input list). It falls short of score 5 because there is no verification/feedback loop on its own findings — no step instructing re-confirming each flagged numerical contradiction against both source sections before reporting it. | 4 / 5 |
Progressive Disclosure | The "Reference Module Integration" section signals all eight one-level-deep reference files with paths and per-file use cases, and all cited paths exist in references/ with no nested references inside them. It misses score 5 because content is not cleanly split — the body's 10-line "Hard Rules" section fully duplicates references/hard-rules.md, creating ambiguity about which is authoritative, and paths are code-formatted rather than linked. | 4 / 5 |
Total | 14 / 20 Passed |