Content
73%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-sequenced, validation-rich planning workflow with proper one-level-deep reference structure and real bundle files behind every link. The dominant cost is redundancy — the Hard Rules section and the A–J section contracts restate step-level requirements repeatedly, inflating token cost without adding instruction.
Suggestions
Collapse Hard Rules 15–23 into a short 'Mandatory outputs' checklist or reference, since they verbatim restate requirements already stated in Steps 3–7 and Sections C.5, D, G, I, and J (e.g. rules 22 and 23 duplicate Section D's disclaimer and Section J's mandate).
Merge the 'Execution — 7 Steps' section with 'Mandatory Output Sections (A–J)' so each step states once what it must produce, instead of describing the deliverable twice (Step 4 = Section C, Step 4.5 = Section I, Step 6 = Section D).
Inline a one-line summary of the 8-field workflow template (the core deliverable format) next to the reference link in Step 6, so the most important output contract is visible without opening references/workflow-step-template.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body never explains MR concepts Claude already knows (good), but it is noticeably padded by restatement: Hard Rules 15–23 restate requirements already given verbatim in Steps 3–7 and Sections C.5/G/I/J (e.g. rule 22 repeats the Dataset Disclaimer rule from Section D, rule 23 repeats Section J's mandate), and the Steps section and the Mandatory Output Sections A–J re-cover the same ground (Step 3 = Section B, Step 4 = Section C, Step 4.5 = Section I). This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than the noticeably-verbose level 2, since the padding is rule duplication, not concept explanation. | 3 / 5 |
Actionability | Guidance is concrete for an instruction-only skill: verbatim redirect script ('This skill designs ... use an appropriate clinical or epidemiology framework'), exact dependency formulas ('exposure GWAS + outcome GWAS + IVW + sensitivity consistency'), pattern-selection tables, and a fully enumerated output contract (Sections A–J with per-section requirements). Per the rubric's code-vs-instruction note, absent code is not penalized. Not a 5: the pivotal 8-field workflow template is entirely deferred to references/workflow-step-template.md, so the core deliverable format is not visible in the body, and rule numbering skips 10, a small internal-consistency slip. | 4 / 5 |
Workflow Clarity | Seven steps that 'always run in order', with an explicit validation checkpoint: Step 5 'Dependency Consistency Check (mandatory before output)' is a checklist ('Does bidirectional design appear without separate IV construction in both directions?') with a feedback loop ('If any dependency inconsistency is found, revise the plan before outputting'). Input validation with a stop-and-redirect, stated-default fallback in Step 1, and the Section J self-critical review add further checkpoints — this matches the top anchor (clear sequence, explicit validation, feedback loops, checklists). | 5 / 5 |
Progressive Disclosure | The body is a genuine overview with eight clearly signaled one-level-deep links ('→ Detailed pattern logic: [references/study-patterns.md]'); all eight referenced files exist, and the reference files contain no further nested links. Not a 5: the main body carries substantial detail (23 Hard Rules and full per-section output contracts) that partly duplicates step content and could be consolidated, so placement is good but not fully clean. | 4 / 5 |
Total | 16 / 20 Passed |