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-only skill with a genuinely actionable eight-step workflow, a mandatory output template, and a clean one-level-deep reference bundle. Its dominant weakness is token efficiency: the same planning cycle is restated across at least four sections, and hard rules and input-validation thresholds are duplicated between SKILL.md and the reference files.
Suggestions
Collapse the "Task", "Core Function", and "Quality Standard" lists into one statement of purpose plus the mandatory output structure — the planning cycle is currently described four times.
Replace the 12 inline "Hard Rules" with a short pointer (e.g., "Obey references/hard-rules.md at every step") and delete the in-body "Input Validation" list, which restates references/input-clarification-thresholds.md; keeping rules in two places invites drift.
Merge "Scope Boundary" and "What This Skill Should Not Do" (they overlap substantially) and add a brief worked mini-example of the memo for one common case (e.g., two-group binary endpoint) to lift actionability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably padded with self-duplication rather than explanations of known concepts: the "Task" list, "Core Function" list, "Mandatory Output Structure" (A–L), and "Quality Standard" sections restate the same planning cycle four times; "Scope Boundary", "What This Skill Should Not Do", and the 12 "Hard Rules" overlap heavily; the in-body "Hard Rules" near-duplicate references/hard-rules.md (10 rules) and "Input Validation" restates references/input-clarification-thresholds.md. This is "several unnecessary or padded sections" (anchor 2), not 3 — the redundancy is extensive and structural — but not 1 since it never teaches concepts Claude already knows. | 2 / 5 |
Actionability | For an instruction-only skill, the guidance is concrete: eight sequenced execution steps, a mandated 12-section output structure with defined contents per section, explicit assumption-quality labels, and enumerated design families, fragility threats, and fallback options. Minor gaps keep it below 5: no worked example or filled-in sample memo, and Section H ("Explain the main reasoning path") leaves the hardest part — the actual sizing logic — unspecified beyond pointing to the design-family reference. | 4 / 5 |
Workflow Clarity | Steps 1–8 are clearly sequenced (clarify → identify driver → select family → audit assumptions → choose stance → build scenarios → identify fragility → produce memo) with an explicit clarification gate in Step 1 and Input Validation, and Section L builds a self-critical review into the output. It stops short of 5 because there is no post-draft feedback loop (e.g., re-audit or revise if assumptions conflict) and steps 2, 6, and 7 are consideration lists rather than decision procedures. | 4 / 5 |
Progressive Disclosure | Five real reference files exist, each exactly one level deep, each introduced in "Reference Module Integration" with a clear purpose ("Use before any long-form answer", "Apply throughout the entire response"), and each verified to contain substantive, non-stub content. Below 5 because content that should live solely in the references is also inlined in SKILL.md — the 12 Hard Rules duplicate references/hard-rules.md and the Input Validation section duplicates references/input-clarification-thresholds.md — creating two sources of truth. | 4 / 5 |
Total | 14 / 20 Passed |