Content
96%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.
An exceptionally well-crafted instruction-only skill body: lean, assumption-respecting prose with concrete formats, limits, and commands, clearly sequenced workflows with real validation checkpoints, and a routing-table-driven progressive disclosure design. The only structural nit is that detailed examples live two levels deep under the section guides, slightly beyond the ideal one-level reference depth.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body contains no concept explanations Claude already knows (no 'what a paper is', no library tutorials) — every line is a directive, a rule, or routing, e.g. "Match the user's situation and load ONLY the needed reference (do not preload all)" and "Never fill gaps by inventing." This is the score-5 anchor (lean, assumes competence, every token earns its place); score 4 would require identifiable over-explanation to trim, and there is none. | 5 / 5 |
Actionability | Concrete, executable guidance throughout: exact placeholder formats ("[XX.X]", "[CITE: sparse-view NeRF methods]"-style), concrete commands ("locate sections with grep (\section, \begin{abstract})"), exact limits ("at most 3 focused questions in one message", "3–7 bullet mini-outline"), and a concrete output format ("Claim: ... | Evidence: ... | Status: supported / needs evidence / weakened"). Per the code-vs-instruction scoring note, an instruction-only skill with guidance this specific is fully actionable — the score-5 anchor equivalent; score 4 would require missing key details, and the routing table plus workflows cover the common cases. | 5 / 5 |
Workflow Clarity | Three clearly sequenced workflows by request size (A quick polish, B section draft, C pre-submission review) with explicit validation checkpoints: "Reverse-outline the result: thesis → topic sentences → evidence; fix anything that doesn't map", the claim-evidence map with status tracking, "If the paragraph's real problem is structural... say so instead of cosmetically polishing", and "Pick engine, build, verify output". This matches the score-5 anchor (clear sequence, explicit validation, feedback loops, checklist for the review workflow); it is not score 4 because validation is explicit rather than implied at each stage. | 5 / 5 |
Progressive Disclosure | Good structure: SKILL.md is a pure overview with a situation→reference routing table, an explicit "load ONLY the needed reference" instruction, and a References section with one-line descriptions; all 25 referenced paths verify to real files. However, detail sits two levels deep — SKILL.md → references/<section>.md → references/examples/<topic>/*.md (e.g. method.md cites 10 example files) — which is beyond the one-level-deep ideal of the score-5 anchor. Fits score 4 (good structure, most content appropriately placed, minor organization gap); not score 3 because navigation is easy and nothing is buried or inlined that belongs in a separate file. | 4 / 5 |
Total | 19 / 20 Passed |