Content
62%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 skill has an unusually well-structured control flow — explicit mode branching, scene sequencing, validation loops, and error-recovery paths — but its craft guidance is hollow at the point of use: all substantive writing references are deferred to files absent from the bundle, and several rules are restated many times across sections. As shipped, following the ACQUIRE step or the AUDIT step fails because the cited resources do not exist.
Suggestions
Bundle the referenced files or inline their essential tables: every path cited in Dependencies and References (resources/anti-ai-checklist.md, sentence-structure-reference.md, academic-verb-tiers.md, hedging-guide.md, and ../_shared/core/*.md) is missing from the skill directory, so the ACQUIRE, AUDIT, and VERIFY steps cannot be executed as written.
Consolidate repeated rules: the weaken-or-remove-unsupported-claims rule, the non-English handoff, and the quote-the-rubric rule each appear 3-6 times across Control-flow features, Transitions, Failure and recovery, Guardrails, and the Canonical workflow path; stating each once (with cross-references) would cut the body substantially.
Add one worked example per mode (e.g., a before/after revision block showing a verb upgrade and hedge recalibration, and a sample Claim-Evidence Map row with Status values) so the verb, hedging, and evidence guidance is actionable even when the reference files are not loaded.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is directive rather than padded with concepts Claude already knows, but key rules repeat many times: "weaken or remove unsupported claims" appears in Control-flow, Transitions, Failure and recovery, Actions, Canonical workflow, and Guardrails; the non-English handoff and quote-the-rubric rules likewise recur 3-4 times. Mostly efficient but could be tightened considerably, matching anchor 3; anchor 2 would require extensive explanation of known material, which is absent. | 3 / 5 |
Actionability | Some guidance is concrete and executable ("Em dashes ≤ 1 per paragraph; semicolons ≤ 2 per 1000 words", the 3-column Claim-Evidence Map, per-mode output formats), but the core craft guidance — verb tiers, hedge calibration, sentence-type patterns — is entirely deferred to resources/ files that are missing from the bundle, and there are no worked before/after examples. This lands on anchor 3 (some concrete guidance but incomplete, key details missing) rather than 4. | 3 / 5 |
Workflow Clarity | The multi-step process is clearly sequenced with explicit validation checkpoints and feedback loops: PREPARE→ACQUIRE→ACT→VERIFY→FINALIZE scenes, a canonical READ→PLAN→DRAFT→AUDIT→REVERSE-OUTLINE→POLISH path, "every fix loops back through the anti-AI checklist before emit", read-aloud verification before emit, defined exit criteria, and a failure-and-recovery table with retry actions. This matches anchor 5; the skill is not a destructive or batch operation, so no cap applies. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and one level deep with a dedicated References section and per-file purpose notes, but the actual bundle contains none of the seven referenced files (resources/{anti-ai-checklist, sentence-structure-reference, academic-verb-tiers, hedging-guide}.md and ../_shared/core/{anti-ai-prose, context-loading, quality-principles}.md), so navigation leads to missing files. Scoring against the actual bundle structure per the guidelines: some structure, but the referenced content is not present and the 178-line body inlines process detail that could be split out — anchor 3. | 3 / 5 |
Total | 14 / 20 Passed |