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 defines a rigorous, well-validated multi-phase workflow with explicit checkpoints, feedback loops, quantified style rules, and clear handoff boundaries. However, its core writing techniques live in referenced files that are missing from the bundle, it contains no worked examples, and the workflow is described redundantly across four parallel representations.
Suggestions
Ship or inline the five missing resources/ files (anti-ai-checklist, sentence-structure-reference, academic-verb-tiers, hedging-guide, abstract-guide) — the ACQUIRE and AUDIT steps currently point to files absent from the bundle, breaking progressive disclosure.
Add worked examples so the protocols are demonstrated rather than only named: a before/after revision pair, one filled Claim-Evidence Map row, and 2-3 evidence-strength-to-hedge calibrations.
Merge the four overlapping workflow representations (Entry, Scenes, Canonical workflow path, Actions table) into a single sequence to remove roughly 40 lines of repetition.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is directive rather than explanatory, but the same workflow is rendered four times — Entry steps, Scenes (PREPARE/ACQUIRE/ACT/VERIFY/FINALIZE), the Canonical workflow path (READ/PLAN/DRAFT/AUDIT/REVERSE-OUTLINE/POLISH), and the Actions table — which could be merged into one sequence. Not 2 because it never pads with concepts Claude already knows; not 4 because the redundant process representations exceed "minor instances" of over-explanation. | 3 / 5 |
Actionability | Concrete elements exist — quantified rules ("Em dashes ≤ 1 per paragraph; semicolons ≤ 2 per 1000 words"), banned words ("definitely, clearly, obviously"), explicit per-mode output formats, and a failure/recovery table — but the core techniques (sentence types, verb tiers, hedge calibration) are delegated to `resources/` files and no worked examples (before/after prose, a filled Claim-Evidence Map) appear inline. Not 4 because key executable detail is missing from the body itself; not 2 because the guidance present is specific and executable. | 3 / 5 |
Workflow Clarity | The multi-step process is explicitly sequenced with validation checkpoints: Scene 4 VERIFY ("apply the shared prose diagnostics and academic checklist"), an iterative feedback loop ("every fix loops back through the anti-AI checklist before emit"), a failure/recovery table with retry actions ("re-run audit"), and explicit exit criteria. Not 4 because validation and error-recovery loops are explicit throughout rather than merely present with minor gaps. | 5 / 5 |
Progressive Disclosure | Structure is good — a dedicated References section, load conditions ("load when drafting, revising, or reviewing an abstract"), one-level-deep pointers — but 5 of the 8 referenced paths (resources/anti-ai-checklist.md, sentence-structure-reference.md, academic-verb-tiers.md, hedging-guide.md, abstract-guide.md) do not exist in the bundle; only the three references/_shared/core/ files are present. Not 4 because dangling pointers to the majority of referenced content is more than a minor organization gap; not 2 because references are clearly signaled, not buried, and the body is well-sectioned. | 3 / 5 |
Total | 14 / 20 Passed |