Content
70%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 body delivers exceptionally concrete, well-sequenced editing guidance with built-in verification loops, but it is a monolith: nearly 800 lines in a single file with substantial meta-commentary, citation apologetics, and development-history notes that inflate the token cost. Moving the word tables and pattern catalog into reference files and trimming the digressions would raise both conciseness and progressive disclosure without losing actionability.
Suggestions
Split the Tier 1/2/3 word tables, pattern catalog, and tolerance matrix into one or more reference files (e.g. references/patterns.md, references/vocabulary.md), keeping SKILL.md as an overview with mode definitions and workflow steps.
Trim meta-commentary that does not change editing behavior: the detector-implementation history ('this was implemented as a detector and reverted'), issue references (#56, #39), roadmap notes, and the academic-citation paragraph in 'What this skill is and isn't'.
Move time-sensitive citations and statistics (Liang et al. 2023, arXiv:2506.07001, 'April 2026 run') into a dedicated references/notes file so they stop competing with the core instructions for context.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~780-line body is noticeably verbose: meta-sections like 'What this skill is and isn't' (three academic citations), 'Caveat worth keeping visible,' reverted-detector histories ('every regex tight enough... stopped matching the tell'), roadmap notes ('the first of four stylometric signals on the roadmap'), and issue references (#56, #39) pad the instruction set. This matches the 2 anchor (several padded sections) rather than 3, because the digressions recur throughout rather than appearing occasionally. | 2 / 5 |
Actionability | Fully concrete guidance: replace/with word tables, before→after rewrite examples ('Imagine a world where every deploy is instant' becomes 'Instant deploys would cut our release cycle from a day to minutes'), per-mode output-format templates, and a runnable validator command. For an instruction-only skill this is copy-paste-level actionable guidance covering the common cases. | 5 / 5 |
Workflow Clarity | Each mode has a numbered job sequence; edit mode ends with an explicit verification step ('re-read the file and confirm the flagged patterns are resolved'), rewrite mode builds in a second-pass audit feedback loop, and the optional mechanical check adds validate-and-fail-nonzero enforcement. Clear sequence with explicit validation and error recovery — matches the 5 anchor. | 5 / 5 |
Progressive Disclosure | No bundle files exist; the ~600 lines of tiered word tables, pattern catalog, and tolerance matrix are all inlined in SKILL.md where a references file would belong, matching the 3 anchor ('content that should be separate is inline'). Not 2: the document itself is well structured with clear, navigable section headers rather than a minimal-structure wall; not 4 because the bulk reference material is not split out at all. | 3 / 5 |
Total | 15 / 20 Passed |