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.
The SKILL.md is a well-organized, prescriptive thin index with clear mode routing, sequenced phases, and validation gates — genuinely strong instructional design. Its critical defect is that the entire detail layer it indexes into (rules/ and templates/) is absent from the bundle, so most references dangle and the executable substance is unreachable, while the only other bundled reference is orphaned.
Suggestions
Ship the referenced detail layer: add the 11 missing files under rules/ (content-routing, placement-resolver, writing-style, claude-md, drift-detection, readme, docs-folder, maintenance, auto-update-loop) and templates/ (claude-md.md, readme.md), or inline the per-mode essentials into SKILL.md so the thin-index design actually resolves.
Inline the minimal executable specifics currently deferred to missing files — e.g., the git diff commands for change detection, the drift-check list, and the README audit rubric items — so update/audit runs are actionable from the body alone.
Reference references/research-sources.md from the body (it is currently an orphan) and state the CLAUDE.md ≤ 200-line threshold once instead of three times across hard rules, the Definition of Done, and anti-patterns.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is a lean thin-index — mode-routing tables, phase lists, and one-liner anti-patterns with a prescriptive-not-descriptive principle, assuming Claude's competence. Minor trimmable padding remains: the "CLAUDE.md ≤ 200 lines / Anthropic's own threshold" rationale is repeated in the init hard rules, the init checklist, and the anti-patterns list, and the AGENTS.md intro explains a spec Claude largely knows. Not 5 because of that duplication; not 3 because the body is otherwise efficient. | 4 / 5 |
Actionability | The body gives concrete decision tables (mode detection, disambiguation, argument parsing) but defers all executable specifics to files that do not exist in the bundle — "see rules/drift-detection.md §1 for git diff commands", "Run the README audit rubric in rules/readme.md §4", "Use templates/claude-md.md" — so no actual command, rubric, or template is reachable. Structurally it matches the 4 anchor's shape, but with the target files missing the guidance never resolves to executable steps, leaving it at the 3 anchor's incomplete-guidance level. | 3 / 5 |
Workflow Clarity | Each mode has a numbered phase sequence with explicit gates: an overwrite confirmation (AskUserQuestion: Overwrite/Merge/Skip-Abort) during init, a dry-run option for update, and per-mode Definition-of-Done checklists ("Treat any unchecked item as a defect") plus P0/P1/P2 priority tiers. Not 5 because several checkpoints delegate their actual procedure to missing rule files (e.g., drift §1 diff commands, readme §4 audit rubric); comfortably above the 3 anchor since checkpoints themselves are explicit. | 4 / 5 |
Progressive Disclosure | The design is the ideal thin-index pattern — one-level-deep, well-signaled references with per-mode loading tables — but scored against the actual bundle, 11 of the 12 referenced paths (all of rules/*.md and templates/*.md) do not exist; only references/archetypes.md ships, and references/research-sources.md is an orphan never cited in the body. Navigation is therefore non-functional as delivered, which drops it below the 3 anchor despite the good on-page structure. | 2 / 5 |
Total | 13 / 20 Passed |