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 is a well-structured orchestration contract with an exemplary bounded verify-repair workflow, but it is not self-sufficient: the actual editing rules, output format, and lens definitions all live in sibling-skill files that are absent from the bundle, so the skill cannot execute its core function as provided. Redundant restatements of pass-counting rules across three sections also leave some tightening room.
Suggestions
Inline the essential editing rules or at least a minimal example of the rewrite-mode Output format (including the four Verification items) instead of delegating entirely to `../avoid-ai-writing/SKILL.md`, which is not present in the bundle — the skill currently cannot perform a rewrite standalone.
Ship the cross-skill dependencies as bundle files (e.g., under `references/`) or correct the sibling paths, since `../avoid-ai-writing/SKILL.md` and `../avoid-ai-writing-router/references/*` do not resolve from this skill's directory.
Consolidate the editing-pass counting rules, which are currently stated in Workflow step 9, the Repair path, and Stop conditions, into a single section to remove redundancy.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Dense, contract-style procedural prose that assumes Claude's competence ("Do not replace one AI cliché with another") and explains nothing the model already knows. Not a 5 because the editing-pass counting rules are stated three times (Workflow step 9, Repair path, Stop conditions) and protection rules repeat across sections. | 4 / 5 |
Actionability | Concrete on process — named handoff operations (ROUTE, FEED, REPAIR, VERIFY), canonical voice profiles ("casual, professional, technical, warm, or blunt"), and bounded repair rules ("Change only the spans implicated by the blocking preservation errors"). Not a 4 because the core editing contract, the rewrite-mode Output format ("including its four Verification items"), and the representation lens are all delegated to files that do not exist in the bundle, leaving key operational details missing as written. | 3 / 5 |
Workflow Clarity | A clear 10-step workflow with explicit validation (step 10 sends before/after content to preservation-verifier) and a bounded feedback loop ("If `pass.index` has reached `pass.max`, do not repair"; "Return to `preservation-verifier` once... Do not cycle again"). This matches the anchor-5 pattern of explicit validation checkpoints plus error recovery. | 5 / 5 |
Progressive Disclosure | The body is well-sectioned (Connection contract, Conditional guard, Workflow, Repair path, Voice handling, Stop conditions, Output) with clearly signaled one-level references. Not a 4 because every reference points outside the bundle to files that do not exist here (`../avoid-ai-writing/SKILL.md`, `../avoid-ai-writing-router/references/handoff-contract.md`, `skill-graph.json`, `agency-role-lenses.md`), so the substantive rulebook is entirely unverifiable and navigation fails. | 3 / 5 |
Total | 15 / 20 Passed |