Content
67%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.
A well-structured, contract-driven instruction skill with concrete routing rules, named reason codes, and clear terminal behavior. Its weaknesses are moderate duplication (the human-reason factor list appears twice, and routing rules are restated) and interpretive steps that point to external skills instead of giving worked examples.
Suggestions
Deduplicate the human-reason factor list: state it once in the evidence lens and have workflow step 3 reference it ('apply the contextual factors listed above') instead of repeating an overlapping list.
Tighten the connection-contract section by merging the repeated router-return rules in 'Terminal behavior' and 'Stop conditions' into a single routing subsection.
Add one short worked example (e.g., a flag, a plausible human cause, and what evidence would reduce uncertainty) to make the interpretive steps of the workflow concrete rather than deferring entirely to ../avoid-ai-writing/SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (no concept explanations) and is domain-specific throughout, but the human-reason factors are listed twice — in the evidence lens ('context mode, genre, second-language writing, technical register, editing software') and again in workflow step 3 ('genre, second-language writing, technical register, deadline pressure, editing tools, typography software') — and the connection-contract section has redundant routing restatements. This matches anchor 3 ('could be tightened') rather than 4, where trimming would be minor. | 3 / 5 |
Actionability | As an instruction-only skill the guidance is concrete: specific router reason codes ('fresh_signal_collection_needed'), named envelope fields ('consequential_authorship_claim'), named evidence types ('source history, drafts, revision logs, direct discussion'), and explicit prohibitions ('Do not rewrite detector scores, invent confidence values'). It is below anchor 5 because steps like 'Explain the strongest signals and plausible human reasons' defer the actual interpretive method to the referenced skill rather than giving specific examples. | 4 / 5 |
Workflow Clarity | The six-step workflow is clearly sequenced with decision checkpoints (step 4's router return when an audit is missing), explicit stop conditions, and terminal-behavior rules that prevent loops. It matches anchor 4 rather than 5 because there is no checkpoint validating the interpretation itself and several steps remain at the level of directives without error-recovery detail. | 4 / 5 |
Progressive Disclosure | No bundle files exist, and the body is well-organized into clearly labeled sections; its external file references (../avoid-ai-writing/SKILL.md, ../avoid-ai-writing-router/references/handoff-contract.md, skill-graph.json, agency-role-lenses.md) are one level deep and signaled with explicit paths. This matches anchor 4 — good structure with minor gaps — rather than 5, since the references are cross-skill pointers rather than a deliberate split of this skill's own content, and the 76-line body exceeds the under-50-line simple-skill case. | 4 / 5 |
Total | 15 / 20 Passed |