Content
63%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 disciplined, low-token overview with explicit validation gates and routing rules, appropriate for a spec-writing entry point. Its weaknesses are that the spec artifact itself is never shown concretely (no ID or frontmatter examples) and the References section does not match the actual bundle.
Suggestions
Add a minimal example of the deliverable: a short spec frontmatter block and one SA or VAC acceptance ID, so 'Define expected behavior, non-goals, interface shape, and acceptance IDs' becomes executable rather than aspirational.
Reconcile the References section with the actual bundle: the listed Plugins/harness-engineering/... paths are not present in this skill's directory, while the bundled references/context-preservation.md is never linked from the body despite the context-disposition policy being invoked.
Deduplicate the guidance repeated across sections: 'do not plan sequencing or implement code' appears in Core Contract, Constraints, and Gotchas, and he-deepen-spec routing appears in both Core Contract and Gotchas.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean and imperative with no explanations of concepts Claude already knows, but "do not plan sequencing or implement code" and the he-deepen-spec routing rule are each repeated across Core Contract, Constraints, and Gotchas. Efficient overall with minor duplication that could be trimmed. | 4 / 5 |
Actionability | Some concrete guidance exists (the exact lint command with path, explicit routing targets, stop conditions), but the core deliverable is left undefined: no example spec artifact, no SA/VAC ID format, no frontmatter shape. This is an instruction-only skill whose guidance is only partially actionable. | 3 / 5 |
Workflow Clarity | A three-step procedure is backed by an explicit Validation section with a runnable gate command and "Stop at the first failed gate", plus a Failure-mode stop-and-ask path. Not 5 because there is no fix-and-revalidate feedback loop and the procedure steps themselves remain high-level. | 4 / 5 |
Progressive Disclosure | The body is a well-sectioned overview with clearly labeled references, but the six listed Plugins/harness-engineering/... paths do not exist in this skill's bundle, and the one bundled file (references/context-preservation.md) is never linked from the body even though the "context-disposition policy" it implements is invoked. Navigation dead-ends, so structure is only partially effective. | 3 / 5 |
Total | 14 / 20 Passed |