Content
73%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, actionable instruction-only skill: sequenced workflow with explicit checkpoints and failure recovery, a concrete output schema, executable validation, and real one-level-deep references. The main drag is redundancy — duplicated redaction rules and heavily overlapping Gotchas/Anti-Pattern sections — plus an orphaned bundle file the body never points to.
Suggestions
Merge the duplicate redaction guidance: 'Redact secrets, private URLs, credentials, customer data, and sensitive operational details' appears in both Execution Boundaries and Constraints — keep it once.
Collapse the overlapping Gotchas and Anti-Pattern sections into one list; items like 'treating a CodeRabbit summary or Slack message as current PR state' are stated nearly verbatim in both.
Either reference references/task-profile.json from the body (e.g., in Inputs or Validation) so the whole bundle is navigable, or remove it from the bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient and does not explain concepts Claude already knows, but there is real duplication that could be tightened: redaction appears twice ('Redact secrets, private URLs, credentials, customer data' in Execution Boundaries and 'Redact secrets, tokens, API keys, credentials, PII' in Constraints), and Gotchas and Anti-Pattern overlap heavily ('Treating every bot message as actionable without stale-thread or latest-head checks' vs 'Treating a CodeRabbit summary, Slack message, or old CI notification as the current PR state'). | 3 / 5 |
Actionability | Concrete, executable guidance throughout: a copy-paste YAML output template with named buckets (fix_now/monitor/informational/blocked) and owner lanes, an executable validation command ('./bin/ask skills audit ... --level strict --json --robot'), and three dialogue Examples. Minor gaps remain in the gathering steps, e.g., 'refresh the live state that matters' leaves the exact tool invocations to judgment rather than naming specific commands. | 4 / 5 |
Workflow Clarity | A clear 7-step sequence with explicit checkpoints ('Refresh live truth before placing any item in fix_now', 'Stop before implementing fixes unless the user asked for follow-through') and a genuine feedback loop in the Failure Mode section: 'If live truth cannot be refreshed, keep the item in blocked... Do not upgrade the item to fix_now from stale notification text.' The skill is read-mostly classification, so the destructive/batch validation cap does not apply. | 5 / 5 |
Progressive Disclosure | The body is a well-organized overview with one-level-deep, clearly signaled references that both exist on disk ('See references/contract.yaml for the compact machine-readable contract and references/evals.yaml for routing checks'). However, the bundle also contains references/task-profile.json, which is never referenced or navigated to from the body, leaving a minor organization gap against the actual bundle structure. | 4 / 5 |
Total | 16 / 20 Passed |