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, largely actionable instruction skill with a clear sequenced workflow, explicit validation gates, and real one-level references. Its weaknesses are redundancy across Constraints/Creation Rules/Execution Boundaries/Anti-Patterns and the absence of a concrete collector invocation command.
Suggestions
Merge Constraints, Creation Rules, and the overlapping redaction rules in Execution Boundaries into a single section, and fold Anti-Patterns that duplicate Gotchas into one list, to cut roughly 20-30 lines.
Add the exact collector invocation command (or a pointer to the collector entrypoint) to Workflow step 1 so "Run or refresh the session collector into /tmp" is executable.
Reference references/task-profile.json from the body (or drop it from the bundle) so every bundled file is discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Guardrail content is spread across four overlapping sections with direct duplication: "Keep created artifacts single-purpose, compact, and executable" (Constraints) versus "Keep created skills single-purpose and executable" (Creation Rules), and redaction rules stated in both Execution Boundaries ("Redact secrets/sensitive data by default") and Constraints ("Redact secrets, tokens, API keys, credentials, PII..."). Anti-Patterns largely restates Gotchas. It is efficient in the sense of not explaining known concepts, but noticeably more than minor trimming is needed to consolidate, so it sits at the 3 anchor rather than 4. | 3 / 5 |
Actionability | Mostly executable guidance: concrete paths (~/.agents/session-collector/, /tmp), named bundle files (aggregate.json, skillify-candidates.json, skill-refactor-handoffs.json), a named existing-owners table with specific skills (pr-green-sweep, autofox triage, technical-writer), a copy-ready YAML output template, and exact validation commands (./bin/ask skills audit ... --level strict --json --robot). It is not a 5 because step 1 ("Run or refresh the session collector into /tmp using a sandbox-safe cache path") never gives an actual command to invoke the collector. | 4 / 5 |
Workflow Clarity | The 8-step Workflow is clearly sequenced with explicit checkpoints ("Validate the new artifact with the repo's narrowest available gate") and a dedicated Failure Mode section ("If validation fails, stop at the first failed gate, classify the failure"). It stops short of the 5 anchor because there is no fix-and-revalidate feedback loop, and the collector-run step is abstract rather than a concrete checkpoint. | 4 / 5 |
Progressive Disclosure | The body is well sectioned with clearly signaled one-level references ("See references/contract.yaml for the compact contract and references/evals.yaml for smoke cases"), both of which exist in the bundle. It is not a 5 because the ~170-line body inlines material (Examples, Existing Owners table) that could be referenced out, and the bundled references/task-profile.json is never referenced from the body. | 4 / 5 |
Total | 15 / 20 Passed |