Content
60%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, audit-first instruction skill with a clear four-step workflow, sensible guardrails, and a defined output format. Its weaknesses are repeated list content across sections and category-level guidance that never reaches concrete commands or example paths for reading the live surface.
Suggestions
Deduplicate the live-state classification list (keep it once, in Workflow step 2, and have Guardrails reference it) and state keep / merge / cut / fix-next once, in the Output Format template.
Add concrete evidence-gathering commands or example paths for each surface (e.g., .github/workflows/, crontab -l, gh run list, .claude/settings.json hooks) so step 1 is executable rather than a category list.
Turn the final Verification section into an explicit validate-and-retry loop (e.g., re-run the proof-path check for any item labeled 'recently verified' before finalizing the recommendation).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly lean telegraphic prose, but it is noticeably duplicated: the live-state list (configured / authenticated / recently verified / stale or broken / missing) appears in both Guardrails and Workflow step 2, and keep / merge / cut / fix-next appears three times (Workflow step 4, Output Format, Verification). It fits 'mostly efficient but could be tightened' better than the verbose anchor, since nothing explains concepts Claude already knows. | 3 / 5 |
Actionability | The guidance is concrete about what to inventory (hooks, GitHub Actions, MCP configs, connectors, wrappers), how to classify, and what the output must contain — but it contains no executable specifics: no commands, file paths, or example invocations (e.g., how to list hooks or enumerate scheduled workflows). Per the rubric's instruction-only note the absence of code is not itself penalized, but the guidance stops at category-level direction, matching 'some concrete guidance but incomplete' rather than 'mostly executable'. | 3 / 5 |
Workflow Clarity | A clearly numbered four-step sequence (inventory → classify → trace proof → recommend) with a read-only guardrail, an explicit ambiguity rule in step 3 ('say so directly instead of pretending the audit is complete'), and a Verification checklist closing the loop. It falls short of the top anchor because the verification steps are assertions to satisfy rather than explicit validate-and-retry feedback loops. | 4 / 5 |
Progressive Disclosure | The skill is a single self-contained file with well-labeled sections (When to Use, Guardrails, Workflow, Output Format, Pitfalls, Verification) and no nested or dangling references. It is above the 'good structure, minor organization gaps' threshold but not a clear top score: at ~135 lines with a Skill Stack section and duplicated category lists, some content could be split or trimmed, and the skill exceeds the under-50-line simple-skill exception. | 4 / 5 |
Total | 14 / 20 Passed |