Content
78%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-built instruction-only audit skill: eleven pattern sections with specific, named controls and cross-references, a clear repeatable workflow with explicit dispositions and an output template, and exemplary progressive disclosure for its one deep-dive topic. The main improvements are trimming the well-known breach narratives, adding executable check commands beyond Pattern 11, and making the 'Clean' disposition verifiable before recording.
Suggestions
Trim the 'What happened' narratives for the most widely documented breaches (Capital One, SolarWinds, Equifax) to one or two sentences each, keeping only the pattern-relevant specifics.
Add a small set of executable grep or CLI check examples for one or two more patterns (as the third-party-scripts reference does) instead of delegating all execution to other audit skills.
Add an explicit validation step to the workflow instructing Claude to substantiate each 'Clean' disposition with evidence (command output, config excerpt) before recording it in the coverage table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is efficient: each pattern is a tight What happened / Audit questions / Check triple with no padding or explanation of basic concepts, and the intro and boundaries are brief. Not 5 because several 'What happened' paragraphs (Capital One, SolarWinds, Equifax) re-tread very widely known incidents in 3-4 sentences that could be trimmed to the pattern-relevant detail. Not 3 because these summaries directly feed the audit questions and the rest of the body earns its tokens. | 4 / 5 |
Actionability | Checks are concrete and specific — 'IMDSv2 enforced on every EC2 launch template (MetadataOptions.HttpTokens: required),' named metadata IPs, 'CVE-2023-34362,' 'Number matching enabled on push MFA,' SRI pinning rules, plus a full markdown output template — and Pattern 11 has a companion reference with copy-paste rg/curl commands. Not 5 because for the other ten patterns execution depth is delegated to other skills ('see iam-audit,' 'see secrets-audit') with no in-bundle commands, leaving minor gaps in copy-paste-ready guidance. | 4 / 5 |
Workflow Clarity | The 'How to use this skill' section gives a clear 4-step sequence (Read, Ask, Map, Decide) with an explicit disposition decision per pattern ('we've confirmed this can't happen' / 'we have a gap' / 'we accept-risk'), a defined output format, and boundary guidance to pivot to incident-triage on active issues. Not 5 because there is no validate-then-record checkpoint — nothing tells Claude how to substantiate a 'Clean' disposition before writing it into the coverage table, so checkpoints are present but implicit in places. | 4 / 5 |
Progressive Disclosure | Clear sectioned overview in SKILL.md, and the single bundle reference is well-signaled ('TP1-TP6, greps and sources in [references/third-party-scripts.md]'), exists as a real file, and is exactly one level deep with its own organized structure (incident summary, checks table TP1-TP6, executable greps, sources). Content is appropriately split: only the pattern needing deep detail was extracted; the inline patterns are compact enough to stay in the body. | 5 / 5 |
Total | 17 / 20 Passed |