Content
92%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.
An exemplary audit skill: maximal-density checklists with per-item grep patterns, copy-paste commands, a rigorous verdict/refutation workflow with validation checkpoints, and clean one-level-deep reference structure with all links resolving. The only slack is minor — an unlinked standards list at the end and a few rationale clauses restating common knowledge — which keeps conciseness at 4 rather than 5.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and checklist-driven with grep patterns and rationale instead of padding, and version-sensitive details (PSP deprecation, MPTCP per-netns behavior) are framed as deprecation notes or deferred to references. It falls short of the 5 anchor because of minor trimmable content: the closing References section lists standards (CIS, NSA/CISA, OWASP, MITRE ATT&CK) without links or use instructions, and a few clauses restate knowledge Claude already has ('K8s Secrets are base64-encoded, NOT encrypted'). | 4 / 5 |
Actionability | Fully executable throughout: per-item grep patterns ('grep -rln "privileged: *true" ...'), a copy-paste one-liners block (trivy, kube-bench, git ls-files pipeline), concrete validation commands ('node references/validate-findings.mjs findings.json'), jq redaction commands, and worked fingerprint examples. Not the 4 anchor because commands are complete and cover the common cases rather than having gaps. | 5 / 5 |
Workflow Clarity | The pipeline is clearly sequenced — Authorization → Scope the Audit (numbered) → checklists → 'Verify, then record' → output format — with explicit validation checkpoints: the refutation pass before confirming, findings.json validation, a needs_validation protocol with blocker+plan, and a 'Verify Fixes at Runtime' section with per-fix checks. Checklists cover the complex process; error-recovery feedback loops are present, so the destructive/batch cap does not apply (and the skill is read-only by design). | 5 / 5 |
Progressive Disclosure | The body keeps the working checklists inline and pushes deep material one level deep into references/ with purpose-signaled links ('the refutation pass in references/verification.md', 'the gate in references/severity-calibration.md'); every referenced path (runtime-exposure.md, findings-format.md, verification.md, severity-calibration.md, vendor-severity.md, enrich-advisories.mjs, validate-findings.mjs, to-sarif.mjs, findings.schema.json) exists in the bundle. Navigation is easy and nothing that belongs in a separate file is inlined. | 5 / 5 |
Total | 19 / 20 Passed |