Content
68%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 lean, well-sequenced enumeration runbook with genuinely useful canonical content (the 21 Rhino Security privesc primitives, concrete aws CLI commands, worked escalation chains). It is held back by undefined tool invocations (iam_policy_audit, kg_add_node) presented without context, unsubstituted placeholders, and a complete absence of validation/error-handling checkpoints for the multi-step workflow.
Suggestions
Explain or ground the iam_policy_audit() and kg_add_node()/kg_add_edge() calls — either name the tool/script they invoke (and bundle it in scripts/) or replace them with concrete, executable equivalents; currently they read as unexplained pseudocode.
Add validation checkpoints to the workflow: handle AccessDenied on get-policy-version gracefully, confirm policy JSON downloaded before auditing, and paginate/iterate explicitly over all attached policies rather than leaving ME/ARN/VID as unsubstituted placeholders.
Consider moving the 21-primitive canonical list to a references/ file (e.g., references/privesc-primitives.md) and keeping a short 'how to check' summary inline, to improve progressive disclosure and keep SKILL.md as an overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is lean throughout: no padding, no explanation of AWS/IAM basics Claude already knows, and every section (identity check, policy retrieval, the 21-primitive list, graph promotion, common chains) carries non-obvious information that earns its place — matching anchor 5's "every token earns its place". | 5 / 5 |
Actionability | The aws CLI commands are concrete but rely on unsubstituted placeholders (ME, ARN, VID), and the guidance mixes in non-executable calls with no explanation: "iam_policy_audit(open("/tmp/policy.json").read())" is an undefined function dropped into a bash flow, and "kg_add_node(... src=<vuln> ...)" is environment-specific pseudocode — matching anchor 3's "some concrete guidance but incomplete; pseudocode instead of executable code" rather than anchor 4's mostly-executable bar. | 3 / 5 |
Workflow Clarity | The five sections form a coherent sequence (identify → gather policies → check primitives → promote to graph → note chains), but validation checkpoints are entirely absent: there is no handling of the extremely common AccessDenied errors on iam get-policy-version, no check that policy retrieval succeeded before auditing, and no verification step after graph promotion — matching anchor 3's "sequence present but checkpoints missing". | 3 / 5 |
Progressive Disclosure | Sections are well-organized with clear headers and the skill is self-contained (no bundle files exist to reference), which per the rubric can approach 5; however, the 21-primitive enumeration (item 11 of the guidelines rewards reference material placed appropriately) is exactly the kind of stable reference data that could live in a references/ file, so organization is good but not optimal — anchor 4. | 4 / 5 |
Total | 15 / 20 Passed |