Content
71%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 strong, dense audit playbook: seven bug classes with concrete vulnerable/fix pairs, a greppable audit sequence with checklists, and near-executable Foundry PoCs plus a remediation reference contract. The main gaps are the fully monolithic single-file structure with no progressive disclosure, placeholder ellipses in several code snippets, and no explicit PoC-confirmation step linking findings to the templates.
Suggestions
Move the Foundry PoC templates, the known-exemplars list, and the full remediation contract into references/ files (e.g., references/poc-templates.md, references/exemplars.md, references/remediation.md), leaving SKILL.md as the bug-class overview plus the audit sequence with one-level-deep pointers.
Replace ellipsis placeholders in the bug-class snippets (ecrecover(h, ...), new Target(...)) with complete minimal snippets, or define the sign()/vm helpers once so the PoC templates are copy-paste runnable.
Add an explicit step 5 to the audit sequence: 'For each suspected finding, write a PoC from the templates below and only report the bug if the replay/malleable transaction succeeds' — turning the templates into a validation checkpoint rather than an appendix.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and efficient: each bug class is a compact vulnerable/safe Solidity pair, and it never explains background concepts Claude already knows (what EIP-712 is, how ecrecover works at length). It is not a 5 because of minor trimmable prose (e.g., 'This is mostly resolved by OZ ERC20Permit, but custom permit impls are often wrong', 'Each has well-known bugs') and the somewhat listy exemplar section, matching anchor 4's 'minor instances of over-explanation'. | 4 / 5 |
Actionability | Concrete grep commands, per-call-site checklists, three Foundry PoC templates, and a full executable OZ remediation contract — mostly executable guidance. It stops short of anchor 5's copy-paste-ready bar because several snippets contain ellipses and undefined helpers (ecrecover(h, ...), new Target(...), sign(privKey, hash), Target targetB) that the reader must fill in, matching anchor 4's 'concrete code with minor gaps'. | 4 / 5 |
Workflow Clarity | The Audit steps section gives a clear 1→4 sequence (find ecrecover call sites → verify payload contents → verify signature validation → grep for dangerous patterns) with verification checklists at each step. It is not a 5 because there is no explicit checkpoint that ties suspected findings to the PoC templates (no 'confirm with a PoC, and only report when it reverts/fails' step), and no feedback loop for the destructive implication of a confirmed finding — anchor 4's 'clear sequence with most checkpoints present; minor validation gaps'. | 4 / 5 |
Progressive Disclosure | The skill is a single ~256-line file with no references/, scripts/, or assets/ bundle at all; everything — PoC templates, exemplar incident list, full remediation contract — is inlined. Headers are clear (better than anchor 2's structure-less wall), but content that clearly belongs in separate reference files is inlined with no external navigation, which is exactly anchor 3's 'some structure... content that should be separate is inline' — and it cannot be a 4-5 since there are no well-signaled one-level-deep references at all. | 3 / 5 |
Total | 15 / 20 Passed |