Content
75%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, highly actionable audit playbook with concrete grep commands, executable Solidity patterns, a severity table, and real-world exemplars. The main gaps are the pseudocode flash-loan PoC and the lack of an explicit error-recovery loop in the audit workflow.
Suggestions
Make the flash-loan PoC executable (or explicitly justify the pseudocode, e.g. 'adapt pool/token addresses to the target') so the Foundry section is copy-paste ready like the rest of the code.
Add a short feedback step to the audit workflow, e.g. 'If the PoC shows no profit, re-check liquidity depth and slippage before downgrading severity.'
Move the known-exemplars list and severity-calibration table into a reference file (e.g. references/exemplars.md) to keep SKILL.md a tighter overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and lean — grep one-liners, compact vulnerable/good Solidity pairs, a severity table, and one-line exemplars — with no re-explanation of concepts Claude already knows. Minor trimming is possible (the comment-heavy pseudocode PoC and dated exemplar list), keeping it just below the every-token-earns-its-place anchor 5. | 4 / 5 |
Actionability | Copy-paste-ready grep commands, executable Solidity for the vulnerable/secure latestRoundData patterns and remediation, and a concrete vm.mockCall staleness test. The flash-loan PoC is explicitly pseudocode ('// Pseudo:') with comment-steps rather than executable code — the main gap preventing a 5. | 4 / 5 |
Workflow Clarity | The 'Audit steps' section gives a clear 1→4 sequence, and step 2 functions as a per-vector checklist (spot price? TWAP window? updatedAt/answeredInRound?) with the stale-price test providing explicit verification ('If we reach here w/o revert → bug'). Missing an explicit error-recovery loop (what to re-check when a PoC disproves the hypothesis) keeps it below 5. | 4 / 5 |
Progressive Disclosure | A single-file skill (~145 lines) with no bundle files and well-organized, clearly headed sections. It exceeds the 50-line simple-skill threshold, and the severity-calibration table and known-exemplars list are natural candidates for a separate reference file — matching the good-structure-with-minor-organization-gaps anchor 4. | 4 / 5 |
Total | 16 / 20 Passed |