安全
Krait
A structured Claude Code skill package for Solidity security auditing with documented methodology, heuristics, and high precision claims.
ZealynxSecurity · 已有 AI 审核记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add ZealynxSecurity/krait技能说明
Krait
AI-assisted security verification for Solidity smart contracts. Not a scanner — a structured methodology with 101 heuristics, 26 analysis modules, and 8 kill gates, tested blind against 50 Code4rena contests at 100% precision. Runs inside Claude Code. Free.
At a Glance
| Current version | v8.2 (kill-gate parity, Impact Premise, rescan + per-contract recall phases) |
| Measured baseline | v8 — 50 contests. v8.1/v8.2 are not yet re-measured (why) |
| Detection angles | 16 per function (4 lenses × 4 mindsets) |
| Heuristics | 43 original + 58 extended (from open-source community) |
| Analysis modules | 15 deep-dive module files + 26 inline modules (A-X) |
| Audit-trail rules | R8 / R10 / R11 / R12 / R15 / R16 — exercised per finding (v8.1) |
| Domain primers | 7 (DEX, Lending, Staking, GameFi, Bridges, Proxies, Wallets) |
| Kill gates | 8 automatic + Impact Premise + 10 FP patterns (identical on both surfaces, parity-tested) |
| Shadow audits | 50 contests, 100% precision, 0 FPs/contest (v8 baseline) |
| Full methodology | METHODOLOGY.md — every technique, publicly documented |
Two Local Surfaces, One Web Platform
This repo ships two Claude Code surfaces. They answer different questions, and you can run either or both.
Audit pipeline (/krait) | Checklist plugin (/krait:scan) | |
|---|---|---|
| Question it answers | "What bugs are in this code?" | "Am I ready for an audit?" |
| Method | Multi-phase adversarial reasoning from first principles | 845 curated checks across 39 DeFi verticals |
| Output | Exploit traces with file:line, severity, suggested fix | Per-check verdicts, importable assessment JSON |
| Bar | Zero false positives — 8 kill gates try to disprove every finding | Coverage — every applicable check gets a verdict |
| Lives in | ` |