GitHub - cloudflare/security-audit-skill: A coding-agent skill for multi-phase security audits with independently verified, machine-readable findings
This GitHub repo provides a coding-agent skill that runs automated security audits in six phases—recon, hunting, validation, reporting, structured output, and independent verification—to identify exploitable vulnerabilities. It uses parallel agents to generate and disprove findings, outputs structured JSON conforming to a schema, and independently verifies each claim against the source code. Each run reads prior findings to skip known issues and improve coverage.
- Cloudflare open-sourced the actual skill behind their own vulnerability discovery harness, structured as six phases: recon, hunting, validation, reporting, structured output, and independent verification.
- Findings are adversarially checked—the agent that finds a bug never validates it, and separate fresh agents re-verify each claim against the source code before reporting.
- A single audit pass only catches about half of total vulnerabilities, so the tool reads prior findings.json runs to skip known issues and improve coverage over repeated runs.
- Output is enforced as machine-readable JSON validated against a schema via a zero-dependency Node.js script, alongside human-readable REPORT.md and FINDINGS-DETAIL.md files.
Cloudflare’s security-audit-skill turns a coding agent into a multi-agent security auditor through six distinct phases: recon, hunting, validation, reporting, structured output, and independent verification. It started as the seed for Cloudflare’s own vulnerability discovery harness and lives in one repo. Each phase spins up parallel or adversarial sub-agents to map the codebase, probe for injection flaws, access-control bugs, business logic errors, crypto misuses, chained attacks and even wildcard scenarios. Findings get double-checked by separate agents before any report is written.
At the end of the run you get three outputs: a human-readable REPORT.md, a detailed FINDINGS-DETAIL.md for medium-and-higher issues, and findings.json validated against report-schema.json via a zero-dependency Node.js script (validate-findings.cjs). Then fresh agents verify each claim against the actual source. The skill skips previously logged issues by reading past findings.json files, so repeated runs explore new paths and improve coverage—tests show a single pass catches about half of total vulnerabilities.
You install it with “npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit” (add “--global” for user-level). Point your agent at the target repo and issue commands like “security audit this codebase” or “find security vulnerabilities in ./src.” Requirements are simple: a coding agent that handles tool calls and parallel sub-agents, plus Node.js for schema checks. Principles include only reporting exploitable bugs with concrete attack steps, strict adversarial validation (the finder never validates its own work), and measuring severity by real impact rather than checklist compliance.