Overview

This repository provides a sophisticated coding-agent skill designed to automate and enhance multi-phase security audits. It transforms a standard coding agent into a dedicated security auditor by orchestrating a series of isolated agents through a structured, six-phase process. This skill is the foundational technology behind Cloudflare's vulnerability discovery harness, offering a robust starting point for developing advanced security auditing systems.

Key Features

• Automated Six-Phase Audit Process: Orchestrates a comprehensive audit covering reconnaissance, coverage-led hunting, candidate validation, structured output, independent record verification, and target-neutral reporting.
• Isolated Agent Execution: Leverages isolated agents for each phase to ensure security and prevent cross-contamination, with adversarial validation as a core principle.
• Machine-Readable & Verifiable Findings: Generates independently verified, machine-readable findings in findings.json, validated against report-schema.json.
• Coverage-Led Hunting: Employs a unique methodology where hunters are assigned from ledger units, and coverage critics actively identify gaps.
• Additive Runs: Supports multiple runs against the same codebase, intelligently targeting gaps, revalidating changes, and preserving existing evidence without treating stale work as covered.
• Comprehensive Documentation: Extensive Markdown files detail each phase, attack classes (including memory safety, AI/LLM, web protocols, client-side, supply chain, cloud, RPC, resource exhaustion, data isolation, and desktop/mobile), and validation logic.
• Zero-Dependency Validators: Includes validate-findings.cjs and validate-coverage-ledger.cjs for robust, dependency-free validation of audit artifacts.
• Flexible Usage: Integrates seamlessly with coding agents via simple natural language prompts, supporting both full audit and guidance modes.

Typical Use Cases

• Automated Vulnerability Discovery: Systematically identify security weaknesses across diverse codebases, from native applications to cloud deployments and LLM-powered systems.
• Security Audit Augmentation: Empower security teams by automating repetitive tasks and providing a structured framework for deeper analysis.
• Continuous Security Monitoring: Integrate into CI/CD pipelines for ongoing security assessments of code changes.
• Developing Custom Security Harnesses: Serve as a powerful, single-repo starting point for building more complex, fleet-wide vulnerability discovery systems.
• Research & Development: Explore the capabilities of AI agents in security auditing and contribute to the advancement of AI-driven security tooling.