What problem does it solve? Manual security reviews of large codebases miss entire attack surfaces because reviewers cannot systematically cover every trust boundary, and unvalidated findings waste triage time on false positives. ## Core Features & Use Cases - Coverage-led hunting waves: Decomposes a codebase into deterministic coverage units tracked in a ledger, assigns them to focused hunter agents, and runs coverage-critic passes to detect gaps before declaring completion. - Domain-specific attack classes: Ships companion references for AI/LLM agents, web protocols, client-side browser code, cloud deployment, memory safety, data isolation, desktop/mobile IPC, supply chain, and resource exhaustion. - Rigorous validation gates: Distinguishes confirmed findings (full source trace plus bounded local reproduction in a sandbox) from needs_validation candidates that require owner-observed deployment facts. - Use Case: Point the audit at a multi-tenant SaaS repository; it maps trust boundaries, hunts injection, access-control, and tenant-isolation defects per subsystem, and returns a prioritized report where every finding carries a source trace and reproduction evidence. ## Quick Start Run a deep security audit of this repository and report only source-validated vulnerabilities with their traces and remediation steps.
