Awesome AI Agents › Red-Teaming Platforms

deonmenezes/mantishack

⭐ 498 Python repository created 2026-04-05

Mantishack is an offensive application-security harness that rebrands and rewires the OpenAI Codex command line agent, written in Rust, into an autonomous vulnerability hunter. It runs a staged detect-then-validate pipeline over a target codebase: reconnaissance, detection, reachability analysis, attacker-simulation validation, chaining of findings into realistic attack paths, gated exploitation, fixing and reporting. The design principle stated by the author is that detection is a commodity while validation precision is the product, so a finding only reaches confirmed status if it survives attacker simulation, and anything rejected must cite a concrete roadblock explaining why. The capability layer is a set of Node-based MCP servers that wrap established scanners rather than reimplementing them: Semgrep, CodeQL database creation and analysis, OSV dependency scanning, TruffleHog secret scanning, Bandit and Trivy. A program-analysis substrate adds heuristic source-to-sink scanning, structural search through ast-grep, and path-condition satisfiability checking backed by the Z3 solver. A findings server holds finding state in code rather than in agent prose, enforcing the no-proof-no-confirm and cited-rejection rules mechanically. An HTTP audit server converts a captured exchange into a bounded, redacted evidence pack with a stable request hash. A canary server exposes decoy tools that raise an alert if the agent ever calls them, acting as a tripwire for prompt injection or hallucinated tool use. Each wrapped scanner reports itself unavailable when the underlying binary is missing instead of fabricating results, so the operator installs Semgrep, TruffleHog, Trivy, Z3, ast-grep, osv-scanner and the CodeQL CLI separately. The author describes it plainly as a rough working harness with gaps rather than polished software, intended for security researchers and bug bounty hunters willing to work with the sharp edges.

https://github.com/deonmenezes/mantishack

agent-harnessai-agentsautonomous-agentsbug-bountyclaude-codemantismcpoffensive-securitysecuritycodexstatic-analysisvulnerability-discovery