Mantis

Google's open-source skills pack that lets AI coding agents autonomously find, reproduce, and patch vulnerabilities in a repository.

Google · Security and safety tooling · open-source

Published in Feed

Overview

Mantis is an open-source AI harness from Google, released on GitHub under the Apache 2.0 license, that installs as a modular set of skills (via `npx skills add google/mantis`) into any coding-agent CLI. It runs a staged pipeline: a threat-modeling step builds a living architecture and trust-boundary model of the repository, researcher agents scan for flaws, review/critic agents filter false positives, and reproduce agents generate sandboxed crash reproducers to ground each finding in evidence before a patch is proposed. The project positions agent-native security review as a composable skill set rather than a standalone scanner, running inside the developer's existing coding-agent workflow.

This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.

Key features

  • Modular skills pack installed into existing coding-agent CLIs via `npx skills add google/mantis`
  • /mantis_threat_model step builds a living threat model of trust boundaries and attacker profiles from the repo's architecture
  • Researcher, review/critic, and reproduce agent stages to find, filter, and evidence vulnerabilities
  • Sandboxed crash reproduction to ground findings before patching
  • Apache 2.0 licensed and open on GitHub

Use cases

  • Automated security review of a codebase inside an AI coding agent session
  • Building and maintaining a living threat model for a repository
  • Reducing false positives in agentic vulnerability scanning via a dedicated critic stage
  • Generating reproducible evidence for discovered vulnerabilities before patching

Pricing

Free and open-source (Apache 2.0); usage cost is whatever the underlying coding-agent CLI and model calls cost.

Pricing reflects the source record and may change. Confirm current terms with the provider.

Dates and sources

Released
Not listed
Record updated
3 Oct 2026
Record last checked
3 Oct 2026

Explore the Feed

Other entries filed under security and safety tooling; inclusion does not imply equivalent features.

  • DefenseClaw Edge Connector

    Cisco's open-source DefenseClaw gains an Edge Connector that allows or blocks AI agent tool calls directly on IoT and robot hardware.

  • Lakera Guard 2.0

    Lakera's real-time guardrail API for LLM apps, updated with multi-agent protection and a new benchmark for detecting indirect prompt injection in tool outputs.

  • Fleming-1

    Sierra's model that listens to a live phone call and scores whether the caller is an AI agent rather than a human.

Browse the Feed