skills.rest
google
googleOFFICIAL
Explore all skills by this author →
Website𝕏@GoogleOSSGitHub

mantis-configure

Configures sandbox mechanisms, AI models, and credentials for Mantis security review pipelines.

865|94|Updated Jun 15, 2026
Search Tags:#security review#gce#gvisor#workflow.json#preflight validation#sandbox configuration#llm credentials
One-click install
npx skills add https://github.com/google/mantis --skill mantis-configure
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: mantis-configure
Source: https://github.com/google/mantis/tree/main/reference/skills/mantis-configure
Command: npx skills add https://github.com/google/mantis --skill mantis-configure

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve?

Setting up an isolated environment for AI-driven security review campaigns requires manually wiring sandbox isolation, LLM credentials, and model endpoints, and misconfiguration is often only discovered after a campaign fails mid-run.

Core Features & Use Cases

  • Sandbox Configuration: Selects between static-only, gVisor container, microsandbox microVM, and GCE ephemeral VM isolation levels based on host capabilities.
  • Model and Endpoint Management: Switches between Gemini, Claude on Vertex AI, MaaS models, and custom OpenAI-compatible endpoints with reasoning effort and timeout controls.
  • Fast Preflight Validation: Runs 1-2 second checks verifying LLM reachability and sandbox readiness before launching campaigns.
  • Use Case: Before running an unattended vulnerability discovery campaign on a new host, auto-detect available isolation mechanisms (KVM, runsc, gcloud), configure the strongest available sandbox, and verify credentials in seconds.

Quick Start

Ask the agent to auto-detect host capabilities and configure the Mantis pipeline with the best available sandbox, then run a preflight test to confirm everything is operational.

Frequently Asked Questions about mantis-configure

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I configure a sandbox for AI security review pipelines?▼

Run the configure script with --auto to detect host capabilities like /dev/kvm, docker/runsc, or gcloud, and it selects the strongest available sandbox. Alternatively, specify one explicitly with --sandbox static-only, gvisor, microsandbox, or gce.

How do I switch LLM models in a workflow.json configuration?▼

Use the --model flag with a provider-prefixed identifier such as gemini-3.7-flash, vertex_ai/claude-opus-5, or openai/{MODEL_ID}. For custom endpoints, add --api-base pointing to your OpenAI-compatible server like a local vLLM instance.

What sandbox isolation options exist for running untrusted AI-generated code?▼

Four tiers are supported: static-only (no execution), gVisor containers via runsc, microsandbox microVMs requiring /dev/kvm, and ephemeral GCE VMs accessed over IAP SSH tunnels. Higher tiers provide stronger kernel or hypervisor isolation.

Can I use a local vLLM or Ollama server with this configuration?▼

Yes, custom OpenAI-compatible endpoints are supported via the openai/{MODEL_ID} model prefix combined with --api-base pointing to your server URL. You can also set --reasoning-effort and --timeout for these endpoints.

How do I verify LLM credentials and sandbox readiness before a campaign?▼

Run the configure script with --test or --preflight to execute fast 1-2 second validation checks. It verifies LLM reachability and sandbox requirements, and --json outputs diagnostics in machine-readable form.

When should I use static-only mode instead of a sandboxed environment?▼

Use static-only when the host lacks docker, KVM, or cloud credentials, or when zero host risk is required. It disables reproducer execution and dynamic patching, limiting analysis to static source review.

Related Skills

Explore More Software Engineering Skills→
anaregdesignanaregdesign
1

azure-app-platform-delivery

Coordinate Azure hosting and production release delivery for React Router and Prisma web apps.

Official
Advanced
devsanthoshmkdevsanthoshmk

sandbox-sdk

Execute untrusted commands and LLM-generated code in isolated Cloudflare Workers sandboxes.

Community
Advanced
xalgordxalgord
954

implementing-aws-iam-permission-boundaries

Configure AWS IAM permission boundaries to delegate role creation while enforcing maximum privilege limits.

Community
Intermediate
latchagentlatchagent
2

enclave-seatbelt

Sandbox macOS agent shells with filesystem and network restrictions.

Official
Advanced