Index of every page on this site: llms.txt
All skills
google avatar

/mantis-launch

@e2205c7
by googlegoogle/mantis2.8k stars
241
Run it yourself

Launches automated vulnerability review campaigns on target files or repositories. Use to initiate Mantis vulnerability review pipelines with automated preflight checks, environment auto-configuration, and runtime overrides (sandboxes, models, endpoints). Don't use for configuring settings without scanning or for manual single-stage reviews.

Use this Skill: https://skilld.dev/gh/google/mantis/mantis-launch

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈90 tokens for metadata: the name and description. ≈1.7k when used: this file.

Description uses 4.5% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈359 characters, or 4.5%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Campaign Launcher (/mantis-launch)

System Goal

Autonomous Security Review Pipeline Launcher. Initiates end-to-end vulnerability discovery, independent verification, exploit viability analysis, crash reproduction, automated patch generation, and risk calibration campaigns across a target file or entire codebase repository.

Automatically detects unconfigured environment placeholders (such as YOUR_PROJECT_ID in GCE sandboxes), auto-resolves active credentials and virtualization capabilities, executes fast preflight sanity checks, and applies runtime overrides before running the pipeline.

Command Definition

  • Command: /mantis-launch
  • Description: Launches automated multi-agent vulnerability discovery and validation campaigns.
  • Execution Commands:
    python3 "${MANTIS_HOME:-/path/to/mantis}/reference/scripts/launch.py" <target_file_or_dir> [flags...]
    "${MANTIS_HOME:-/path/to/mantis}/reference/run.sh" <target_file_or_dir> [flags...]

Path Anchoring Requirement (CRITICAL): The launch scripts reside within the Mantis installation directory at reference/scripts/launch.py and reference/run.sh. You MUST invoke these scripts via an absolute path or via $MANTIS_HOME. NEVER execute ./reference/run.sh or python3 reference/scripts/launch.py using a relative path inside audited target repositories.

  • CLI Options:
    • target (positional): Path to a single source file (e.g. src/auth.py) or a root repository directory (e.g. . or /path/to/repo).
    • --sandbox / -s: Override sandbox mechanism (static-only, gvisor, microsandbox, gce).
    • --model / -m: Override AI model (e.g. gemini-3.8-flash, vertex_ai/claude-opus-5, vertex_ai/zai_org/glm-5.2-maas, openai/{MODEL_ID}).
    • --api-base: Custom endpoint URL for OpenAI-compatible LLM deployments (e.g. http://localhost:8000/v1).
    • --reasoning-effort: Reasoning effort level (low, medium, high).
    • --timeout: LLM request timeout in seconds.
    • --db / -d: Custom path to SQLite knowledge database (default: knowledge.db).
    • --depth: Continuous audit depth from 1 (fast shallow triage) to 100 (exhaustive full-stack + firmware/RTL coverage with multi-pass deep reasoning). Defaults to 30 when omitted. Explicit flags (--scan-mode, --passes, --reasoning-effort, --token-budget, --max-time) always take precedence, and existing budget guardrails remain enforced.
    • --scan-mode: Scan scope strategy (auto, whole, file-by-file, uncovered, cross-functional, changed). auto (default) uses --depth planning (whole up to the depth whole-file limit, else risk-ranked batches rotating onto uncovered/changed files on repeat runs). Use changed (without --resume) on the same knowledge.db to scan newly added/edited files and verify whether prior findings touching them are still_present or fixed.
    • --resume [RUN_ID]: Resume an interrupted or partially failed run (latest run for target when RUN_ID is omitted), inheriting its recorded scan mode. Cannot be combined with a different --scan-mode.
    • --workflow / -w: Path to custom workflow.json layout definition.
    • --preflight-only / --test / --preflight: Run preflight checks and exit without starting the campaign.
    • --probe / --probe-llm: Actively probe LLM reachability and provider credentials during preflight with a minimal test prompt (test, max 256 tokens).
    • --interactive: Launch interactive configuration wizard before execution.
    • --dry-run: Display launch plan and indexed files without calling AI models.
    • --no-auto-configure: Disable automatic detection and resolution of default placeholders.

Automated Auto-Healing & Preflight

Before starting a security campaign, mantis-launch:

  1. Placeholder Auto-Detection: Inspects workflow.json for unconfigured defaults (e.g. project: "YOUR_PROJECT_ID").
  2. Capability Auto-Healing: If unconfigured, automatically detects host capabilities (GCP project from gcloud, /dev/kvm for microVMs, or runsc for gVisor) and auto-updates workflow.json or falls back safely to static-only.
  3. Preflight Sanity Check: Runs a ~1s preflight check verifying that LLM credentials are valid and the selected sandbox environment is operational.

Common CLI Workflows

1. Launch Standard Review on Target File or Repository

# Scan a specific file
"$MANTIS_HOME/reference/run.sh" src/server/auth.py

# Scan an entire repository (defaults to --depth 30)
"$MANTIS_HOME/reference/run.sh" .

# Dial scan depth down for quick triage or up for exhaustive multi-pass coverage
"$MANTIS_HOME/reference/run.sh" . --depth 10
"$MANTIS_HOME/reference/run.sh" . --depth 100

2. Launch with Static Analysis Only (Zero Sandbox Requirements)

"$MANTIS_HOME/reference/run.sh" . --sandbox static-only

3. Launch with Specific Model (e.g. Claude or Custom OpenAI Server)

# Vertex AI Claude
"$MANTIS_HOME/reference/run.sh" . --model vertex_ai/claude-opus-5

# Local vLLM / Ollama server
"$MANTIS_HOME/reference/run.sh" . --model openai/custom-model --api-base http://localhost:8000/v1

4. Verify Preflight Readiness Without Scanning

python3 "$MANTIS_HOME/reference/scripts/launch.py" . --preflight-only

5. Incremental Scan & Fix Verification

# Resume failed/unstarted campaigns of the baseline run (inherits recorded scan mode)
"$MANTIS_HOME/reference/run.sh" . --resume

# After fixing code, start a new change-scoped run (do NOT pass --resume)
"$MANTIS_HOME/reference/run.sh" . --scan-mode changed

6. Inspect Results After Launch

All findings, regression verification checks (regression_checks), exploit reproduction logs, verified patches, and risk calibration scores are recorded in knowledge.db. Query guidance using mantis-advise:

python3 "$MANTIS_HOME/reference/scripts/advise.py" --file src/server/auth.py

Input/Output Contract

  • Reads:
    • Target source code files (under target path or repository).
    • workflow.json (declarative graph layout and config).
  • Writes:
    • knowledge.db (findings, regression_checks, campaign_spend, coverage_files, campaign_artifacts, risk_scores, learnings tables).
    • sessions.db (ADK session state trajectories), beside knowledge.db unless sessions_db_path / MANTIS_SESSIONS_DB says otherwise. It grows with every run and is pruned by nobody; each run ends by stating its size. Only a paused or interrupted run reads its sessions again (--resume).
    • Formatted terminal report and execution logs.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 warning22d3 checks · Risk SAFE
  • Gen Agent Trust Hub22d

    This skill provides a launcher for automated vulnerability review campaigns. It includes features for environment auto-configuration and preflight checks, which are standard considerations for security auditing tools. While it executes local commands and interacts with external model endpoints, these activities are aligned with the skill's primary purpose.

  • Socket22d

    No alerts

  • Snyk22d

    Risk: MEDIUM · 1 issue

Signed by skilld at e2205c7. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated 5 days ago

README badge

README badge for google/mantis/mantis-launch