Argus - ADK Threat-Modeling Agent

Argus turns PRDs, design docs, feature docs, RFCs, ADRs, and architecture notes into developer-facing threat model reports. It uses a Google ADK multi-agent workflow for security reasoning, exact-reference skills for grounding, and deterministic tools for schema validation, model element binding, OWASP severity calculation, and final Markdown rendering.

Install

The package is named argus-agent; the command is argus.

uv tool install argus-agent

Or:

pip install argus-agent

For local development from this repository:

uv sync
uv tool install --reinstall .

Quick Start

Argus accepts one readable UTF-8 text document. It does not ingest directories, GitHub URLs, OpenAPI specs, Terraform, or binary files in this version.

Bash:

export GEMINI_API_KEY="your-api-key"
argus path/to/design.md

PowerShell:

$env:GEMINI_API_KEY = "your-api-key"
argus path/to/design.md

cmd.exe:

set GEMINI_API_KEY=your-api-key
argus path\to\design.md

If --out is omitted, Argus writes the report to <input>.threatmodel.md.

argus path/to/design.md --out path/to/design.threatmodel.md

Configuration

Config precedence is:

CLI flags > exported environment > .env > built-in defaults

GEMINI_API_KEY is required and is intentionally environment-only. Do not pass API keys as CLI arguments.

Cheap smoke test:

argus path/to/design.md --model gemini-3.1-flash-lite

Use separate models for deeper and lighter stages:

argus path/to/design.md --pro-model gemini-2.5-pro --flash-model gemini-2.5-flash

Use a local .env file:

GEMINI_API_KEY=your-api-key
ARGUS_PRO_MODEL=gemini-2.5-pro
ARGUS_FLASH_MODEL=gemini-2.5-flash
ARGUS_TEMPERATURE=0.1

Use 1Password CLI:

op run --env-file=.env -- argus path/to/design.md --out path/to/design.threatmodel.md

Progress Output

Interactive terminals show a progress bar with the current stage and elapsed time. Non-TTY execution prints concise stage lines. Disable progress output with:

argus path/to/design.md --no-progress

Workflow stages:

input document
  -> ingestion_agent
  -> architecture_zone_agent
  -> entry_point_agent
  -> scenario_enumeration_agent
  -> schema_validation_agent
  -> element_binding_agent
  -> false_positive_validation_agent
  -> control_challenge_agent
  -> risk_rating_agent
  -> report_agent
  -> Markdown report

Output

The report includes:

• Architecture summary and data-flow view.
• Confirmed threats ranked by severity.
• STRIDE category, affected element, attack path, and mitigating controls.
• Ruled-out findings when the false-positive gate rejects a candidate.
• Assumptions, open questions, coverage notes, and limitations.

Security Design

• Agents perform document understanding, zone mapping, entry-point discovery, scenario enumeration, false-positive validation, control challenge, and risk-factor assignment.
• schema_validation_tool enforces structured stage outputs.
• element_validation_tool filters threats that do not reference real model elements.
• risk_rating_tool converts agent-assigned OWASP factor scores into deterministic severity.
• report_render_tool renders the final Markdown report; the report agent does not freehand it.
• Exact-reference OWASP, ATLAS, AIUC-1, and risk-rating skill corpora are packaged under argus.skill_corpus.

Troubleshooting

Missing API key:

GEMINI_API_KEY is required. Set it in the environment or a .env file.

Fix it by exporting GEMINI_API_KEY, creating .env, or using op run --env-file=.env.

Wrong command shape:

argus path/to/design.md

Do not use argus run; the installed CLI is intentionally flat.

Model capacity or rate-limit errors:

argus path/to/design.md --model gemini-3.1-flash-lite

Unreadable input:

Input must be a readable UTF-8 text document.

Convert the source document to text or Markdown before running Argus.

Local Development

uv sync
uv run pytest
uv run ruff check .

The live Gemini test is opt-in so normal test runs remain deterministic:

ARGUS_RUN_LIVE=1 uv run pytest -m live

Project Layout

argus/
  agents/          - ADK workflow, stage schemas, runner, and tools
  models.py        - SystemModel, Actor, Component, DataFlow, Control
  threats.py       - ProposedThreats, VerifiedThreats, element binding
  rating.py        - RatedThreat, ThreatStatus, rate_threats
  risk.py          - OWASP Risk Rating factor-to-severity matrix
  context.py       - SystemModel to structured agent context
  security/        - untrusted input guard
  skill_corpus/    - packaged exact-reference skills
  skills.py        - skill loader
  skills_router.py - just-in-time skill selection from SystemModel tags
  report.py        - Markdown report renderer
  config.py        - runtime config resolver
  cli.py           - Typer CLI
tests/             - unit, integration, and opt-in live tests

Contributing and Security

See CONTRIBUTING.md for development setup and PR guidance.

If you discover a vulnerability, follow SECURITY.md. Do not open a public issue for security reports.

Releasing

uv version --bump patch
uv version 0.1.1

License

This project is licensed under the MIT License. See LICENSE.