promptheus-ai 0.1.0

Proactive Red-team Operator for Model PenTesting & Heuristic Exploit Utility System

PROMPTHEUS

Proactive Red-team Operator for Model PenTesting & Heuristic Exploit Utility System

Steals fire from the gods. LLM red-team security auditing.

---

AI agents are being given more power every day — calling tools, reading files, sending emails, executing code. Traditional security testing wasn't designed to find LLM-specific vulnerabilities: prompt injection, goal hijacking, tool abuse, memory poisoning.

PROMPTHEUS fills that gap.

---

Install

Option A — pip (requires Python 3.10+)

pip install promptheus-ai
promptheus init   # first-run setup wizard

To upgrade later:

pip install --upgrade promptheus-ai

Option B — Binary download (no Python required)

Download the latest release for your platform from the Releases page:

Platform	File
macOS	promptheus-macos
Windows	promptheus-windows.exe
Linux	promptheus-linux

macOS/Linux: Make executable and run:

chmod +x promptheus-macos
./promptheus-macos init

Windows: Run from PowerShell:

.\promptheus-windows.exe init

---

Quick Start

# First-time setup (choose your AI provider + enter API key)
promptheus init

# Scan an AI API endpoint
promptheus scan -u https://your-ai-app.com/chat

# Audit a codebase with AI agents
promptheus scan --mode agent --target-path /path/to/repo

# Review security of recent commits
promptheus pr-review --path /path/to/repo --last 5

No API key? Run without one — Judge defaults to Mock mode (always returns Safe), useful for testing the pipeline.

---

What it does

Payload Attack Engine

50 security payloads across 6 attack vectors:

Vector	Examples
Prompt injection	Direct override, Base64, YAML, sandwich, translation trap
Tool abuse	Shell exec, file read/write, SSRF, API key generation, privilege escalation
Multi-turn attacks	Conditional trigger, deferred execution, implicit consent
Memory attacks	Cross-session exfiltration and poisoning
Identity & auth	System prompt extraction, developer prompt leak, approval bypass
Infra attacks	Denial of Wallet, unsafe code execution, lazy error handling

Each payload includes a judge_expectation rubric — the LLM Judge classifies results as Critical, High, Medium, or Safe.

Agent Scan — Full Codebase Security Audit

6 AI agents analyze your codebase:

Agent	Output
Architecture Assessment	SECURITY.md
Threat Modeling (OWASP ASI 2026)	THREAT_MODEL.json
Code Review	VULNERABILITIES.json
Report Generator	scan_results.json
DAST Validation (optional)	DAST_VALIDATION.json
Fix Remediation (optional)	Inline suggestions

PR / Commit Review

Risk-based security triage on git diffs — integrated into your development workflow.

promptheus pr-review --path . --last 1
promptheus pr-review --path . --range main..feature/new-agent --severity high

---

All commands

promptheus init                          # First-run setup wizard
promptheus config show                   # View current config (API key masked)

# Scan
promptheus scan -u <url>                 # Attack an AI API endpoint
promptheus scan --mode agent \
  --target-path <path> \
  --model sonnet                         # Full codebase audit with AI agents

# PR review
promptheus pr-review --path <repo> \
  --last <N>                             # Review last N commits
promptheus pr-review --path <repo> \
  --range <base..head>                   # Review commit range

Full options: promptheus --help or promptheus scan --help.

---

Supported AI Providers

No lock-in — works with any OpenAI-compatible endpoint:

Provider	Notes
Claude (Haiku / Sonnet / Opus)	Recommended for agent scan
OpenRouter	200+ models via one API key
OpenAI	GPT-4o, GPT-4o-mini
Groq	Free tier, fast inference
Ollama	Local models, no API key needed
GLM / Zhipu AI	Alternative provider
Custom	Any OpenAI-compatible endpoint

See docs/ai-providers.md for full setup instructions per provider.

---

Interfaces

PROMPTHEUS ships across multiple surfaces — pick what fits your workflow:

Interface	Description	Guide
CLI	Full-featured terminal tool	apps/cli/README.md
Desktop App	Native app (macOS/Windows/Linux), bundled CLI, no Python needed	apps/desktop/README.md · Download
Web Dashboard	Streamlit UI — enter URL, run scan, view results visually	apps/dashboard/README.md
REST API	FastAPI server for programmatic access	apps/api/README.md
Slack Bot	@RedTeamBot attack @TargetBot — scan directly in Slack	apps/slack_bot/README.md

---

Documentation

Doc	Description
Architecture	How PROMPTHEUS uses Claude Agent SDK — agent pipeline, orchestration, data flow
Scanner Config	Cost limits, threat-aware scanning, design decisions, artifact trust
AI Providers	Setup guide for Claude, OpenAI, Groq, Ollama, OpenRouter, GLM
DAST Guide	Dynamic Application Security Testing — how it works and how to enable it
PR Review Pipeline	Architecture of the PR/commit review flow
Maintenance Guide	Dead code detection, pruning strategy, test coverage
Timeout Configuration	Agent scan timeout settings and troubleshooting
Streaming Mode	Real-time output streaming for agent scans
Example Reports	Sample scan outputs across different models and targets

---

Contributing

Pull requests welcome. Please open an issue first to discuss major changes.

git clone https://github.com/dungsnowaxe/prometheus-red-team-bot
cd prometheus-red-team-bot
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"
pytest

---

Inspiration

PROMPTHEUS was inspired by SecureVibes by Anshuman Bhartiya — an AI-powered security scanner that pioneered the multi-agent approach to codebase vulnerability scanning. Thanks for lighting the way.

---

License

MIT — free to use, modify, and distribute.