<!-- lyrie-shield: ignore-file (documentation, not vectors) --> <div align="center">

🛡️ Lyrie

The autonomous security agent.

Pentests apps. Defends agents. Researches binaries. Trains itself. One daemon.

Install · Quick Start · Commands · ATP · Security

</div>

---

🆕 What's New in v3.3.0

• Self-Heal Loop: closed end-to-end — detect → remediate → revalidate → attach, with automated remediation-PR revalidation and rogue-AI hooks
• Compression Dashboard: attack-compression telemetry aggregated across store, server, and a live UI
• Living SBOM: generate, revalidate, and schedule software bill-of-materials with typed manifests
• ATP Vaccine & Kill-Switch: agent vaccine, trust attestation, and kill-switch integration for the Agent Trust Protocol
• AAV Expansion: Crescendo + TAP multi-turn adversarial attack strategies (alongside PAIR / GCG / AutoDAN)
• 1,977 tests passing across @lyrie/atp, @lyrie/core, @lyrie/gateway, @lyrie/mcp, @lyrie/ui
• Fully backward compatible with v3.x — no migration required

Since v3.1.0

• v3.2.0 — ATP CLI, standalone MCP Shield middleware, lyrie hack trust score, OAST audit
• v3.2.1 — memory-recall importance fix
• v3.2.2 — stability & maintenance

See CHANGELOG.md for the complete list.

---

What is Lyrie?

Lyrie is an autonomous security agent built by OTT Cybersecurity LLC. It runs end-to-end pentests, red-teams LLM endpoints, scans code and live URLs, and ships with the Agent Trust Protocol (ATP) — the first open cryptographic standard for AI agent identity.

Two installs, one tool:

| Component | Language | Install | What it does | |---|---|---|---| | lyrie-omega | Python | pip install lyrie-omega | CLI for scanning, pentesting, red-teaming, governance | | @lyrie/atp | TypeScript/Node | npm install @lyrie/atp | Agent Trust Protocol SDK — cryptographic agent identity |

---

🚀 Install

# Option 1: one-line installer (installs both)
curl -sSL https://lyrie.ai/install.sh | bash

# Option 2: install separately
pip install lyrie-omega
npm install @lyrie/atp

After install:

lyrie init                  # one-time setup wizard
lyrie doctor                # verify everything works

---

⚡ Quick Start

# Scan a live URL for security misconfigurations
lyrie scan https://app.example.com

# Run a 7-phase autonomous pentest
lyrie hack https://app.example.com
lyrie hack ./myapp                          # local source tree
lyrie hack ./myapp --stage scan --output report.json

# AI red-team an LLM endpoint
lyrie redteam https://api.openai.com/v1/chat --strategy crescendo --dry-run

# Check CVSS score
lyrie cvss 'CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H'

# Self-diagnostic
lyrie doctor

---

📋 Commands

All 25 commands are real and tested. Run lyrie <command> --help for details.

Core security

lyrie hack <target>              # 7-phase autonomous pentest (URL or local path)
lyrie scan <target>              # Scan file/dir/URL for vulnerabilities
lyrie redteam <endpoint>         # AI red-team an LLM endpoint
lyrie cvss <vector>              # CVSS v3.1 scoring
lyrie exploit --cve <id>         # SMT-backed exploit feasibility
lyrie validate --target <url>    # Agentic exploitability validation
lyrie intel --repo <url>         # GitHub OSS forensics evidence collection
lyrie smt --check <expr>         # Z3 SMT solver interface

Binary analysis (Omega)

lyrie omega analyze <binary>     # Static binary analysis
lyrie omega rop <binary>         # ROP gadget search
lyrie omega smt <binary>         # SMT constraint analysis
lyrie omega replay <session>     # Replay recorded session

Identity & trust (ATP)

lyrie atp verify <agent-id>      # Verify agent identity + scope
lyrie atp badge --show           # Display compliance badge
lyrie atp receipt <session-id>   # Audit trail for a session

Operations

lyrie init                       # First-time setup wizard
lyrie doctor                     # Self-diagnostic (env, deps, keys, network)
lyrie auth setup                 # Configure API keys interactively
lyrie auth set --key NAME        # Set a specific key (prompts securely)
lyrie auth list                  # Show configured keys (redacted)
lyrie config show                # Show config file contents
lyrie config path                # Print config file path

Automation & lifecycle

lyrie daemon --threat-watch      # Continuous threat detection
lyrie service install            # Install as system service (launchd/systemd)
lyrie service status             # Service status
lyrie cron list                  # List scheduled jobs
lyrie cron add "*/5 * * * *" "lyrie scan https://example.com"

Governance & compliance

lyrie governance assess --interactive     # NIST AI RMF 8-question assessment
lyrie governance permissions tools.json   # Audit tool permissions for risk
lyrie tools audit                         # Risk assessment of installed tools
lyrie memory integrity-check              # Detect tampered memories

Self-improvement

lyrie evolve dream               # Full cycle: score → extract → prune → summarize
lyrie evolve stats               # Domain breakdown
lyrie evolve train --export atropos       # Export training data

Models & migration

lyrie models list                # List available LLM aliases
lyrie models route <task-type>   # Show routing decision (cyber, code, seo, trading)
lyrie models health              # Health-check all model providers
lyrie migrate --detect           # Auto-detect existing agent platforms
lyrie migrate --from openclaw    # Import from another platform

Skills

lyrie skills list                # List installed skills
lyrie skills search <query>      # Search skill library
lyrie skills install <skill-id>  # Install a skill
lyrie skills run <skill-id>      # Execute a skill

---

🛡️ Capabilities

Autonomous pentesting (lyrie hack)

7-phase pipeline: recon → fingerprint → scan → exploit → PoC → report. Works on live URLs and local source trees. Outputs SARIF for GitHub Code Scanning.

URL security scan (lyrie scan <url>)

Checks every site for:

• Security headers (CSP, HSTS, X-Frame-Options, etc.)
• TLS version and cert expiry
• Common exposed paths (.env, .git/config, /admin, etc.)
• Server version disclosure

AI red-teaming (lyrie redteam)

5 attack strategies against LLM endpoints:

• crescendo — gradual escalation
• tap — tree-of-attacks-with-pruning
• pair — prompt automatic iterative refinement
• gcg — gradient-based suffix attack (full: H200 required)
• autodan — genetic algorithm black-box (full: GPU required)

Agent Trust Protocol (ATP)

Open cryptographic standard for AI agent identity. Ed25519 signatures, delegation chains, revocation lists, multisig. Spec at atp.lyrie.ai. 143 tests passing.

Lyrie Shield (Rust)

Production-grade security engine: hash-signature scanning, heuristic analysis, WAF, rogue-AI detector. 31 tests passing.

---

🔐 ATP — Agent Trust Protocol

The first open cryptographic standard for AI agent identity. Think TLS for agents.

import { issueCertificate, verifyAic } from '@lyrie/atp';

// Issue a scoped certificate
const aic = await issueCertificate({
  subjectPublicKey: agentPubKey,
  scope: { tools: ['scan', 'read'], maxBudget: 100 },
  issuerPrivateKey: rootKey,
  ttlSeconds: 3600,
});

// Verify it
const result = await verifyAic(aic, trustAnchor);
if (result.valid) {
  // Agent is authorized
}

Full spec: atp.lyrie.ai · Whitepaper PDF

---

🔑 Configuration

# Interactive setup
lyrie auth setup

# Or set individual keys
lyrie auth set --key ANTHROPIC_API_KEY    # prompts securely (no shell history)
lyrie auth set --key OPENAI_API_KEY
lyrie auth set --key GITHUB_TOKEN

# View configured keys (redacted)
lyrie auth list

Keys are stored at ~/.lyrie/config.json with chmod 600 (user-only).

Known keys: ANTHROPIC_API_KEY, OPENAI_API_KEY, GITHUB_TOKEN, LYRIE_LICENSE_KEY, CODEQL_CLI, CODEQL_QUERIES.

---

🏗️ Architecture

┌─────────────────────────────────────────────────┐
│                  lyrie CLI                       │
│  (Python — lyrie-omega, this repo)               │
└────────────┬─────────────────────────────────────┘
             │
   ┌─────────┴─────────┐
   ▼                   ▼
┌──────────┐      ┌──────────────┐
│ omega    │      │  @lyrie/atp  │
│ engine   │      │  (Node.js)   │
│ (Rust +  │      │              │
│  Python) │      │  Ed25519     │
│          │      │  delegation  │
│ CodeQL,  │      │  revocation  │
│ SMT, ROP │      │  multisig    │
└──────────┘      └──────────────┘

• packages/atp/ — TypeScript Agent Trust Protocol SDK (npm: @lyrie/atp)
• packages/omega-suite/ — Python CLI + analysis engines (PyPI: lyrie-omega)
• packages/shield/ — Rust security scanner (WAF + rogue-AI + threat scoring)

---

✅ Quality

• ATP: 143 tests passing
• **Core: