Lyrie — Autonomous Security Agent CLI + ATP SDK

Practical Notes

• GitHub: 532 stars · 52 forks; pushed 2026-05-12 (verified via GitHub API).
• README lists “25 commands are real and tested” and includes lyrie hack (7-phase pentest) and lyrie redteam strategies.
• README claims ATP spec has 143 tests passing and “Lyrie Shield” has 31 tests passing.

Main

What to do with Lyrie beyond a demo:

1. Use lyrie doctor as your “preflight” before long scans—treat failures as environment issues, not model issues.
2. Split workflows into:
   • scan for broad coverage
   • hack for deeper, staged work (URL or local source tree)
   • redteam for LLM endpoint testing
3. For agent platforms, ATP gives you a vocabulary for identity + scope: certificates, delegation, and audit trails. Use it to reduce “mystery tools” risk.

If you already have tools, Lyrie is most valuable as a unifying CLI + trust layer, not as yet another scanner.

FAQ

Q: What are the two installs? A: Per README: lyrie-omega is the Python CLI and @lyrie/atp is the TypeScript SDK for Agent Trust Protocol.

Q: What’s the quickest sanity check? A: Run lyrie init once, then lyrie doctor to verify setup (README examples).

Q: Can it red-team LLM endpoints? A: Yes. README includes lyrie redteam <endpoint> --strategy crescendo --dry-run.