Auto-SecAge is red-team research tooling for one sealed, offline benchmark: the Kaggle AI Agent Security – Multi-Step Tool Attacks environment (AgentDojo lineage). Everything it generates targets that benchmark's mock tools and synthetic fixtures:

• URLs use the reserved .invalid TLD (RFC 2606) and are never resolved.
• Secrets are fixture-style strings from the competition's own secret.txt.
• There is no live HTTP, no live mail, and no real filesystem outside the sandbox's isolated fixture tree.

Auto-SecAge does not acquire, modify, or attack real systems, and it is not a general-purpose exploitation framework.

Authorized evaluation and defensive research: understanding how a tool-using agent can be talked into an unsafe tool call, and how a guardrail and scorer disagree about what counts as one. Findings are replay-dependent. Expert review remains mandatory — see DISCLAIMER.md.

• Do not point these arms at a third-party agent, product, or deployment.
• Do not file exploit proof-of-concepts against third-party agents in this repo's issue tracker.
• Do not repurpose the catalog against systems you are not authorized to test.

If you believe something here is unsafe, mis-scoped, or leaks material it should not, please report it privately to the maintainer rather than opening a public issue, and allow reasonable time to respond before disclosure.