-
Notifications
You must be signed in to change notification settings - Fork 2
research: integrate Promptfoo for automated agent red-teaming #1523
Copy link
Copy link
Closed
Labels
researchResearch-driven improvementResearch-driven improvementsecuritySecurity-related issueSecurity-related issue
Description
Research
Promptfoo (github.com/promptfoo/promptfoo) is an open-source CLI for automated agent red-teaming with 50+ vulnerability types: prompt injection, jailbreaks, tool misuse, authorization bypass. YAML config, CI/CD integration. 127 Fortune 500 users.
Works as a black-box tester — can target Zeph's daemon HTTP endpoint (/a2a) and ACP HTTP+SSE transport without any Rust SDK.
Proposal
- Create Promptfoo test config (YAML) targeting daemon
/a2aendpoint - Define red-team scenarios: prompt injection via tool outputs, tool misuse escalation, sandbox bypass attempts, memory poisoning
- Add to CI as optional security gate (non-blocking initially)
Sources
- https://github.com/promptfoo/promptfoo
- https://www.promptfoo.dev/docs/red-team/agents/
- AgentAssay behavioral fingerprinting: https://arxiv.org/html/2603.02601
- Anthropic Petri framework for autonomous red-teaming
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
researchResearch-driven improvementResearch-driven improvementsecuritySecurity-related issueSecurity-related issue