promptfoo-framework-owasp-agenticlisted

# OWASP Top 10 for Agentic Applications The OWASP Top 10 for Agentic Applications (announced Black Hat Europe 2025) covers the most critical security risks for AI agents: ASI01-ASI10. Agents introduce unique risks via autonomous decision-making, persistent memory, tool/API access, and multi-agent coordination. Use `owasp:agentic` for comprehensive testing. ## Use this skill when - Testing AI agents against OWASP Agentic Top 10 (ASI01-ASI10) - Testing agent goal hijack, tool misuse, identity/privilege abuse - Testing agentic supply chain, unexpected code execution, memory poisoning - Testing insecure inter-agent communication, cascading failures - Testing human-agent trust exploitation, rogue agents ## Do not use this skill when - Testing OWASP LLM Top 10 → use `promptfoo-framework-owasp-llm` - Testing OWASP API Security → use `promptfoo-framework-owasp-api` - Testing general agent security → use `promptfoo-redteam-agents` - Testing MCP servers → use `promptfoo-redteam-agents` (MCP section) ## 🛡️ Edge cases (mandatory handling) - **Agents differ from LLM apps** — autonomous decision-making, persistent memory, tool access, multi-agent coordination create unique risks. - **ASI04 (Supply Chain) includes MCP** — `mcp` plugin for compromised tools/plugins/servers. - **ASI06 (Memory Poisoning) includes `agentic:memory-poisoning`** — tests stateful agent memory corruption. - **ASI07 (Inter-Agent Communication) is multi-agent specific** — `indirect-prompt-injection`, `hijacking`,