promptfoo-framework-owasp-llmlisted

# OWASP LLM Top 10 The OWASP Top 10 for LLM Applications lists the top 10 critical vulnerabilities. Promptfoo maps each to plugins via `owasp:llm`. Includes Gen AI Red Team best practices (4 phases: Model, Implementation, System, Runtime). ## Use this skill when - Testing OWASP LLM Top 10 (2025) compliance - Testing prompt injection (LLM01), PII disclosure (LLM02), supply chain (LLM03) - Testing excessive agency (LLM06), system prompt leakage (LLM07) - Testing RAG/vector weaknesses (LLM08), misinformation (LLM09), unbounded consumption (LLM10) - Running OWASP Gen AI red team best practices (4 phases) ## Do not use this skill when - Testing OWASP Agentic Applications → use `promptfoo-framework-owasp-agentic` - Testing OWASP API Security → use `promptfoo-framework-owasp-api` - Testing NIST AI RMF → use `promptfoo-framework-nist-ai-rmf` - Testing general LLM app security → use `promptfoo-redteam-llm` ## 🛡️ Edge cases (mandatory handling) - **`prompt-extraction` plugin requires `systemPrompt` config** — must provide the actual system prompt for LLM07 testing. - **LLM03 (Supply Chain) needs ModelAudit + dynamic testing** — static scanning + behavioral comparison across model versions. - **LLM04 (Poisoning) can't be directly prevented** — test for effects (bias, inconsistency) not the poisoning itself. - **LLM05 (Output Handling) uses assertions not plugins** — `not-contains` assertions for script tags, HTML, etc. - **LLM08 (Vector/Embedding) includes RAG poisoning CLI** — `pr