promptfoo-redteam-foundation-modelslisted

# Red Team Foundation Models with Promptfoo LLM security starts at the foundation model level. Assess base/fine-tuned models with two complementary approaches: (1) **dynamic** red team scans against live models using the `foundation` plugin + canonical strategies, and (2) **static** scanning of model files with ModelAudit for trojans, malicious pickle payloads, and embedded executables. Use HarmBench for standardized safety benchmarking. ## Use this skill when - Baseline security assessment of a foundation or fine-tuned model - Comparing multiple foundation models side-by-side (ASR comparison) - Scanning model files for trojans, malicious pickle, embedded executables (`promptfoo scan-model`) - Running HarmBench standardized benchmarks (400 harmful behaviors) - Testing fine-tuned models against their base for safety regression - Contributing results to promptfoo.dev/models ## Do not use this skill when - Testing an LLM application with prompts/guardrails/context → use `promptfoo-redteam-llm` - Testing RAG, agents, MCP, or multi-input apps → use the corresponding skill - Setting up CI/CD drift detection → use `promptfoo-redteam-supply-chain` (this skill establishes the baseline) - Testing guardrails → use `promptfoo-redteam-guardrails` ## 🛡️ Edge cases (mandatory handling) - **Skipping static scans for downloaded models** — pickle deserialization executes arbitrary code; ALWAYS scan with `promptfoo scan-model --strict` before deploying any downloaded model. - **Only testin