agent-platform-tuning

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.75). Outsider free text can enter the LLM context via dataset preview: Phase 1.0/1.1 instructs reading and showing “a few samples” from a Hugging Face dataset (references/hf_datasets.md) or from user-specified/auto-discovered files in GCS, and those samples are arbitrary outsider-authored text that the agent would render into chat/LLM context for column selection.