tao-generate-referring-expressions

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The prompt instructs the assistant to collect API keys (e.g., Gemini/api_key, OpenAI-compatible api_key) and to place them into config fields or command-line overrides, which requires handling and potentially emitting secret values verbatim in generated specs/commands.

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

• Third-party content exposure detected (medium risk: 0.65). Outsider free text can enter the LLM context via the VLM prompts in Steps 0/1/2/3 using the per-image KITTI label lines (parsed from data.kitti_label_dir or data.input_annotations_jsonl), which are user-supplied and may be authored by outsiders; the pipeline reads those text files at runtime and includes their contents in the LLM request.