tao-generate-referring-expressions

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.80). This skill explicitly instructs the agent to "collect" API keys and to set fields like vlm.gemini.api_key or vlm.*.api_key (and shows usage in a CLI example), which requires the LLM to accept and potentially include secret values in generated configs/commands — an insecure pattern unless handled strictly via environment variables, but the prompt allows direct collection/storage.

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

• Third-party content exposure detected (medium risk: 0.30). The skill’s runtime workflow ingests outsider-authored free text when it reads KITTI label .txt files (object types/descriptions) from user-provided directories or resumes from data.input_annotations_jsonl, and those text fields are then sent to the VLM for generation/verification.