audio-transcribe

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

• Third-party content exposure detected (high risk: 0.90). The skill explicitly sends transcript excerpts to external LLM providers (AWS Bedrock / Anthropic / OpenAI) for “LLM cleanup” and speaker-role verification (see SKILL.md Phase 3 and the use of scripts/llm_utils.py and verify_speakers.py), and those untrusted third-party LLM responses are parsed and used to change speaker labels and post-processing decisions.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill's optional MiMo path requires cloning and later importing/executing third‑party code from the GitHub repo (setup_mimo.sh: https://github.com/XiaomiMiMo/MiMo-V2.5-ASR.git) and uses Hugging Face weights (https://huggingface.co/XiaomiMiMo/MiMo-V2.5-ASR), so externally fetched code/assets control runtime behavior and are required for the --lang mimo mode.