The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill makes network requests to https://api.qwenlm.com/v1/asr to perform audio transcription. This service is the official API endpoint for the Qwen AI model, and its use is consistent with the skill's primary function.
[CREDENTIALS_UNSAFE]: API credentials are managed through environment variables such as COZE_QWEN3ASR_API_7598467069021093924. This follows recommended security practices for secret management, and no hardcoded keys or tokens were found.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input in the form of transcribed audio content. Ingestion points: Untrusted data enters the agent's context via the transcriber.transcribe call documented in SKILL.md. Boundary markers: Absent; instructions do not specify the use of delimiters for the transcribed text. Capability inventory: The skill includes scripts/asr_transcriber.py which performs network operations (ASR API) and local file reads (audio files), but it lacks high-risk capabilities like arbitrary command execution or dynamic code evaluation. Sanitization: Absent; the transcription results are passed directly to the LLM. While this presents an attack surface, the risk is minimal given the restricted nature of the skill's outputs (text formatting and rewriting).

qwen3-asr-assistant