Skills
Navigate
skills/stvlynn/skills/qwen-asr/Gen Agent Trust Hub

qwen-asr

Fail

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: HIGHCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The client script scripts/asr.py uses subprocess.Popen to execute the local service/start.sh script, managing the lifecycle of the transcription service.
  • [PERSISTENCE]: SKILL.md includes manual instructions for users to configure a macOS launchd plist (com.qwen.asr.plist) to ensure the local service persists across system reboots.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and transcribes audio from external URLs via the /transcribe_url endpoint, creating a potential attack surface for instructions hidden in speech. 1. Ingestion points: audio_url parameter in scripts/asr.py and service/main.py. 2. Boundary markers: Absent. 3. Capability inventory: Local file writing (temporary audio files), network fetching, and subprocess execution. 4. Sanitization: Absent for transcribed audio content.
  • [EXTERNAL_DOWNLOADS]: The service downloads model weights from a well-known mirror for the trusted HuggingFace platform (hf-mirror.com) during its initial execution.
Recommendations
  • HIGH: Downloads and executes remote code from: unknown (check file) - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 9, 2026, 08:46 AM