fun-asr
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The script
scripts/fun_asr_cli.pyusessubprocess.runto callffmpegandffprobefor necessary audio processing tasks, including mono conversion and duration probing. These executions use list-based arguments and are required for the skill's stated functionality. - [EXTERNAL_DOWNLOADS]: The skill identifies dependencies on standard Python packages
boto3,requests, andpython-dotenv. It communicates with official Alibaba Cloud DashScope API endpoints and S3-compatible storage providers for task submission and file uploads. - [DATA_EXFILTRATION]: The skill is designed to upload local audio files to a user-configured S3 bucket and send presigned URLs to Alibaba Cloud's transcription service. This is a functional requirement for the service and uses user-provided credentials from environment variables.
- [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing external audio content and instructing the agent to read the resulting transcription.
- Ingestion points: Audio files ingested by
scripts/fun_asr_cli.pyand the transcription output read by the agent as instructed inSKILL.md. - Boundary markers: Absent.
- Capability inventory: Execution of system binaries (
ffmpeg,ffprobe), network communication viarequests, and cloud storage interaction viaboto3. - Sanitization: No sanitization or filtering of transcribed text is performed before it is presented to the agent.
Audit Metadata