whisper
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill guides the installation of well-known Python packages including openai-whisper, transformers, and torch from official registries. It also references official documentation and repositories from OpenAI and Arxiv.
- [COMMAND_EXECUTION]: Instructions include using the whisper command-line interface and the ffmpeg utility for audio manipulation. These are standard and necessary tools for the skill's purpose.
- [PRIVILEGE_ESCALATION]: Documentation mentions using sudo for installing system-level dependencies like ffmpeg on Linux. This is a common administrative requirement for environment setup.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and transcribe untrusted audio data, which presents a surface for indirect prompt injection. 1. Ingestion points: Transcription of audio.mp3 and other media via model.transcribe and the whisper CLI in SKILL.md. 2. Boundary markers: Not explicitly demonstrated in the code examples. 3. Capability inventory: Includes shell command execution (ffmpeg, whisper) and file system access. 4. Sanitization: No specific transcription sanitization logic is included in the examples.
Audit Metadata