transcribing-slack-audio
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes audio recordings from Slack, which are untrusted external inputs. The generated transcript is subsequently treated by the AI agent as a direct instruction.
- Ingestion points: Audio files provided as Slack attachments are read and processed in
scripts/transcribe.py. - Boundary markers: The script outputs the transcript prefixed with a file name header
=== <file name> ===. - Capability inventory: The skill utilizes
subprocess.runto invoke local audio processing tools (ffmpeg,ffprobe, andwhisper-cli). - Sanitization: The script performs basic cleanup of the transcript (e.g., removing noise markers) but does not sanitize the content for adversarial prompts or instructions.
- [COMMAND_EXECUTION]: The
scripts/transcribe.pyscript executes system commands to perform audio conversion and transcription. - Evidence: Calls to
subprocess.runare used to invokeffmpeg,ffprobe, andwhisper-cli. The calls are implemented securely by passing arguments as a list, which prevents shell injection.
Audit Metadata