azure-ai-transcription-py
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- Indirect Prompt Injection Surface: The skill processes external audio data for transcription, which represents a potential surface where spoken instructions could influence an agent when the resulting text is processed.
- Ingestion points: External data enters the context via
content_urlsin batch transcription jobs andsend_audio_filein streaming sessions as seen inSKILL.mdandreferences/non-hero-scenarios.md. - Boundary markers: The current examples do not include explicit instructions or markers to distinguish transcribed content from agent commands.
- Capability inventory: The skill is capable of outputting transcribed text to the console or as data structures, as demonstrated in the transcription loops in
SKILL.md. - Sanitization: There is no evidence of text sanitization or filtering in the provided implementation examples.
- External Data Ingestion: The skill is designed to retrieve audio content from remote storage URLs for processing. This is an expected feature for cloud-based batch transcription services.
- Secure Credential Management: The instructions correctly recommend using environment variables (
TRANSCRIPTION_KEY) orDefaultAzureCredentialrather than hardcoding credentials, which is a positive security practice for identity management.
Audit Metadata