The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted audio files to produce text transcripts, creating a surface where malicious instructions embedded in audio could influence the agent's behavior.
Ingestion points: Audio data is read from the filesystem in scripts/stt.py and uploaded to the transcription API.
Boundary markers: None; the skill returns the raw transcript text or JSON structured data without explicit delimiters to the agent.
Capability inventory: The skill utilizes network access to communicate with the transcription service and filesystem access to read source audio, write output transcripts, and manage its configuration file.
Sanitization: No sanitization is performed on the audio binary or the resulting transcript string.
[SAFE]: All network operations are directed to the vendor's official API endpoint (noiz.ai). The skill follows security best practices for local credential storage, using a configuration file with restrictive 0600 permissions to prevent unauthorized access by other local users.

speech-to-text