The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses coli asr to process audio files and checks for environment tools like ffmpeg using shell commands. These operations are essential for the skill's transcription functionality.
[EXTERNAL_DOWNLOADS]: The skill requires the @marswave/coli Node.js package and suggests ffmpeg installation. Additionally, the coli tool automatically downloads speech recognition models (~60MB) to the ~/.coli/models/ directory upon its first run.
[INDIRECT_PROMPT_INJECTION]: The transcription text derived from user-provided audio files is passed back to the AI for a "polish" step. This creates a potential surface where malicious instructions spoken in the audio could target the AI's refinement process.
Ingestion points: Raw transcription output from the coli asr tool in SKILL.md.
Boundary markers: The instructions in Step 4 provide guidance to the AI to preserve meaning and speaker intent while fixing formatting, acting as a soft constraint against redirection.
Capability inventory: The skill has capabilities to execute the coli CLI and write markdown transcript files to the current directory.
Sanitization: There is no explicit sanitization step for the transcription text before it is processed by the AI for refinement.

asr