The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media content to generate text transcripts and summaries. This creates a surface for indirect prompt injection attacks where malicious instructions could be embedded in the media content. * Ingestion points: Video, audio, PDF, and image files processed in scripts/digest.py. * Boundary markers: SKILL.md contains a warning for the agent to treat transcript and on-screen text as data, not instructions. * Capability inventory: The skill can write files, execute subprocesses (ffmpeg, ASR tools), and make network requests for cloud-based transcription. * Sanitization: The skill uses standard JSON parsing and serialization to handle tool outputs.
[EXTERNAL_DOWNLOADS]: The skill documentation and runtime error messages suggest installing the uv tool using a script from astral.sh, which is a well-known service in the Python ecosystem. It also suggests installing ffmpeg via official system package managers.
[COMMAND_EXECUTION]: The skill executes media processing tools such as ffmpeg, ffprobe, and various ASR binaries via subprocess.run. These executions use argument lists and avoid shell processing, minimizing command injection risks.

media-digest