The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local script execution and system binaries to perform its tasks.
It executes various Python and Shell scripts within the scripts/ directory (e.g., package_youtube.py, generate_voice.py, create_new_session.sh) to automate the workflow.
It uses ffmpeg and ffprobe for heavy audio/video processing, including complex filtering and metadata extraction.
[CREDENTIALS_UNSAFE]: The voice-synthesis module requires and verifies the presence of GOOGLE_APPLICATION_CREDENTIALS. While it does not hardcode secrets, it directs the agent to interact with sensitive environment variables used for cloud authentication.
[EXTERNAL_DOWNLOADS]: The youtube-packaging skill contains logic for uploading content to the external domain salars.net. This is identified as a vendor-owned resource belonging to the author 'randysalars'.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection through its data ingestion workflows.
Ingestion points: session-creation/SKILL.md and ssml-generation/SKILL.md ingest natural language 'topic descriptions' and 'journey concepts'.
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands are documented for the processing of these inputs.
Capability inventory: The agent has permissions to execute subprocesses (ffmpeg, python3) and write files to the local system.
Sanitization: There is no evidence of sanitization or validation of the input strings before they are interpolated into the generated SSML scripts or manifests.

Production Operations