The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The scripts/upload_youtube.py script automatically searches for Google OAuth client secret files in the user's ~/Downloads/ and ~/.config/google/ directories. This behavior broadens the access scope for sensitive credential files to standard user download locations.
[COMMAND_EXECUTION]: The skill executes system utilities including ffmpeg and ffprobe via subprocess.run to process and merge audio and video files. While these calls incorporate user-supplied text for titles and subtitles, the skill includes an escape_drawtext function to sanitize these inputs and mitigate shell injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external URLs, tweets, and articles to generate podcast scripts, creating a surface for indirect prompt injection.
Ingestion points: Data enters the agent's context through WebFetch and PDF reading tools during the source collection phase (SKILL.md, Step 1).
Boundary markers: There are no explicit instructions or delimiters provided to the agent to disregard instructions embedded within the fetched source materials.
Capability inventory: The skill possesses capabilities for file writing, network operations to OpenAI and Google APIs, and shell command execution for media processing.
Sanitization: While technical escaping is performed for ffmpeg parameters, there is no semantic sanitization or filtering of the external content before it is processed for script generation.
[EXTERNAL_DOWNLOADS]: The skill fetches data from external sources and interacts with the official OpenAI and Google APIs for text-to-speech generation and video uploads. These operations target well-known and established services.

podcast