The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive credential files to manage API access. Evidence: scripts/auth_write.py and scripts/update_metadata.py read and write OAuth tokens and client secrets in the ~/.config/youtube-skills/ directory (or a path specified by YT_OAUTH_DIR). While these are used for legitimate API interactions with a trusted provider, they constitute access to sensitive filesystem locations.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted metadata from the YouTube API which could theoretically contain malicious instructions designed to influence the agent.
Ingestion points: Video titles and descriptions are fetched in scripts/update_metadata.py using the youtube.videos().list method.
Boundary markers: The skill presents a markdown table diff to the user for review but does not use explicit prompt boundary markers or "ignore embedded instructions" warnings when interpolating metadata into the conversational context.
Capability inventory: The skill can write backups to the local filesystem (reports/backups/) and update remote YouTube metadata via youtube.videos().update.
Sanitization: The skill validates metadata length and filters angle brackets (<, >) before submitting updates to the API.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute local Python scripts to manage authentication and metadata updates. Evidence: SKILL.md provides multiple commands to run python3 scripts/auth_write.py and python3 scripts/update_metadata.py to perform the skill's primary functions.

youtube-metadata-updater