The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to write and execute Python scripts (youtube_auth.py and youtube_pull.py) in the user's environment at runtime rather than providing static code.
[CREDENTIALS_UNSAFE]: The skill manages highly sensitive OAuth2 credentials, including client secrets and access tokens, and instructs the agent to store and retrieve them from a predictable path (~/.youtube-skills/). While this is a functional requirement, the predictable path makes these credentials a known target for potential data exposure.
[INDIRECT_PROMPT_INJECTION]: The skill fetches external, user-generated content (YouTube comments) via the commentThreads.list API. This data is intended to be passed to other skills, creating a vulnerability surface where malicious instructions embedded in comments could influence agent behavior.
Ingestion points: Fetched via commentThreads.list in Step 3 and saved to JSON files.
Boundary markers: The instructions do not specify any delimiters or warnings to treat this external content as untrusted.
Capability inventory: The generated scripts have file-write capabilities (metrics/<VIDEO_ID>.json) and network access (Google API calls).
Sanitization: No sanitization or filtering of comment text is described.
[COMMAND_EXECUTION]: The skill requires the agent to install and use external libraries (google-api-python-client, google-auth-oauthlib) and execute terminal commands to perform the OAuth2 flow and data retrieval.

youtube-connect