The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill uses a script (extract-cookies.mjs) to extract YouTube session cookies from the user's Chrome browser and store them in a local file named .cookies.json. This process handles sensitive session tokens that could be misused if the local environment is compromised.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands using Node.js to check authentication status and manage session cookies via the extract-cookies.mjs script.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data from YouTube video transcripts, creating a potential vulnerability to indirect prompt injection.
Ingestion points: Transcripts are retrieved using the youtube_get_transcript tool as defined in SKILL.md and referenced in references/gemini-tools.md.
Boundary markers: The instructions lack explicit boundary markers or warnings to the agent to ignore instructions that might be embedded within the retrieved transcript text.
Capability inventory: The skill has the ability to write files to the local system (youtube_download, youtube_clip) and execute shell commands.
Sanitization: There is no evidence of sanitization, filtering, or escaping applied to the transcript data before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill includes tools (youtube_download and youtube_clip) that facilitate downloading media content from external YouTube servers to the local file system.

youtube