The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions in references/auth-setup.md that explicitly direct the agent to circumvent platform-level security redaction. It advises the agent to write sensitive information, such as access_token and api_key values, to temporary files specifically to prevent them from being detected or redacted by tool-output security filters.
[PERSISTENCE]: The instructions in references/auth-setup.md guide the agent to persistently store credentials by modifying shell configuration files like ~/.bashrc and ~/.zshrc. This allows the skill's environment variables and potential modifications to survive across different shell sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (YouTube transcripts) which presents an attack surface where malicious instructions could be embedded in video captions to influence the agent's behavior.
Ingestion points: Transcripts are fetched from the external transcriptapi.com service as defined in SKILL.md.
Boundary markers: There are no instructions to use delimiters or ignore instructions that might be contained within the transcript data.
Capability inventory: The agent is granted capabilities for network access (HTTP GET/POST) and file system modification (writing to profile and temporary files).
Sanitization: The skill does not implement or recommend any sanitization, validation, or filtering of the transcript content before it is analyzed or summarized by the agent.

video-transcript