The Agent Skills Directory

[PROMPT_INJECTION]: The authentication guide in references/auth-setup.md instructs the agent to bypass security redaction filters. It explicitly tells the agent to write sensitive tokens and API keys to temporary files instead of printing them to the tool output, specifically to circumvent platform mechanisms that redact sensitive strings (like sk_ keys or access_token values) from the model's view.
[PERSISTENCE]: The skill directs the agent to maintain persistence by modifying the user's environment. In references/auth-setup.md, the agent is instructed to store the TRANSCRIPT_API_KEY in shell profile files (e.g., .bashrc, .zshrc) or other persistent configuration files to ensure the credential survives across different sessions.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from YouTube transcripts, which represents an injection surface.
Ingestion points: Transcript content fetched via the /api/v2/youtube/transcript endpoint defined in SKILL.md.
Boundary markers: The instructions do not provide delimiters or "ignore instructions" warnings for the fetched transcript data.
Capability inventory: The skill has network access (to transcriptapi.com) and file system access (writing and reading temporary files during the auth flow).
Sanitization: No sanitization or validation of the transcript content is specified before the agent processes or summarizes it.
[COMMAND_EXECUTION]: The references/curl-examples.md file provides multiple shell commands (curl) for the agent to execute. These commands include the use of environment variables and headers, which are standard for API interaction but represent a capability that requires monitoring.

transcriptapi