The Agent Skills Directory

[PERSISTENCE]: The skill's setup documentation (references/auth-setup.md) instructs the agent to persistently store the TRANSCRIPT_API_KEY in shell profiles such as ~/.bashrc or ~/.zshrc. This technique ensures the credential remains available across different sessions and shell environments on the host system.
[OBFUSCATION]: The skill includes explicit instructions in references/auth-setup.md to circumvent agent runtime security controls. It directs the agent to write sensitive API responses (containing access_token and api_key) to temporary files specifically to avoid being caught by automated redaction filters in platforms like Hermes or Claude Code, ensuring the agent can retrieve the secret even when the platform attempts to hide it from output.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process YouTube transcripts from an external API (transcriptapi.com). This external, untrusted content presents an indirect prompt injection surface. 1) Ingestion points: The YouTube transcript endpoint (api/v2/youtube/transcript) as documented in SKILL.md. 2) Boundary markers: No instructions are provided to delimit or treat the transcript text as untrusted content. 3) Capability inventory: The skill uses curl for network communication and has instructions to modify local configuration files and shell profiles. 4) Sanitization: The instructions do not include any steps for sanitizing or filtering the external transcript text before it is processed by the agent.
[COMMAND_EXECUTION]: The skill relies on executing curl commands to interact with the YouTube transcript API, including registration and verification flows that handle sensitive authentication tokens.

yt