The Agent Skills Directory

[OBFUSCATION]: The skill includes instructions in references/auth-setup.md to evade platform-level security redaction of sensitive credentials. It directs the agent to write HTTP responses containing API keys and bearer tokens to temporary files instead of standard output, specifically to circumvent automated security filters designed to detect and mask secrets during tool execution.
[PERSISTENCE]: In references/auth-setup.md, the agent is instructed to persistently store the TRANSCRIPT_API_KEY environment variable by modifying system configuration files like shell profiles (e.g., ~/.bashrc, ~/.zshrc) or dedicated environment files. This modification ensures the API key remains available across different sessions and shell environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted text from YouTube transcripts which could contain adversarial instructions, presenting an indirect prompt injection risk.
Ingestion points: Transcripts are fetched from https://transcriptapi.com/api/v2/youtube/transcript as specified in SKILL.md and references/curl-examples.md.
Boundary markers: The instructions do not define any delimiters or provide guidance to ignore instructions embedded within the fetched transcript data.
Capability inventory: The skill uses curl for network operations and provides instructions for local file system modifications to store credentials.
Sanitization: No validation or sanitization of the fetched transcript content is mentioned or implemented in the skill instructions.

subtitles