The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bash script (search_youtube.sh) which uses curl to interact with a specific API endpoint (ai-factory.frevana.com). The script securely handles user input by using environment variables within an embedded Python block to construct the JSON payload, avoiding shell injection risks.
[CREDENTIALS_UNSAFE]: The skill handles a FREVANA_TOKEN for authentication. It follows safe practices by reading from environment variables or a specific flag and explicitly instructs the agent not to echo the token back to the user.
[DATA_EXFILTRATION]: While the skill makes external network requests, they are directed to the service's official API endpoint (ai-factory.frevana.com) to fulfill the primary purpose of the skill (searching YouTube). It does not access or transmit sensitive local files.
[DYNAMIC_EXECUTION]: The script uses embedded Python blocks to safely process JSON data. These blocks use standard libraries and do not execute untrusted code or use unsafe functions like eval() on external input.

youtube-search