The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes information from external X and LinkedIn profiles to rank connections and create outreach drafts, which could expose the agent to malicious instructions hidden in those profiles. Ingestion points: External social media profiles and activity history accessed via tools like x-api, lead-intelligence, and browser control. Boundary markers: The skill does not implement specific delimiters for input data but enforces a 'Review Pack' format that requires user confirmation. Capability inventory: The agent utilizes browser control and desktop automation to interact with LinkedIn and Apple Mail. Sanitization: There are no instructions provided for sanitizing or escaping the profile data before it is processed.
[COMMAND_EXECUTION]: The skill employs desktop automation for Apple Mail and browser control for LinkedIn tasks, allowing the agent to perform actions within these local and web-based applications.
[SAFE]: The skill defines robust safety defaults that prevent the automatic pruning of connections and strictly prohibit the auto-sending of DMs, invites, or emails, ensuring the user maintains full control.

connections-optimizer