The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a Node.js relay script (relay.mjs) to execute the cursor-agent and git command-line tools. The script implements secure execution practices, including regex-based validation for model and session identifiers, and specific protections for Windows shell environments such as character blacklisting (%, !, quotes, newlines) and rigorous argument quoting to prevent command injection.- [EXTERNAL_DOWNLOADS]: The skill documentation provides instructions for users to manually install the cursor-agent CLI from an official, well-known service (cursor.com). It does not perform any automated downloads or background execution of external code at runtime.- [DATA_EXFILTRATION]: The relay script writes logs, event streams, and task artifacts to the system's temporary file directory. It does not perform network operations, hardcode credentials, or attempt to exfiltrate sensitive data from the host environment.- [PROMPT_INJECTION]: The skill provides structured guidance for orchestrating tasks through separate implementer sessions. It includes specific recommendations for boundary markers in task briefs and a 'review-and-land' workflow that ensures all generated output is manually verified and committed by the orchestrator, preventing autonomous or unsafe repository modifications.

cursor-delegate