The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute a local utility script, scripts/cursor-companion.sh, which serves as a bridge to the Cursor external agent. The execution logic safely handles user input by passing it as a positional parameter to a subshell, preventing shell injection vulnerabilities.
[DATA_EXFILTRATION]: The skill documentation addresses the risk of data exposure by providing explicit instructions to use .cursorignore. This ensures that sensitive directories and files, such as .ssh, .aws, and .env, are excluded from being read by the external inference process.
[PROMPT_INJECTION]: The skill identifies a potential indirect prompt injection surface where external project data could influence the agent's behavior. It mitigates this risk by instructing the host AI to analyze and summarize the external output (3-5 lines) rather than displaying it verbatim, acting as a security and brevity filter.

cursor-ask