The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes robust defensive instructions in its mail and messaging components. It explicitly directs the agent to ignore any command-like text found within external data (emails, chats, etc.), identifying them as potential prompt injection attempts rather than genuine user intent.
[DATA_EXFILTRATION]: A proactive security scanner is implemented for the web application deployment feature. During the 'dry-run' phase, the tool automatically scans the local directory for sensitive files, including .env files, SSH keys, AWS/cloud provider credentials, and private keys (.pem, .key). It specifically warns the user and requires confirmation if these risks are detected before any data is uploaded to public endpoints.
[SAFE]: The skill acts as a secure router for Lark/Feishu operations, maintaining strict separation between instructional context and data processing. It enforces best practices such as identity confirmation, permission checks, and mandatory 'dry-run' previews for high-risk operations like bulk deletions or public deployments.

lark