The Agent Skills Directory

[SAFE]: The skill implements a robust security model for high-risk operations. It specifically defines an 'exit 10' protocol that requires the agent to stop, present risk details to the user, and obtain explicit consent before re-executing with a confirmation flag. This prevents automated or accidental data loss.
[SAFE]: Instructions explicitly restrict file system access to relative paths (e.g., --file, --output), which mitigates potential directory traversal vulnerabilities. It further recommends using stdin for data input to avoid shell escaping issues.
[SAFE]: The guidelines strictly forbid displaying sensitive credentials like appSecret or accessToken in plain text. It also defines a clear 'user vs bot' identity model to ensure the principle of least privilege is applied based on the task context.
[SAFE]: All authentication workflows (e.g., auth login) utilize standard OAuth2 device flows with separate 'split-flow' steps, ensuring the agent never handles credentials directly and that users must authorize via official Lark verification URLs.

lark-shared