tool-permissions

AI agent permissions are controlled by defining permission modes and rule-based confirmations for tool access. You can set allow, ask, deny, or accept-edit behaviors for specific shell commands, file paths, and edits to enforce workspace governance.

The best way to protect sensitive environment files is by configuring security guardrails with deny rules for specific paths. This prevents accidental writes to files like .env and keys while still allowing the agent to perform routine reads and safe operations.

Yes, you can switch between default prompting, read-only planning, trusted edit workflows, or fully sandboxed execution. These permission modes let you safely restrict the agent to reading code without executing risky commands or writing files.

Workflow governance is configured by applying rule-based access control that matches tool categories and command conditions. You set the behavior to ask before shell execution, ensuring the agent requires explicit confirmation for potentially unsafe commands.

Accept-edit rules should be used when you want to auto-accept edits in source files without manual confirmation, while allow rules broadly permit tool access. Use accept-edit for trusted edit workflows and allow for routine safe reads.

Yes, command safety enforces ordered fallback behavior to determine allow, ask, deny, or accept-edit decisions. This ensures the agent evaluates matching tool categories, path conditions, and command rules sequentially to apply the correct permission.