optimus:permissions

To prevent a Claude Code agent from deleting files outside your project directory, configure path-restriction hooks with allow and deny rules. This enforces safe agent autonomy by blocking destructive operations like arbitrary file deletion while permitting reads and edits within the project scope.

Yes, you can restrict Git push operations to specific branches for an AI agent by applying branch-aware Git operation restrictions. This security configuration blocks destructive commands like git push --force to protected branches, mitigating risks associated with AI-driven development.

Configure Claude Code permissions with granular allow and deny rules to stop routine approval prompts. This streamlines your workflow by eliminating manual confirmation for safe, common operations while maintaining strict guardrails against dangerous commands.

Limitations of using allow and deny rules for AI agent code execution include the strict confinement to a specified project directory. Operations requiring writes or modifications to external locations are blocked automatically and require explicit confirmation to proceed.

Yes, Claude Code works with tiered security configurations for precious file protection. This approach enforces multiple layers of guardrails, automatically blocking accidental deletion of critical project files and preventing unauthorized writes to external paths.