safety-hooks

Claude Code safety hooks enforce security by blocking dangerous commands, protecting secrets, and logging operations. They integrate across pre-tool, post-tool, and stop hooks to prevent data leakage and unsafe actions during tool executions.

You prevent secret exfiltration by applying pre-tool and post-tool hooks that use pattern-based blocking to intercept unsafe actions. This setup automatically audits tool usage, protecting sensitive data from leaking during automated operations.

Yes, you can configure project-wide safety hooks through settings.json integration and YAML frontmatter-driven configuration. This approach implements verification and defines explicit guardrails to enforce deterministic safety checks across your Claude Code environment.

The best way to audit tool usage is by applying deterministic safety hooks across pre-tool, post-tool, and stop phases. This approach automatically logs operations, preventing unsafe actions while providing a clear audit trail of tool executions.

No external dependencies are required to set up deterministic safety hooks. The system relies on Claude Code's native settings.json integration and YAML frontmatter configuration to implement verification and explicit toxicity safeguards directly.