Summary

Add a tool:before or exec:before hook event that fires before a tool executes, allowing hooks to intercept, modify, or block tool calls.

Use Case

I want to block specific dangerous commands (like gog gmail send, gog gmail delete) while allowing everything else to run freely. Currently:

• The exec-approvals system only supports allowlist mode (no denylist)
• The tool_result_persist hook fires after execution (too late to block)
• There's no way to intercept commands before they run

Proposed Solution

Add a new hook event:

// Event structure
{
  type: 'tool',
  action: 'before',
  toolName: string,        // e.g., 'exec', 'read', 'write'
  toolArgs: object,        // tool arguments
  command?: string,        // for exec: the command being run
  sessionKey: string,
  timestamp: Date,
  // Hook can set this to block execution
  blocked?: boolean,
  blockReason?: string
}

Example Handler

const handler: HookHandler = async (event) => {
  if (event.type !== 'tool' || event.action !== 'before') return;
  
  if (event.toolName === 'exec' && event.command?.match(/gog gmail (send|delete)/)) {
    event.blocked = true;
    event.blockReason = 'Email send/delete requires manual approval';
  }
};

Alternatives Considered

1. Wrapper scripts - Works but fragile, requires PATH manipulation
2. Allowlist mode - Requires maintaining a huge allowlist of "safe" commands
3. Agent guardrails - Soft guidance only, not enforced

Environment

• OpenClaw 2026.1.30
• macOS