Feature Request: Blocking Hooks (PreToolUse that can prevent execution)

Summary

Request for Claude Code to support blocking hooks that can prevent tool execution, not just log it.

---

Current Behavior

Hooks like PreToolUse are informational only:

• Hook runs before tool execution
• Hook can log, warn, or output messages
• Hook cannot prevent the tool from executing
• Non-zero exit code is noted but doesn't block

---

Requested Behavior

Allow hooks to block tool execution based on exit code:

PreToolUse hook exits 0     → Tool executes normally
PreToolUse hook exits non-0 → Tool execution BLOCKED, message shown to user

---

Use Cases

1. Compliance Enforcement

Organizations with security policies need to enforce rules, not just document them:

• Block writes to sensitive paths
• Require approval before external API calls
• Prevent accidental secret exposure

2. Human-in-the-Loop Gates

Some actions should require explicit human approval:

• Pushing to remote repositories
• Sending external communications
• Modifying production systems

3. Safe Automation

For autonomous/unattended Claude Code sessions:

• Prevent runaway actions
• Enforce iteration limits
• Block known-dangerous patterns

---

Proposed Implementation Options

Option A: Exit Code Enforcement

# .claude/hooks/PreToolUse
# Exit 0 = allow, Exit 1 = block

if [[ "$TOOL_NAME" == "Bash" ]]; then
  if echo "$TOOL_INPUT" | grep -q "rm -rf /"; then
    echo "BLOCKED: Dangerous command pattern"
    exit 1  # This should BLOCK execution
  fi
fi
exit 0

Option B: Policy File

// .claude/policy.json
{
  "rules": [
    {
      "tool": "Bash",
      "pattern": "git push.*--force",
      "action": "block",
      "message": "Force push requires manual execution"
    }
  ]
}

Option C: Hook Response Format

// Hook outputs JSON to stdout
{
  "action": "block",
  "reason": "External communication requires approval",
  "allow_override": true
}

---

Security Considerations

• Hooks should be tamper-evident (hash verification)
• Block decisions should be logged
• User should see clear feedback when blocked
• Consider "allow override" flag for soft blocks vs hard blocks

---

Business Impact

Without blocking hooks:

• Security policies are advisory only
• Compliance frameworks (ISO 27001, SOC 2) require enforceable controls
• Organizations can't safely use Claude Code for sensitive workflows
• Autonomous agents can't be trusted without guardrails

---

Happy to discuss further or provide more detail on any of these use cases.