• warn — shows message, allows operation
• warn is too weak — the model can reason past it The warn action relies on the model's self-discipline to follow the warning message. As documented in issue #46991, the model can reason past warnings by deciding conversation context implies approval. The ask action would put the decision in the user's hands via a native Claude Code permission prompt, which the model cannot bypass.

TL;DR

To implement the requested feature, modify the hookify rule engine to support an action: ask type that returns a permissionDecision: "ask" response, prompting the user for approval.

Guidance

• Update the core/rule_engine.py file to handle the new action: ask type and return the proposed JSON response with permissionDecision: "ask".
• Add support for the action: ask type in the hookify rule parser to correctly interpret the new action type in rule files.
• Test the new feature with a sample rule file, such as the provided confirm-systemctl example, to ensure it behaves as expected.
• Consider adding documentation and examples for the new action: ask type to help users understand its usage and benefits.

Example

# In core/rule_engine.py
if rule['action'] == 'ask':
    return {
        "hookSpecificOutput": {
            "hookEventName": "PreToolUse",
            "permissionDecision": "ask"
        },
        "systemMessage": rule['message']
    }

Notes

The implementation details may vary depending on the existing code structure and requirements of the hookify plugin. Additionally, it's essential to ensure that the new feature is properly tested and validated to avoid introducing any regressions or security vulnerabilities.

Recommendation

Apply workaround by modifying the core/rule_engine.py file to support the new action: ask type, as this will provide the desired functionality without requiring a full upgrade or rewrite of the existing codebase.