using-superpowers

To prevent AI agents from bypassing available tools, enforce mandatory skill invocation before responding to any request. This requires strict adherence to execution protocols, ensuring the agent checks for and utilizes relevant process skills rather than relying on base knowledge.

Mandatory skill invocation in prompt engineering is a protocol requiring AI assistants to check for and execute relevant skills before generating a response. It addresses tool use bypassing by prioritizing predefined workflows and knowledge integration over base model capabilities.

You enforce strict checklist management for an LLM workflow by providing structured guidance that mandates precise instruction following. This involves setting up a decision tree that identifies red flags and rationalizations, ensuring the AI executes required checklists.

AI assistants ignore available skills and answer directly due to common rationalizations for not using tools. This behavior is corrected by enforcing a protocol that requires checking and invoking any potentially relevant skill, even with a low probability of applicability.

The best way to structure AI agent tool use for complex tasks is implementing a structured decision tree for skill invocation. This framework provides clear guidance on when and how to invoke skills, ensuring precise instruction following and proper knowledge integration.