using-superpowers

Instruction prioritization defines a strict hierarchy where user instructions override skill directives, which in turn override default system prompts. This ensures the AI agent respects user commands while maintaining the mandatory skill invocation workflow.

Yes, you can adapt skill tool usage across different AI platforms like Gemini CLI and Claude Code. The skill provides specific platform adaptation guidance to ensure consistent tool access and workflow execution across various agentic environments.

To prevent prompt injection in agentic workflows, enforce strict adherence to skill applicability checks before executing actions. This mechanism validates whether a skill truly applies to the current task, blocking malicious directives from hijacking the AI response process.

You should trigger mandatory skill checks whenever there is even a slight possibility that a skill applies to the current task. For example, if a user asks about deployment, the AI must invoke a deployment skill to fetch platform-specific instructions before attempting an answer.

A limitation of forcing AI agents to always invoke skills is the potential for processing overhead on simple queries. However, the system prioritizes user instructions over skill directives, allowing users to bypass mandatory checks when direct responses are explicitly required.