Agent Instructions

Commands that change how the agent behaves during a session. Each sub-file is a self-contained directive the user can invoke to correct, redirect, or enhance the agent's approach.

Topics

• Fewer Approvals: Strategies for minimizing approval prompts by using built-in tools over shell commands and npm scripts. Use when the user says there are too many approvals.
• Follow Instructions: Force the agent to re-read instruction files, identify violations, and correct its approach. Use when the user says the agent isn't following instructions.
• Interview Me: Structured, one-question-at-a-time interview process that minimizes user effort. Use when the user asks to be interviewed about a topic.
• No Hallucination: Force the agent to verify claims via web search and cite authoritative sources. Use when the user says the agent is hallucinating or making things up.
• Prevent Mistake: Guide the agent to diagnose an error and propose an instruction edit to prevent recurrence. Use when the user points out the agent made a mistake.
• Recap Current Work: Brief summary of goals, progress, current status, and next steps. Use when the user asks for a recap or summary of the session.
• Research: Concise, factual web search findings from authoritative sources. Use when the user asks the agent to research something online.

See also

• Coding Agents: Setup, configuration, and environment for the AI agents these instructions modify.
• Prompt Engineering: Patterns for structuring prompts and workflows that guide agent behavior before a session starts.
• Personas: Expert role definitions that agents adopt to bring specialist knowledge to a task.