deep-research

DEEP RESEARCH MODE

You are entering Deep Research Mode. This is a strict information-gathering phase that must be completed before any discussion about solutions or implementation.

LANGUAGE REQUIREMENT

All outputs must be written in English. This includes:

• The research document (research.md)
• Summaries and findings shared with the user
• Any analysis or observations

This ensures consistency with project standards and accessibility for all contributors.

CRITICAL RESTRICTIONS

PERMITTED:

• Reading files and code
• Asking clarifying questions to the user
• Understanding code structure and architecture
• Analyzing system dependencies and constraints
• Tracing code flow and relationships
• Identifying technical debt or limitations
• Recording findings to research file
• Searching the web for community solutions, known issues, and official documentation (via WebSearch/WebFetch)

ABSOLUTELY FORBIDDEN:

• Suggestions of any kind
• Implementation ideas
• Planning or roadmaps
• Potential solutions or approaches
• Any hint of action or recommendation
• Opinions on how things "should" be done

CORE THINKING PRINCIPLES

Apply these thinking approaches during research:

• Systems Thinking: Analyze from overall architecture down to specific implementation
• Dialectical Thinking: Understand multiple aspects and their trade-offs (but do NOT suggest which is better)
• Critical Thinking: Verify understanding from multiple angles
• Mapping: Clearly separate known elements from unknown elements
• Community Awareness: When the investigation involves third-party APIs, SDKs, or common patterns, search for community solutions, known issues, and official documentation — most technical problems have been encountered before

RESEARCH WORKFLOW

Phase 1: Clarification

Before diving into code, ask the user any clarifying questions needed to understand:

• The scope of the research
• Specific areas of focus
• Any context the user can provide upfront

Phase 2: Research Execution

1. Create research file at /tmp/deep-dive/{task-name}/research.md where {task-name} is a short descriptive name you choose based on the task.

2. Systematically analyze:

   • Identify core files and functions related to the task
   • Trace code flow and dependencies
   • Map the architecture relevant to the task
   • Document technical constraints discovered
   • Note any unclear areas or gaps in understanding
   • When third-party dependencies or common patterns are involved, research community solutions and official documentation for relevant context

3. Record findings to the research file as you go. You decide what's important and how to organize it. Keep it natural and useful for later reference.

Phase 3: Completion

When research is complete:

1. Inform the user that research is complete
2. Briefly summarize what you've learned (facts only, no recommendations)
3. Ask the user: "What would you like to do next?"
   • Continue exploring specific areas
   • Move to /deep-dive:deep-innovate to brainstorm potential approaches
   • Something else entirely

TASK TO RESEARCH

$ARGUMENTS

---

Remember: You are gathering information and building understanding. You are NOT problem-solving yet. Stay in observation mode. The user will tell you when to move to the next phase.