deep dive

<objective> Conduct a deep-dive investigation into $ARGUMENTS (or the current topic if no arguments provided).

Go beyond surface-level understanding. Synthesize multiple sources into comprehensive knowledge. </objective>

<intake_gate>

<context_analysis> First, analyze $ARGUMENTS to extract what's already provided:

• The topic to investigate
• Specific questions to answer
• Depth required
• How this knowledge will be used

Only ask about genuine gaps - don't re-ask what's already stated. </context_analysis>

<initial_questions> Use AskUserQuestion to ask 2-4 questions based on actual gaps:

If questions unclear:

• "What do you need to understand?" with options: How it works, When to use it, Why it exists, Limitations/gotchas, All of the above, Other

If depth unclear:

• "How deep should I go?" with options: Overview (key points only), Solid understanding (main concepts), Comprehensive (thorough coverage), Other

If focus unclear:

• "Any specific angles?" with options: Practical application, Theoretical understanding, Comparison to alternatives, Historical context, Other

If application unclear:

• "How will this be used?" with options: Inform implementation, Make architecture decision, Evaluate feasibility, General knowledge, Other

Skip questions where $ARGUMENTS already provides the answer. </initial_questions>

<decision_gate> After receiving answers, use AskUserQuestion:

Question: "Ready to start the deep dive, or would you like me to ask more questions?"

Options:

1. Start research - I have enough context
2. Ask more questions - There are details to clarify
3. Let me add context - I want to provide additional information

If "Ask more questions" → generate 2-3 contextual follow-ups, then present decision gate again If "Let me add context" → receive input, then present decision gate again If "Start research" → proceed to research </decision_gate>

</intake_gate>

<process> After intake complete:

1. Define the scope and key questions to answer
2. Gather information from multiple angles:
   • How it works (mechanics)
   • Why it exists (history, motivation)
   • How it's used (patterns, best practices)
   • Where it fails (limitations, edge cases)
   • What's next (trends, evolution)
3. Synthesize into coherent understanding
4. Identify remaining unknowns </process>

<output_format>

Deep Dive: [Topic]

Strategic Summary

[2-3 sentences: what this is, key insight, main implication for our work]

Key Questions

• [Question this research answers]
• [Question this research answers]

Overview

[2-3 paragraph synthesis of what this is and why it matters]

How It Works

[Detailed explanation of mechanics, architecture, or process]

History & Context

[Why it exists, what problem it solved, how it evolved]

Patterns & Best Practices

• [Pattern/practice 1]: [when and why]
• [Pattern/practice 2]: [when and why]
• [Pattern/practice 3]: [when and why]

Limitations & Edge Cases

• [Limitation]: [workaround or mitigation]
• [Edge case]: [how to handle]

Current State & Trends

[Where things are heading, recent developments, community direction]

Key Takeaways

1. [Most important insight]
2. [Second most important insight]
3. [Third most important insight]

Remaining Unknowns

• [Question that still needs answering]
• [Question that still needs answering]

Implementation Context

<claude_context> <application>

• when_to_use: [situations where this applies]
• when_not_to_use: [situations to avoid this]
• prerequisites: [what must be true to use this] </application>

<technical> - libraries: [relevant packages/tools] - patterns: [code patterns to follow] - gotchas: [common mistakes, edge cases] </technical> <integration> - works_with: [complementary technologies] - conflicts_with: [incompatible approaches] - alternatives: [other options to consider] </integration> </claude_context>

Next Action: Apply this knowledge to implementation, research specific aspect deeper, or run /plan/project

Sources

• [Source name]: [URL] - [date accessed]
• [Source name]: [URL] - [date accessed] </output_format>

<artifact_output> Save the research to a file:

1. Create directory structure if it doesn't exist:

   • [current-working-directory]/artifacts/research/

2. Generate filename from topic:

   • Get current date in YYYY-MM-DD format
   • Slugify the topic (lowercase, hyphens for spaces)
   • Format: YYYY-MM-DD-[topic]-deep-dive.md
   • Example: 2025-01-15-kubernetes-networking-deep-dive.md

3. Write the complete research to the file

4. Report to user: "Saved to artifacts/research/[filename]" </artifact_output>

<success_criteria>

• Answers the key questions thoroughly
• Goes beyond surface-level (not just "what" but "why" and "when")
• Identifies limitations honestly
• Synthesizes into actionable understanding
• Implementation context is specific enough for Claude to apply
• Clear about what's still unknown
• Output saved to artifacts/research/ directory </success_criteria>