Agent Skills: Deep Dive

Deep Dive

A two-phase workflow for technical research and solution design. Phase 1 gathers facts without suggesting solutions. Phase 2 explores multiple approaches based on those facts.

All artifacts are written to deep-dive/{task-name}/.

Operation: deep research

Usage: deep research [task description]

Information gathering phase. Analyze the codebase without suggesting solutions.

Restrictions

PERMITTED:

• Reading files and code
• Asking clarifying questions
• Understanding code structure and architecture
• Analyzing dependencies and constraints
• Recording findings

FORBIDDEN:

• Suggestions or recommendations
• Implementation ideas
• Planning or roadmaps
• Any hint of action

Workflow

1. Clarification - Ask questions to understand the scope
2. Research - Systematically analyze relevant code
3. Documentation - Record findings to deep-dive/{task-name}/research.md
4. Completion - Summarize findings (facts only), ask what to do next

research.md Structure

# Research: {task-name}

## Scope
What was investigated and why.

## Findings

### Architecture
- Current structure and patterns observed

### Dependencies
- Libraries, services, and internal modules involved

### Constraints
- Technical limitations, compatibility requirements

### Data Flow
- How data moves through the relevant components

## Open Questions
- Unresolved items that need clarification

Operation: deep innovate

Usage: deep innovate [task description]

Creative brainstorming phase. Explore multiple approaches based on research findings.

Prerequisites

Research document must exist at deep-dive/{task-name}/research.md

Restrictions

PERMITTED:

• Discussing multiple solution ideas
• Evaluating advantages and disadvantages
• Exploring architectural alternatives
• Comparing technical strategies
• Considering trade-offs

FORBIDDEN:

• Concrete planning with specific steps
• Implementation details or pseudo-code
• Committing to a single solution
• File-by-file change specifications

Workflow

1. Review - Read research document, summarize key findings
2. Exploration - Generate 2-3 distinct solution approaches
3. Analysis - Document pros/cons, trade-offs for each
4. Documentation - Create deep-dive/{task-name}/innovate.md
5. Discussion - Present approaches, gather user feedback

innovate.md Structure

For each approach, document:

# Innovation: {task-name}

## Research Summary
Key findings from the research phase.

## Approach A: {name}

### Core Concept
Philosophy and high-level idea.

### Advantages
- ...

### Challenges / Risks
- ...

### Compatibility
How it fits with existing architecture.

### Scalability & Maintainability
Long-term considerations.

## Approach B: {name}
(same structure)

## Approach C: {name}
(same structure)

## Comparison Matrix

| Criterion        | Approach A | Approach B | Approach C |
|------------------|------------|------------|------------|
| Complexity       |            |            |            |
| Risk             |            |            |            |
| Time to deliver  |            |            |            |
| Maintainability  |            |            |            |

## Next Steps
Awaiting user decision before proceeding.