/think - Planning Orchestrator

Purpose

Orchestrate implementation planning with automated design exploration followed by SOW and Spec generation.

Template Reference

Use for structure and format guidance:

• SOW structure: @../templates/sow/workflow-improvement.md
• Spec structure: @../templates/spec/workflow-improvement.md
• Summary structure: @../templates/summary/review-summary.md

Input Resolution

Same as /sow - see /sow.md for details:

1. Explicit argument: Use directly
2. Research context: Auto-detect and use topic/findings
3. Neither: Ask user for task description

Execution Flow

Step 0: Requirements Clarification (Conditional)

Skip when: Research context exists, detailed description provided, or follow-up task.

When to ask: Vague description, missing success criteria, unknown constraints.

Questions (5-7 from categories)

Category	Question Focus
Purpose	Main goal, problem solved, who benefits
Users	Primary users (end users/internal/API)
Priority	MoSCoW (Must/Should/Could/Won't)
Success	"Done" definition, metrics
Constraints	Deadline, dependencies, restrictions
Scope	What's explicitly excluded
Edge Cases	Special scenarios to handle

Known Information Filter

Before asking, extract known info from user input and skip already-answered questions:

Input: "Add OAuth authentication for API consumers by next sprint"

Known: Purpose ✓, Users ✓, Constraints ✓
Ask: Priority, Success Criteria, Scope, Edge Cases

Save Q&A to: .claude/workspace/qa/[timestamp]-[topic].md

Step 1: Implementation Design Exploration

Step 1a: Recommended Approach with Plan Agent (Opus)

First, get a recommended approach with deep analysis:

Task({
  subagent_type: "Plan",
  model: "opus",
  description: "SOW/Spec-aligned implementation analysis",
  prompt: `
Feature: "${featureDescription}"
${qaResults ? `Q&A results:\n${qaResults}\n` : ""}

Collect with confidence markers [✓]/[→]/[?]:

1. **Current State**: Quantitative baseline (file count, lines, patterns)
2. **Problems**: Classified by confidence level
3. **Solution Design**: Recommended approach with rationale
4. **Scope**: Files to modify (2-5), dependencies, phase plan
5. **Acceptance Criteria**: Measurable criteria with verification
6. **Risks**: Classified by confirmed/potential/unknown

Output format: Tables with evidence for each item.
  `,
});

Why Opus: Deep architectural analysis and comprehensive trade-off evaluation.

Step 1b: Alternative Approaches with code-architect Agents

Then, launch 3 code-architect agents in parallel for comparison:

// Agent 1: Minimal changes approach
Task({
  subagent_type: "feature-dev:code-architect",
  description: "Minimal changes approach",
  prompt: `
Feature: "${featureDescription}"
${qaResults ? `Q&A results:\n${qaResults}\n` : ""}

Design with MINIMAL CHANGES focus:
- Smallest change, maximum reuse
- Extend existing components where possible
- Minimal refactoring required

Output: Complete blueprint with file:line references
  `,
});

// Agent 2: Clean architecture approach
Task({
  subagent_type: "feature-dev:code-architect",
  description: "Clean architecture approach",
  prompt: `
Feature: "${featureDescription}"
${qaResults ? `Q&A results:\n${qaResults}\n` : ""}

Design with CLEAN ARCHITECTURE focus:
- New abstractions and interfaces
- Clear separation of concerns
- Testable, maintainable structure

Output: Complete blueprint with file:line references
  `,
});

// Agent 3: Pragmatic balance approach
Task({
  subagent_type: "feature-dev:code-architect",
  description: "Pragmatic balance approach",
  prompt: `
Feature: "${featureDescription}"
${qaResults ? `Q&A results:\n${qaResults}\n` : ""}

Design with PRAGMATIC BALANCE focus:
- Speed + quality balance
- Good boundaries without over-engineering
- Fits existing architecture

Output: Complete blueprint with file:line references
  `,
});

After Agents Complete

1. Review Plan recommendation - Opus-powered deep analysis
2. Compare with alternatives - code-architect approaches
3. Present comparison - Show trade-offs table with Plan recommendation highlighted
4. Ask user - Which approach to proceed with

Step 2: Display Analysis Results

Show structured analysis (non-blocking), then auto-continue:

🎯 Implementation Analysis

## Current State | Problem Summary | Recommended Approach
## Implementation Scope | Key Risks

Proceeding with SOW generation...

Data maps directly to SOW sections.

Step 3-5: Generate Documents

// Step 3: SOW
SlashCommand({ command: '/sow "[task description]"' });

// Step 4: Spec (includes IDR determination)
SlashCommand({ command: "/spec" });

// Step 5: Review (optional)
Task({ subagent_type: "sow-spec-reviewer", model: "haiku" });

Step 4a: IDR Requirement Determination

During Spec generation, determine IDR requirement based on:

## IDR Determination Logic

Check these conditions:

1. **SOW exists?** → If yes, IDR required (traceability)
2. **Files to modify ≥ 3?** → If yes, IDR required (complexity)
3. **Architectural decisions?** → If yes, IDR required (documentation)
4. **New patterns introduced?** → If yes, IDR required (knowledge transfer)

If NONE of above → IDR can be skipped

Output to spec.md Section 11:
| idr_required | true/false | [specific reason] |

This determination is then used by /code, /audit, /polish, /validate.

Step 6: Generate Summary

Create concise review summary following Golden Master structure:

# Summary: [Feature Name]

## 🎯 Purpose | 📋 Change Overview | 📁 Scope

## ❓ Discussion Points | ⚠️ Risks | ✅ Key Acceptance Criteria

## 🔗 Links to SOW/Spec

Output: .claude/workspace/planning/[same-dir]/summary.md

Output

All documents saved to: .claude/workspace/planning/[timestamp]-[feature]/

├── sow.md     # Statement of Work
├── spec.md    # Specification
└── summary.md # Review Summary ← Start here

When to Use

Situation	Command
Full planning (after research)	/research → /think
Full planning (explicit)	/think "task description"
SOW only	/sow
Spec only (SOW exists)	/spec
View existing plans	/plans

Example

# After /research (recommended)
/research "user authentication options"
/think  # Auto-detects research context

# With explicit argument
/think "Add user authentication with OAuth"

Related Commands

• /sow - SOW generation only
• /spec - Spec generation only
• /plans - View planning documents
• /code - Implement based on spec