/plan

Thorough planning with research. Uses multiple researcher agents for comprehensive analysis.

Syntax

/plan:hard [task]

How It Works

  1. Pre-Creation Check: Checks for active plan in .claude/active-plan
  2. Research Phase: Multiple researcher agents (max 2) research in parallel
  3. Codebase Analysis: Reads docs; uses /scout if codebase-summary.md unavailable or >3 days old
  4. Gather & Plan: Main agent gathers research/scout reports, passes to planner subagent
  5. User Review: Asks for approval

Output Structure

plans/
└── YYYYMMDD-HHmm-plan-name/
    ├── research/
    │   └── researcher-XX-report.md  # Research findings
    ├── reports/
    │   └── XX-report.md
    ├── scout/
    │   └── scout-XX-report.md
    ├── plan.md                      # Overview (<80 lines)
    └── phase-XX-name.md             # Phase details

Research Limits

  • Max 2 researcher agents in parallel
  • Max 5 tool calls per researcher
  • Reports ≤150 lines each

When to Use

  • Unfamiliar technology/library
  • Complex architectural decisions
  • Need best practices research
  • External integrations

Comparison

CommandResearchSpeed
/planFullMedium
/plan:fastNoneFast
/plan:hardDeepSlow

Important: Does NOT start implementation. Use /code after approval.

Key Takeaway: Use /plan:hard when you need comprehensive research and detailed planning for complex tasks.