---

name: plan description: This skill should be used when the user requests help building a complex multi-phase project plan, resuming an existing plan from CURRENT_PLAN.md, or executing projects that require research, planning, building, validation, and cleanup phases. Use when users say "create a plan", "help me plan", show a CURRENT_PLAN.md file, or need structured project execution with memory optimization between phases.

Plan Skill

If you are a Plan subagent: Invoke this skill immediately using the Skill tool.

Quick Reference

Use for: Multi-phase projects with research/planning/building/validation phases Don't use for: Single-step tasks, quick fixes, minor changes

Core Commands

Copy# Initialize
python3 <skill_path>/scripts/orchestrator.py --dir . init "PROJECT_GOAL" [--phases research planning building]

# Status check
python3 <skill_path>/scripts/orchestrator.py --dir . status

# Update phase
python3 <skill_path>/scripts/orchestrator.py --dir . update <phase> completed --outputs "file1.md" "file2.md"

# Generate phase template
python3 <skill_path>/scripts/phase_templates.py <phase_name> .plans/<phase_name>/plan.md

# Cleanup
python3 <skill_path>/scripts/orchestrator.py --dir . cleanup

Workflow

Starting New Plan

1. Clarify goal - Ask questions if unclear
2. Initialize - Run init command with project goal
3. Generate templates - Create phase plan templates as needed
4. Execute phases - Follow phase template → create outputs → update status
5. Iterate - Repeat until all phases complete

Resuming Existing Plan

1. Check status - Run status command
2. Load minimal context:
   • Read CURRENT_PLAN.md
   • Read .plans/<current_phase>/plan.md
   • Read previous phase's phase_summary.md ONLY
   • Load knowledge base files on-demand
3. Continue execution - Pick up where previous instance stopped

Memory Management Rules

CRITICAL - Always follow these rules to prevent high CPU usage:

Context Loading Protocol

• ✅ DO: Load previous phase's phase_summary.md (<500 words)
• ✅ DO: Load knowledge base files only when specifically needed
• ✅ DO: Use .plans/knowledge-base/ for cross-phase references
• ❌ DON'T: Load full previous phase context
• ❌ DON'T: Read all phase directories
• ❌ DON'T: Load outputs from non-adjacent phases

Phase Output Requirements

Every phase MUST create:

1. Phase-specific outputs (findings, plans, logs, etc.)
2. phase_summary.md - Concise summary under 500 words
3. Knowledge base files for cross-phase references (if needed)

Summary Cascade Pattern

CopyResearch → phase_summary.md → Planning → phase_summary.md → Building → ...

Each phase sees only the previous summary, not full context.

Phase Quick Reference

Iteration Pattern

If phase reveals issues:

1. Document reason in phase_summary.md
2. Update CURRENT_PLAN.md with iteration note
3. Return to appropriate phase
4. Create iteration_01_plan.md
5. Resume forward progress

Limit: Max 2-3 iterations per phase

Best Practices

DO:

• Create concise summaries (<500 words)
• Store shared references in knowledge-base/
• Load summaries first, details only if needed
• Update CURRENT_PLAN.md after each phase

DON'T:

• Load full previous phase context
• Skip phase_summary.md creation
• Mix phase concerns
• Read unnecessary files

Structure Created

Copy.
├── CURRENT_PLAN.md           # Central progress tracking
└── .plans/
    ├── knowledge-base/       # Cross-phase references
    ├── research/
    │   ├── plan.md
    │   ├── research_findings.md
    │   └── phase_summary.md
    ├── planning/
    │   ├── plan.md
    │   └── phase_summary.md
    └── [other phases...]

Detailed Documentation

For comprehensive phase workflows, see references/phase_workflows.md. Load this file ONLY when you need detailed guidance for a specific phase.