Plan Do Check Act

Manual Setup

If you prefer manual setup:

# 1. Create workspace directory
mkdir -p ~/pdca-workspace

# 2. Create a test plan
cd ~/pdca-workspace
manage_plan.py create "My First Plan" "Task 1" "Task 2" "Task 3"

# 3. Verify it works
manage_plan.py check plan.md

# 4. (Optional) Set up git for progress tracking
git init

Security Notes

• Workspace Scope: Only operates on plan*.md files in the configured workspace directory
• Destructive Operations: cleanup command deletes archived files older than specified days (default 7)
• User Confirmation: Archive operation prompts for confirmation if plan is not completed
• Dry Run: Use cleanup --dry-run to preview before deleting
• Autonomous Invocation: Should require user confirmation before cleanup/archive operations

Core Process

1. Plan

When receiving a user task, create plan.md:

• Break down the task into executable subtasks
• Record the session ID
• Mark priorities and dependencies

2. Do

Execute tasks according to plan.md:

• Check off each completed task [x]
• Mark in-progress tasks as [>]
• Record issues during execution

3. Check

After all tasks are completed:

• Check for any missed items
• Verify output meets expectations
• Identify areas for improvement

4. Act

Based on check results:

• Fix identified issues
• Update plan.md with learnings
• Archive or iterate

plan.md Template

# Task Plan: [Task Name]

**Session:** discord:1488795270769676318
**Created:** YYYY-MM-DD HH:MM
**Status:** IN_PROGRESS | COMPLETED | PAUSED

## Current Task

> Task 3 - Implement Login API (in progress)

## Checklist

- [x] Task 1 - Database Design
- [x] Task 2 - User Model Creation
- [>] Task 3 - Implement Login API (in progress)
- [ ] Task 4 - Implement Registration API
- [ ] Task 5 - Unit Testing

## Notes

Records during execution...

## Check Results

Fill in after completion...

Trigger Scenarios

Core Logic: Step Analysis

After receiving a task, quickly analyze how many steps are needed:

1-2 steps → Execute directly, no plan

• "Rename a file"
• "Delete that test file"

3-5 steps → Optional plan (ask user)

• "Help me build a login feature"

5+ steps → Auto trigger PDCA

• "Build a complete user system with registration and login"
• "Refactor this project from React to Vue"

Step Analysis Dimensions

Keyword Triggers

Trigger regardless of step count when user explicitly requests:

• "break down", "decompose", "plan", "checklist"
• "plan", "checklist", "todo", "track"

Session Binding

plan.md Records Session ID

**Session:** discord:1488795270769676318

Categorized Display After Wake

Uncompleted plans for current session:
1. plan.md - User System Development (2/5)

Uncompleted plans for other sessions:
2. plan-blog.md - Blog Setup (1/3) ← telegram:123456

Interruption Recovery

Two Interruption Scenarios

Progress Inference (Active Check)

1. Check git status - Uncommitted changes
2. Check file modification time - Files within 10 minutes
3. Compare task description - Keyword matching
4. Identify task type - Code/Test/Deploy/Docs/General
5. Ask user confirmation - Final confirmation

Recovery Flow

Wake → on-start → Read plan.md → Check files within 10 min
                                    ↓
                    "login_api.py modified 3 min ago, continue?"

Consumption Mechanism

Status Flow

CREATED → IN_PROGRESS → PAUSED ↔ RESUMED → COMPLETED → ARCHIVED

Execution Rules

• Consume one task at a time
• Check off immediately upon completion [x]
• Skip and record if blocked

Concurrency Control

• Only one active plan at a time
• When new task arrives: if incomplete plan exists → ask "Finish previous or create new?"

Destruction Mechanism

Archive Strategy

Cleanup Commands

# Archive completed
manage_plan.py archive

# Cleanup expired files (>7 days)
manage_plan.py cleanup --days 7

# List all plans
manage_plan.py list

Usage

Agent Workflow

Trigger Conditions

Automatically trigger PDCA when user:

• Asks about incomplete tasks: "What's not done?", "What's pending?", "Any unfinished tasks?"
• Asks about progress: "How's it going?", "Progress check", "What's the status?"
• Asks about the plan: "What's the plan?", "Show me the plan"
• Submits a complex task (5+ steps)
• Explicitly requests: "break down", "create a plan", "make a checklist"

Plan Phase

1. Analyze steps with analyze_steps.py
2. If 5+ steps → create plan.md
3. Output plan content to user

Do Phase

1. Execute tasks one by one
2. After each task → update plan.md
3. Report progress to user

Example:

「✓ Task 1 complete (Database Design)
→ Starting Task 2 (User Model Creation)」

Check Phase

1. When user asks "What's not done?" → run manage_plan.py on-start
2. Output incomplete plans with progress
3. Example response:

「You have 1 incomplete plan:

**User System Development** (2/5 complete)
✓ Database Design
✓ User Model Creation
○ Implement Login API (in progress)
○ Implement Registration API
○ Unit Testing

Next task: Implement Login API」

Act Phase

1. Fix identified issues
2. Archive completed plans

Create Plan

# Simple create
manage_plan.py create "User System" "Database Design" "User Model" "Login API"

# With session
manage_plan.py create "User System" "Task 1" "Task 2" \
  --session "discord:1488795270769676318"

# Specify output file
manage_plan.py create "Blog Setup" "Buy Domain" "Deploy" \
  --output plan-blog.md

Update Status

# Complete tasks 1 and 2
manage_plan.py update plan.md --completed 0,1

# Mark current task (in progress)
manage_plan.py update plan.md --current 2

Pause/Resume

# Pause
manage_plan.py pause plan.md --reason "Waiting for API response"

# Resume
manage_plan.py resume plan.md

Session Start Check

# Check incomplete plans
manage_plan.py on-start . "discord:1488795270769676318"

Progress Check

# Check current progress
manage_plan.py check plan.md

# Detailed progress check (with git and files)
python3 scripts/check_progress.py . plan.md

Scripts

Configuration

Default workspace: ~/.openclaw/workspace/ Plan files: plan.md or plan-*.md Archive directory: archive/

Security:

• Only operates on plan*.md files in the workspace
• Does not modify non-plan files
• Archive/cleanup only affects archive/ directory
• Use --dry-run to preview cleanup before deleting