Plan Generator Skill
This skill creates standalone plan files for complex projects that enable autonomous, phased execution with real-time progress tracking.
When to Use This Skill
Use this skill when:
- •User asks to create a project plan or PLAN file
- •User wants to break down a complex project into phases and steps
- •User needs a self-contained planning document for large initiatives
Do NOT use this skill when:
- •User wants to EXECUTE an existing plan (use
plan-executorskill instead) - •User references an existing PLAN file to continue work
Plan File Naming
Plans are feature/project-specific and can be named flexibly:
- •Single feature:
PLAN-{feature}.md(e.g.,PLAN-auth.md,PLAN-api.md) - •Main project plan:
PLAN.md - •Multiple plans can coexist in one repository for different features
Always ask the user what filename they want or suggest based on the project/feature name.
Plan File Structure
When creating a plan file, follow this exact structure:
1. Instructions for Claude Code (TOP OF FILE)
This section MUST be first. Keep it concise (~15 lines):
## Instructions for Claude Code When working on this plan: 1. **Use the `plan-executor` skill** to execute this plan 2. **Work autonomously** through phases without stopping for approval (unless blocked) 3. **Document findings** in the Notes & Decisions section **Execution Mode:** [direct|worker] **Auto-continue:** [yes|no] **Commit after phase:** [yes|no] (include PLAN file: [yes|no])
2. Current Phase Indicator
Simple status at the top:
- •
**Current Phase:** Phase 1: [Phase Name] - •
**Current Phase:** COMPLETE
3. Project Overview
- •Clear description of the end goal
- •Context about what the project achieves
- •Key requirements (bulleted list)
- •Any constraints or dependencies
4. Phase Breakdown
Each phase should include:
## Phase N: [Phase Name] Brief description of what this phase accomplishes. ### Step N.1: [Step title] Detailed, actionable step description with context and requirements. ### Step N.2: [Another step title] Description for this step...
Step Format Notes:
- •Steps are markdown headings (
### Step N.X: Title), NOT checkbox list items - •Each step ID follows the format
{phase}.{step}(e.g.,0.1,1.2,2.3) - •Include inline examples within phases when helpful (e.g., config structures, code snippets)
5. Notes & Decisions Section
Major section after all phases: ## Notes & Decisions
Template structure for each phase:
### Phase N: [Phase Name] (Status) - Implementation details and approach taken - Why decisions were made - Deviations from original plan with explanations - Technical details for future reference
6. Completion Checklist
Final section with overall project readiness criteria:
## Completion Checklist Before marking this project complete: - [ ] All phases marked complete - [ ] All tests passing - [ ] Build successful - [ ] Documentation complete
Generating the Plan File
When this skill is invoked:
- •
Determine filename: Ask user for desired filename or suggest based on project (e.g., "I'll create PLAN-auth.md for this authentication feature")
- •
Ask about preferences:
a) Execution mode:
- •Direct mode: Execute steps in the main session (simpler, good for small plans)
- •Worker mode: Delegate to sub-agents (recommended for large/complex plans)
b) Auto-continue preference:
- •Yes: Automatically start the next phase without asking
- •No: Ask for confirmation before each new phase (default)
c) Commit preferences:
- •Yes, include PLAN file: Commit code changes AND updated PLAN file together
- •Yes, exclude PLAN file: Commit only code changes
- •No: User will handle commits manually
- •
Gather project details from the user if not already provided
- •
Break down the project:
- •Identify logical phases (typically 3-8 phases, including Phase 0 for setup if needed)
- •For each phase, identify concrete steps (3-10 steps per phase)
- •Consider including inline examples (configs, schemas) in relevant phases
- •
Create the plan file with:
- •Instructions section at the TOP (referencing
plan-executorskill) - •Current Phase indicator with execution preferences
- •Complete project overview
- •All phases with detailed, actionable steps
- •Empty Notes & Decisions section
- •Completion checklist
- •(JSON state will be initialized automatically by plan-executor on first run)
- •Instructions section at the TOP (referencing
- •
Make it standalone: The file should contain enough context that referencing it in a fresh session gives Claude complete understanding of the project
Key Principles
- •Instructions first: Always put Claude instructions at the top of the file
- •Reference plan-executor: The instructions section should tell Claude to use the
plan-executorskill - •Detailed steps: Each step should be actionable with clear requirements
- •Include examples: Inline examples (configs, schemas, code snippets) make steps clearer
- •Self-referential: Plan should reference "this file" not a hardcoded filename
Example Usage
Example 1: Single Feature Plan
User: "Create a plan for building a JWT authentication system with refresh tokens"
Claude should:
- •Suggest: "I'll create
PLAN-auth.mdfor this authentication feature" - •Ask about execution mode, auto-continue, and commit preferences
- •Break down into phases like:
- •Phase 0: Dependencies and project structure
- •Phase 1: JWT token generation and validation
- •Phase 2: Refresh token implementation
- •Phase 3: Middleware integration
- •Phase 4: Testing
- •Create detailed steps for each phase with inline examples (e.g., JWT payload structure)
- •Generate PLAN-auth.md with all structure including the JSON state block
- •Inform user: "I've created PLAN-auth.md. Reference it with
@PLAN-auth.mdto begin execution"
Example 2: Main Project Plan
User: "Create a plan for building a CLI tool with subcommands and configuration support"
Claude should:
- •Suggest: "I'll create
PLAN.mdfor this project" - •Follow same process as Example 1
- •User can later create additional feature-specific plans alongside (e.g., PLAN-config.md)