project-planner
npx skills add https://github.com/petestewart/claude-skills --skill project-planner
Agent 安装分布
Skill 文档
Project Planner Skill
Purpose
Generate a comprehensive PLAN.md file at the start of any new project. This plan serves as the single source of truth that an Orchestrator agent (or human) can use to drive the entire build.
When to Use
- Starting a new project from scratch
- When the user says “plan this project”, “create a plan”, or “help me scope this”
- When a project needs structured planning before implementation
Required Inputs
Before generating the plan, gather these from the user (ask if not provided):
| Input | Description |
|---|---|
PROJECT_NAME |
Short identifier for the project |
GOAL_ONE_SENTENCE |
Single sentence describing what we’re building |
TARGET_USER_CONTEXT |
Who will use this and in what context |
HARD_CONSTRAINTS |
Non-negotiable requirements (tech stack, deadlines, etc.) |
NICE_TO_HAVE |
Optional features if time permits |
NON_GOALS |
What is explicitly out of scope |
Behavior Rules
- Only create
PLAN.md– Do not implement code or create other files - Do not describe future agents or tmux usage – Keep the plan implementation-agnostic
- Make explicit assumptions – If details are missing, document assumptions clearly
- Write for an Orchestrator – The plan should be actionable by another agent or developer
PLAN.md Structure
Use these exact section headings in this exact order:
1. Overview
Short description of what we’re building and what success looks like.
2. Non Goals
What is explicitly out of scope.
3. Assumptions
Assumptions made to proceed with planning.
4. Constraints
Technical, time, tooling, and platform constraints.
5. Architecture Sketch
- High-level architecture and data flow
- Key modules and their responsibilities
- External integrations
- Specific enough to guide implementation, but not pseudo-code heavy
6. Definition of Done
Objective criteria determining the app is complete:
- Build requirements
- Test requirements
- Run requirements
- Basic user validation
7. Task Backlog
This section is the source of truth for work.
Create 10-25 initial tickets following this exact format:
### Ticket: T### Title
- **Priority:** P0 | P1 | P2
- **Status:** Todo
- **Owner:** Unassigned
- **Scope:** [Small, bounded slice of work]
- **Acceptance Criteria:** [Testable criteria]
- **Validation Steps:** [Commands or concrete checks]
- **Notes:** [Empty initially]
Ticket Rules:
- Each ticket must be independently completable
- Scope must be small and bounded
- Acceptance criteria must be testable
- Validation steps must be commands or concrete checks
8. Open Questions
Unknowns requiring decisions, each with context.
9. Discovered Issues Log
Start with this line:
New issues must be appended here with a timestamp and brief context.
This section is append-only during development.
Output Rules
- Write only the
PLAN.mdcontent - Do not wrap in markdown fences
- Do not include commentary outside the plan
- Place the file in the repository root
Example Usage
User: “I want to build a CLI tool for managing dotfiles”
Claude should:
- Ask for any missing inputs (constraints, nice-to-haves, etc.)
- Create
PLAN.mdin the repo root with all sections populated - Stop – do not begin implementation
Validation
After creating the plan, verify:
- All 9 sections are present with correct headings
- 10-25 tickets exist in the backlog
- Each ticket has all required fields
- Definition of Done has testable criteria
- File is saved as
PLAN.mdin repo root