project-init（项目初始化）

Project Init

Overview

This skill transforms a project idea into a fully scoped, documented project blueprint through structured discovery, research, and user collaboration. The output is a project directory containing comprehensive documentation that serves as the source of truth for implementation.

Workflow

The workflow has two major stages: Discovery (understanding what to build) and Initialization (creating the project documentation). Complete Discovery fully before moving to Initialization.

Stage 1: Discovery

Step 1: Gather Project Details

Begin by understanding the project at a high level. Ask the user to describe:

• What the project is (purpose, problem it solves, target users)
• Core features and functionality
• Tech stack preferences (languages, frameworks, databases, infrastructure)
• Project-type-specific details:

- Web apps: pages/routes needed, authentication, data models - CLI tools: example commands/usage, input/output formats - APIs: endpoints, data schemas, integrations - Mobile apps: screens, navigation flow, platform targets - Libraries/SDKs: public API surface, target consumers - Games: mechanics, platforms, networking model

Use the AskUserQuestion tool to gather this information. Start with broad questions, then drill into specifics based on answers. Do not overwhelm — ask 1-3 focused questions per round.

Step 2: Research

Conduct online research using the WebSearch tool to investigate:

• Similar projects — find existing implementations, open-source examples, and established patterns for this type of project
• Technical best practices — architecture patterns, library choices, common pitfalls for the chosen tech stack
• Trade-offs — evaluate alternative approaches discovered during research
• Implementation details — specific APIs, protocols, or techniques the project will need

Synthesize findings internally. Identify anything the user may have missed, potential risks, and areas needing clarification.

Step 3: Self-Reflection

Before proceeding to the interview, pause and reason through:

• Connections between user requirements and research findings
• Gaps in the current understanding
• Architectural implications of the stated requirements
• Potential scope risks or technical challenges
• Questions where the answer is NOT obvious from prior context

Step 4: Interview

Use the AskUserQuestion tool to clarify anything unresolved from Steps 1-3. Topics may include:

• Technical implementation choices surfaced by research
• UI/UX decisions (for visual projects)
• Concerns or trade-offs discovered during research
• Feature prioritization (MVP vs. nice-to-have)
• Design preferences and constraints
• Integration requirements
• Deployment and infrastructure needs

Rules for the interview:

• Do NOT ask questions whose answers are obvious from previous responses or research
• Do NOT ask about preferences already documented in references/user-preferences.md — read and apply those directly
• Continue interviewing until confident that the project is fully understood
• Each round should surface genuinely new information

Read references/user-preferences.md for standing preferences that apply to all projects (monorepo structure, Docker, Makefiles, Bash scripts). Apply these automatically without asking.

Stage 2: Initialization

Proceed only after reaching full confidence in the project understanding. Initialization creates the project directory and documentation files — nothing else.

Step 1: Create Project Directory

mkdir -p <project-name>/docs

Create only the project root and docs/ directory. Do NOT run any framework initializers, package managers, or scaffolding tools — those are roadmap tasks.

Step 2: Create SPEC.md

Write <project-name>/docs/SPEC.md containing:

• Project Overview — name, purpose, problem statement, target users
• Architecture — high-level system design, component diagram (ASCII/Mermaid), data flow
• Tech Stack — languages, frameworks, databases, infrastructure with rationale for each choice
• Project Structure — directory layout following monorepo conventions (see references/user-preferences.md)
• Core Modules — description of each major module/component, its responsibilities, and interfaces
• Data Models — key entities, schemas, relationships
• API Surface — endpoints, contracts, or public interfaces (as applicable)
• External Integrations — third-party services, APIs, dependencies
• Key Decisions — architectural decisions made during discovery with rationale

Step 3: Create STYLES.md (Conditional)

Only create this file for projects with a visual/frontend component (web apps, mobile apps, desktop apps, games with UI).

Write <project-name>/docs/STYLES.md containing a concise styling specification for a modern, professional, Apple-esque design:

• Design Philosophy — clean, minimal, purposeful, premium feel
• Color Palette — primary, secondary, accent, semantic colors (success/warning/error/info), neutral scale, dark mode variants
• Typography — font families (system stack or specific), size scale, weight scale, line heights
• Spacing & Layout — spacing scale, grid system, breakpoints, max-widths, container padding
• Component Styling — buttons, inputs, cards, modals, navigation patterns, hover/focus/active states
• Motion & Animation — transition durations, easing curves, animation principles
• Iconography — icon style, library recommendation, sizing
• Shadows & Elevation — shadow scale, usage guidelines
• Border Radius — radius scale and usage

Base recommendations on modern design trends: clean whitespace, subtle depth through shadows, refined color palettes, smooth micro-interactions.

Step 4: Create ROADMAP.md

Read references/roadmap-format.md for the canonical format, phase structure, and action item template.

Write <project-name>/docs/ROADMAP.md following these rules:

Header: # <project-name> Development Roadmap followed by ---

Phases (in order):

1. Phase 1: Project Setup — Infrastructure, tooling, project scaffolding. Must include tasks for:

- Running framework initializers (e.g., npx create-next-app@latest, go mod init) - Monorepo directory structure creation - Makefile with standard targets (dev, run, test, check, plus project-specific targets) - Docker and docker-compose setup - CI/CD pipeline configuration - Linting, formatting, pre-commit hooks - For frontend apps: global styles setup, shared component library scaffold, page stubs with "X Page Coming Soon" text - Environment configuration templates

1. Phase 2: Research (optional) — Only include for complex projects requiring deep technical investigation before implementation. Each task produces a deliverable in ./research/. Skip for straightforward projects.
2. Phase 3: MVP (or Phase 2 if Research is skipped) — Core features required for initial launch. Break into logical subsections with numbered action items.
3. Phase 4: Nice to Have (or Phase 3) — Features, polish, and production hardening that elevate the project beyond MVP.
4. Phase 5: Future (or Phase 4) — High-level directions, not scoped tasks. Uses the Description/Features/Rationale format instead of Description/Requirements/Implementation Notes.

Action item format for Phases 1-4: Each item gets exactly 3 subsections:

• Description: What and why
• Requirements: - [] checklist of concrete deliverables
• Implementation Notes: Technical guidance, references, trade-offs

Action item format for Future phase: Each item gets exactly 3 subsections:

• Description: What this direction is about
• Features: Potential features and capabilities
• Rationale: Why this direction is worth pursuing

Number action items as <phase>.<item> (e.g., 1.1, 1.2, 3.1, 3.2). Give each item a descriptive title after the number (e.g., ### 1.1 Monorepo Structure and Tooling).

Action items should be granular enough to be individually actionable but broad enough to avoid excessive fragmentation. Target 3-8 action items per phase for most projects.