project-onboarding（项目入职）

When to Use

Invoke this skill when:

• Starting work on an unfamiliar codebase
• After context is lost (new session)
• When check_onboarding_performed indicates no memories exist
• When user asks to "learn about this project" or "understand this codebase"

Onboarding Workflow

Step 1: Check Existing Knowledge

First, check if onboarding was already performed:

List files in: .claude/context/memory/
Look for: project-structure.md, build-commands.md, test-commands.md

If memories exist, read them and skip to Step 6 (Validation).

Step 2: Project Discovery

First, classify the project:

Greenfield vs Brownfield Detection

For Brownfield Projects:

1. Respect Ignore Files: Check .gitignore and .claudeignore BEFORE scanning
2. Efficient File Triage:

- Use git ls-files to list tracked files (respects.gitignore) - For large files (>1MB): Read only head/tail (first and last 20 lines) - Skip binary files, node_modules, build artifacts

1. Infer Tech Stack: Analyze manifests before asking questions
2. Context-Aware Questions: Base questions on discovered patterns

# Efficient file listing (respects .gitignore)
git ls-files --exclude-standard -co | head -100

# For non-git projects with manual ignores
find . -type f \
  -not -path '*/node_modules/*' \
  -not -path '*/.git/*' \
  -not -path '*/dist/*' \
  -not -path '*/build/*' \
  | head -100

For Greenfield Projects:

• Create fresh context artifacts
• Use interactive-requirements-gathering skill for setup

Analyze the project root to identify:

1. Package Manager & Language:

- package.json - Node.js/JavaScript/TypeScript - pyproject.toml, requirements.txt - Python - Cargo.toml - Rust - go.mod - Go - pom.xml, build.gradle - Java - composer.json - PHP

1. Project Type:

- Frontend, Backend, Fullstack, Library, CLI, Mobile, Monorepo

1. Framework Detection:

- Parse dependencies for frameworks (React, Next.js, FastAPI, etc.)

Step 3: Build System Analysis

Identify how to build/run the project:

1. Check package.json scripts (Node.js): {"scripts": {"dev": "...", "build": "...", "start": "...", "test": "..."}}
2. Check Makefiles (Python, Go, Rust): build: test: lint:
3. Check pyproject.toml (Python): [tool.poetry.scripts] [tool.poe.tasks]
4. Document discovered commands:

- Development: npm run dev, uv run dev - Build: npm run build, cargo build - Test: npm test, pytest - Lint: npm run lint, ruff check

Step 4: Directory Structure Mapping

Map key directories:

Identify:

• Entry points (index.ts, main.py, app.py)
• Component directories
• API routes
• Database models

Step 5: Create Onboarding Memories

Save discovered information to persistent memories:

Memory: project-structure.md

# Project Structure

## Overview

- Project Type: [fullstack/backend/frontend/library]
- Primary Language: [TypeScript/Python/Go/Rust]
- Framework: [Next.js/FastAPI/Express/etc.]

## Key Directories

- Source: `src/`
- Tests: `test/`
- Config: `.claude/`

## Entry Points

- Main: `src/index.ts`
- API: `src/api/`

## Important Files

- Configuration: `package.json`, `tsconfig.json`
- Environment: `.env.example`

Memory: build-commands.md

# Build Commands

## Development

- Start dev server: `npm run dev`
- Watch mode: `npm run watch`

## Build

- Production build: `npm run build`
- Type check: `npm run typecheck`

## Clean

- Clean build: `npm run clean`

Memory: test-commands.md

# Test Commands

## Unit Tests

- Run all: `npm test`
- Watch mode: `npm test -- --watch`
- Coverage: `npm test -- --coverage`

## E2E Tests

- Run: `npm run test:e2e`

## Linting

- Lint: `npm run lint`
- Fix: `npm run lint:fix`

Step 6: Validation

Validate discovered information:

1. Test Commands (if safe):

- Run npm --version or equivalent to verify package manager - Run npm run --silent to list available scripts - Do NOT run build or test without user permission

1. Verify Paths:

- Confirm key directories exist - Verify entry points are correct

Step 7: Report Summary

Output a concise summary:

## Onboarding Complete

**Project**: [name]
**Type**: [fullstack/backend/etc.]
**Framework**: [Next.js/FastAPI/etc.]

**Quick Commands**:
- Dev: `npm run dev`
- Test: `npm test`
- Build: `npm run build`

**Key Locations**:
- Source: `src/`
- Tests: `test/`
- API: `src/api/`

**Memories Created**:
- .claude/context/memory/project-structure.md
- .claude/context/memory/build-commands.md
- .claude/context/memory/test-commands.md

Agent Actions:

1. Check for existing memories in .claude/context/memory/
2. If no memories, run project discovery
3. Analyze package.json, directory structure
4. Create memory files
5. Report summary

Output:

## Onboarding Complete

**Project**: agent-studio
**Type**: Multi-agent orchestration framework
**Framework**: Claude Code + Custom agents

**Quick Commands**:
- Validate: `node .claude/tools/cli/validate-agents.mjs`
- Test hooks: `node .claude/hooks/routing/router-enforcer.cjs` (uses `.claude/lib/routing/routing-table.cjs`)

**Key Locations**:
- Agents: `.claude/agents/`
- Skills: `.claude/skills/`
- Memory: `.claude/context/memory/`

**Memories Created**: 3 files

</usage_example>

Iron Laws

1. ALWAYS check for existing onboarding memories before running discovery — duplicate onboarding creates contradictory knowledge entries that mislead future agents.
2. NEVER assume standard conventions without verification — run commands to confirm what actually works in this specific project environment.
3. ALWAYS write discovered commands and paths to persistent memory files — session context resets, but named memory files survive indefinitely.
4. ALWAYS verify discovered commands with a safe test step before documenting — incorrect commands in onboarding memories mislead every agent that reads them.
5. NEVER complete onboarding without revisiting memories when the project evolves — stale onboarding artifacts are more dangerous than no onboarding.

Anti-Patterns

Memory Protocol (MANDATORY)

Before starting: Read .claude/context/memory/learnings.md

After completing:

• New pattern discovered -> .claude/context/memory/learnings.md
• Issue encountered -> .claude/context/memory/issues.md
• Decision made -> .claude/context/memory/decisions.md

ASSUME INTERRUPTION: If it's not in memory, it didn't happen.