本文へ移動
cccskills
無料GitHub で公開

shared-monorepo-turborepo

Turborepo, workspaces, package architecture, @repo/* naming, exports, tree-shaking

インストール方法を見る

含まれるファイル(7)

  • SKILL.md9.2 KB
  • examples/caching.md4.1 KB
  • examples/core.md3.8 KB
  • examples/packages.md8.0 KB
  • examples/workspaces.md3.6 KB
  • metadata.yaml351 B
  • reference.md11.1 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

Monorepo Orchestration with Turborepo

Quick Guide: Turborepo 2.x for monorepo orchestration. Task pipelines with dependency ordering. Local + remote caching for massive speed gains. Workspaces for package linking. Syncpack for dependency version consistency. Internal packages use @repo/* naming, explicit exports fields, and workspace:* protocol.


<critical_requirements>

CRITICAL: Before Using This Skill

All code must follow project conventions in CLAUDE.md (kebab-case, named exports, import ordering, import type, named constants)

(You MUST define task dependencies using dependsOn: ["^build"] in turbo.json to ensure topological ordering)

(You MUST declare all environment variables in the env array of turbo.json tasks for proper cache invalidation)

(You MUST set cache: false for tasks with side effects like dev servers and code generation)

(You MUST use workspace:* protocol for internal package dependencies)

(You MUST use @repo/* naming convention for ALL internal packages)

(You MUST define explicit exports field in package.json - never allow importing internal paths)

(You MUST mark React as peerDependencies NOT dependencies in component packages)

</critical_requirements>


Auto-detection: Turborepo configuration, turbo.json, monorepo setup, workspaces, Bun workspaces, syncpack, task pipelines, @repo/* packages, package.json exports, workspace dependencies, shared configurations

When to use:

  • Configuring Turborepo task pipeline and caching strategies
  • Setting up workspaces for monorepo package linking
  • Enabling remote caching for team/CI cache sharing
  • Synchronizing dependency versions across workspace packages
  • Creating new internal packages in packages/
  • Configuring package.json exports for tree-shaking
  • Setting up shared configuration packages (@repo/eslint-config, @repo/typescript-config)

When NOT to use:

  • Single application projects (use standard build tools directly)
  • Projects without shared packages (no monorepo benefits)
  • Very small projects where setup overhead exceeds caching benefits
  • Polyrepo architecture is preferred over monorepo
  • Projects already using Nx or Lerna (don't mix monorepo tools)
  • App-specific code that won't be shared (keep in app directory)

Key patterns covered:

  • Turborepo 2.x task pipeline (dependsOn, outputs, inputs, cache)
  • Local and remote caching strategies
  • Workspaces for package linking
  • Syncpack for dependency version consistency
  • Environment variable handling in turbo.json
  • Package structure and @repo/* naming conventions
  • package.json exports for tree-shaking
  • Named exports and barrel file patterns
  • Internal dependencies with workspace protocol

Detailed Resources:


<philosophy>

Philosophy

Turborepo is a high-performance build system designed for JavaScript/TypeScript monorepos. It provides intelligent task scheduling, caching, and remote cache sharing to dramatically reduce build times. Combined with workspaces, it enables efficient package management with automatic dependency linking.

</philosophy>
<patterns>

Core Patterns

Pattern 1: Turborepo Task Pipeline with Dependency Ordering

Define task dependencies and caching behavior in turbo.json to enable intelligent build orchestration and caching.

Key Concepts

  • dependsOn: ["^build"] - Run dependency tasks first (topological order)
  • outputs - Define what files to cache
  • inputs - Specify which files trigger cache invalidation
  • cache: false - Disable caching for tasks with side effects
  • persistent: true - Keep dev servers running

Minimal Example

{
  "tasks": {
    "build": {
      "dependsOn": ["^build"],
      "env": ["DATABASE_URL", "NODE_ENV"],
      "outputs": ["dist/**", ".next/**", "!.next/cache/**"]
    },
    "dev": { "cache": false, "persistent": true }
  }
}

Key: dependsOn: ["^build"] ensures topological execution, env declares variables for cache invalidation, cache: false for side-effect tasks.

See examples/core.md for full good/bad comparison examples.


Pattern 2: Caching Strategies

Turborepo's caching system dramatically speeds up builds by reusing previous task outputs when inputs haven't changed.

What Gets Cached

  • Build outputs (dist/, .next/, framework-specific directories)
  • Test results (when cache: true)
  • Lint results

What Doesn't Get Cached

  • Dev servers (cache: false)
  • Code generation (cache: false - generates files)
  • Tasks with side effects

Cache Invalidation Triggers

  • Source file changes
  • Dependency changes
  • Environment variable changes (when in env array)
  • Global dependencies changes (.env, tsconfig.json)

Setup: Link a Vercel account (or self-hosted cache), then set TURBO_TOKEN and TURBO_TEAM environment variables to enable remote cache sharing.

See examples/caching.md for remote caching configuration and CI integration examples.


Pattern 3: Workspaces for Package Management

Configure workspaces to enable package linking and dependency sharing across monorepo packages.

Key Concepts

  • Root package.json declares "workspaces": ["apps/*", "packages/*"]
  • Internal deps use "@repo/ui": "workspace:*" protocol for automatic linking
  • Standard structure: apps/ for deployable apps, packages/ for shared code

See examples/workspaces.md for full good/bad comparison examples and syncpack configuration.

</patterns>
<performance>

Performance Optimization

Cache Hit Metrics:

  • First build: ~45s (5 packages, no cache)
  • Cached build: ~1s (97% faster with local cache)
  • Affected build: ~12s (73% faster, only changed packages rebuild)
  • Team savings: Hours per week with remote cache enabled

Optimization Strategies:

  • Set globalDependencies for files affecting all packages (.env, tsconfig.json) to prevent unnecessary cache invalidation
  • Use inputs array to fine-tune what triggers cache invalidation for specific tasks
  • Enable remote caching to share artifacts across team and CI
  • Use --filter with affected detection (--filter=...[HEAD^]) to only run tasks for changed packages
  • Set outputs carefully to exclude cache directories (e.g., !.next/cache/**)

Force Cache Bypass:

# Ignore cache when needed
bun run build --force

# Only build affected packages
bun run build --filter=...[HEAD^1]
</performance>

<decision_framework>

Decision Framework

New code? → Shared across 2+ apps? → packages/ (else keep in app)
Monorepo? → Builds > 30s or caching matters? → Use Turborepo

For comprehensive decision trees and package creation criteria, see reference.md.

</decision_framework>


<red_flags>

RED FLAGS

High Priority Issues:

  • Missing dependsOn: ["^build"] for build tasks (breaks topological ordering)
  • Missing env array in turbo.json (causes cache misses across environments)
  • Caching dev servers or code generation (incorrect outputs reused)
  • Default exports in library packages (breaks tree-shaking)
  • Missing exports field in package.json (allows internal path imports)

Common Mistakes:

  • Hardcoded versions instead of workspace:* for internal deps
  • React in dependencies instead of peerDependencies
  • Giant barrel files re-exporting everything (negates tree-shaking)
  • Running full test suite without --filter=...[HEAD^] affected detection

Gotchas:

  • dependsOn: ["^task"] runs dependencies' tasks; dependsOn: ["task"] runs same package's task
  • --filter=...[HEAD^] requires fetch-depth: 2 in GitHub Actions
  • Exclude cache directories in outputs: !.next/cache/**

For detailed anti-patterns and checklists, see reference.md.

</red_flags>


<critical_reminders>

CRITICAL REMINDERS

All code must follow project conventions in CLAUDE.md

(You MUST define task dependencies using dependsOn: ["^build"] in turbo.json to ensure topological ordering)

(You MUST declare all environment variables in the env array of turbo.json tasks for proper cache invalidation)

(You MUST set cache: false for tasks with side effects like dev servers and code generation)

(You MUST use workspace:* protocol for internal package dependencies)

(You MUST use @repo/* naming convention for ALL internal packages)

(You MUST define explicit exports field in package.json - never allow importing internal paths)

(You MUST mark React as peerDependencies NOT dependencies in component packages)

Failure to follow these rules will cause incorrect builds, cache misses, broken dependency resolution, and tree-shaking failures.

</critical_reminders>

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Hugging Face Inference SDK patterns for TypeScript/Node.js — InferenceClient setup, chat completion, text generation, streaming, embeddings, image generation, audio transcription, translation, summarization, and Inference Endpoints

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

LiteLLM proxy server setup, TypeScript client patterns via OpenAI SDK, model routing, fallbacks, load balancing, spend tracking, virtual keys, and production deployment

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

Serverless GPU compute platform for AI model deployment — web endpoints, GPU functions, model serving, and TypeScript client patterns

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

Local LLM inference with the Ollama JavaScript client -- chat, streaming, tool calling, vision, embeddings, structured output, model management, and OpenAI-compatible endpoint

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

Replicate SDK patterns for TypeScript/Node.js -- client setup, predictions, streaming, webhooks, file handling, model versioning, deployments, and training

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

Together AI SDK patterns for TypeScript — client setup, chat completions, streaming, structured output, function calling, embeddings, image generation, fine-tuning, and OpenAI-compatible endpoints

日本語の概要は準備中です。原文の説明を表示しています。

agents-inc/skills242026年9月8日 更新

agents-inc のスキルをすべて見る

このスキルの問題を報告する