MCPとツール呼び出し

OrchestKit

yonatangross/orchestkit

A comprehensive AI development toolkit and Claude Code plugin providing production-ready patterns, specialized agents, and automated quality gates for full-stack development.

★ 222スター
⑂ 21フォーク数
101未解決の問題
TypeScript言語
MITライセンス
Q@project.QualityScore編集部スコア

プロジェクトのスクリーンショット

OrchestKitのスクリーンショット OrchestKitのスクリーンショット OrchestKitのスクリーンショット OrchestKitのスクリーンショット OrchestKitのスクリーンショット OrchestKitのスクリーンショット

概要

OrchestKit is a unified Claude Code plugin designed to give AI coding assistants persistent knowledge of production patterns, eliminating the need to repeatedly explain tech stacks and preferences. It bundles skills, specialized agents, and lifecycle hooks into a single installation. Skills load on-demand with zero overhead, while hooks operate automatically across 29 lifecycle events to enforce quality gates, git safety, and security protocols. The toolkit includes a setup wizard that scans a codebase, detects the tech stack, recommends appropriate skills, and configures MCP servers.

主な機能

  • 105 on-demand skills for full-stack development
  • 36 specialized AI agents for task routing
  • 218 automated hooks for quality gates and git safety
  • Personalized onboarding wizard (/ork:setup) that detects tech stack and configures MCP servers
  • Local-only event logging with no hardcoded remote telemetry endpoints
  • Strict blocklist preventing AI from writing to sensitive files (e.g., id_rsa, .pem, .env)
  • Compatibility with other AI agents via skills.sh

要件、インストール、クイックスタート

For Claude Code:
/plugin marketplace add yonatangross/orchestkit
/plugin install ork

For other agents (Cursor, Codex, OpenCode, etc.):
npx skills add yonatangross/orchestkit

使用方法

After installation, run /ork:setup to start the personalized onboarding wizard which scans your codebase, detects your tech stack, recommends skills, and configures MCP servers. Key commands include /ork:auto (route a goal to the right skill), /ork:implement (full-stack implementation with parallel agents), /ork:review-pr (PR review with parallel agents), /ork:commit (conventional commit with pre-checks), /ork:doctor (health check), and /ork:remember (save to persistent memory).

モデルの互換性とユースケース

Requires Claude Code version 2.1.220 or higher (which supports Opus 5 as the default Opus, xhigh effort, dynamic workflows, sandbox.network.strictAllowlist, native binary, and hardened Bash rules).

ライセンスとリスクに関する注意事項

MIT License. The toolkit is free and nothing is gated behind paid tiers.

Editorial verification 2026-08-02: repository URL, owner, description, license and repository statistics were reviewed. License metadata: MIT. README was fetched for the channel draft; re-check repository dependencies, releases and model terms before production use.

リリースとメンテナンス

v9.5.0 (2026-08-01): Nudge TaskCreate on 3+ step work, fix count and command drift in docs, keep user-typed fork skills interactive on CC 2.1.218+, dead E2E wiring, gitignore-aware gates. v9.4.0: Add python half of phantom-symbol gate, skill-delta ratchet gate, clear 94 dead skill preload declarations. v9.3.0: Key Tavily guidance on the tvly CLI rail, fix 5 shipped bugs, stop output-validator killing sibling SubagentStop hooks. v9.2.1/v9.2.0: Propagate INCONCLUSIVE through --changed fan-out, derive flow graph for skill reference pages, skip ASK tiers in bypassPermissions mode. v9.1.1/v9.1.0: Retro-hardening mechanisms, re-anchor visual-style rule. v9.0.2: Remove Claude PR Review workflow.

MemPalace

mempalace/mempalace

★ 58KPython

QwenPaw

agentscope-ai/qwenpaw

★ 33.9KPython

Composio

composiohq/composio

★ 29.5KTypeScript