oh-my-kimi

Kimi-native multi-agent orchestration harness for the Kimi Code CLI
_{Turn one prompt into a planned, reviewable project. Parallel agents, live quality gates, DESIGN.md-aware UI, and zero-config safety hooks.}

What is oh-my-kimi? oh-my-kimi (OMK) wraps the Kimi Code CLI (Kimi K2.6) with a multi-agent orchestration layer. It spins up parallel coding teams in isolated Git worktrees, enforces lint / typecheck / test / build gates before completion, and provides a real-time terminal HUD to monitor progress.

npm install -g @oh-my-kimi/cli
omk init
omk doctor
# omk demo # Coming soon — try the examples below

✅ Stable Release — v1.1.0 is ready for daily use. Core orchestration, chat harness, and quality gates are solid. Some advanced features (parallel, goal) may still evolve.

Korean / English / Chinese / Japanese

Table of Contents

• Quick Start
• Examples
• GitHub Release Snapshot
• Repository Topics
• One Prompt to Landing Page
• Korean
• English
• Chinese
• Japanese
• Customization
• Acknowledgements

Quick Start

# 1. Install
npm install -g @oh-my-kimi/cli

# 2. Initialize a project
mkdir my-project && cd my-project
omk init

# 3. Verify your environment
omk doctor

# 4. Run a planned workflow
omk run "Build a Next.js landing page with dark mode and contact form"

Examples

Case studies with reproducible prompts, actual outputs, and honest limitations.

One-prompt landing page — Next.js + Tailwind landing page from a single sentence

Each example includes:

• Prompt — exactly what was sent
• RUN_REPORT.md — what the agents produced
• Known limitations — what broke or needed manual fix

GitHub Release Snapshot

Current GitHub-ready version: 1.1.0

What's New in v1.1.0

What's New in v0.4.0

GitHub Markdown checklist

• GitHub Actions / package version / npm / stars / forks / issues badges are visible at the top.
• Mermaid architecture diagrams render in GitHub-flavored Markdown.
• Repository topic badges below match the recommended GitHub topics.
• README logo PNG display width increased to 720 px for a stronger GitHub landing page.
• Screenshots for HUD, parallel UI, and status-line gauges are embedded with alt text.
• I18n utilities and multi-language README sections are present.
• New PNG assets (omk-hud-1.png, omk-statusline-gauge.png, readme-in.png, etc.) are included in the repo.

Repository Topics

These topics are also mirrored in package.json keywords for npm/GitHub discoverability.

Recommended GitHub topics:

kimi, kimi-cli, kimi-code, kimi-k2, ai-agent, coding-agent, multi-agent, agentic-coding, orchestration, dag, task-graph, ensemble, mcp, model-context-protocol, lsp, typescript, nodejs, cli, developer-tools, worktree

One Prompt to Landing Page

@oneprompt.mp4 prompt → COS landing page. 10–20 min. Kimi-native multi-agent CLI behind it.

Korean

✅ Stable Release v1.1.0 — Kimi Code CLI를 worktree 기반 코딩 팀으로 변환하세요. DESIGN.md 기반 UI 생성, AGENTS.md 호환성, 실시간 품질 게이트, 병렬 HUD를 제공합니다.

Features

🆕 v1.0.0 Highlights (Stable)

• omk chat — 오케스트레이션 경로, 퇴장 배너(Run ID, 재개, workers, MCP, skills), cockpit/tmux 지원이 포함된 인터랙티브 Kimi 세션
• Chat 전용 first-run star prompt (OMK_STAR_PROMPT) — cockpit 자식 프로세스 중복 제거
• 성능 — cockpit, doctor, hud, ensemble, dag, run, MCP 서버 전반에 병렬 I/O 최적화 적용
• omk cockpit — 병렬 TODO/에이전트 렌더링, git 변경사항, 히스토리를 포함한 실시간 컴팩트 대시보드
• omk hud — 목표 점수, ETA 예측, 상태 오류 복구 힌트가 포함한 풀 터미널 대시보드
• 안전 및 품질 — 엄격한 lint, typecheck, 234 tests, smoke test, 패키지 감사, 시크릿 스캔 전부 통과
• 오케스트레이션 — 재시도, 실패 시 건너뛰기, 폴백 역할, 증거 게이트, 앙상블 후보가 포함된 DAG 스케줄러
• 메모리 — 로컬 그래프 메모리(기본값), 선택적 Neo4j, 선택적 Kuzu 백엔드

🆕 v0.4.0 Highlights

• omk specify — GitHub Spec Kit 연동 (init, workflow, preset, version)
• omk dag from-spec [spec-dir] — Spec Kit tasks.md를 OMK DAG JSON으로 변환 (의존성 추론 + 역할 기반 라우팅)
• omk parallel --from-spec <dir> — Spec 기반 DAG를 병렬 실행기로 실행
• omk feature / bugfix / refactor / review — --spec-kit 지원 워크플로우 프리셋
• omk summary / omk summary-show — 실행 요약 및 report.md 생성
• omk index / omk index-show — 프로젝트 인덱싱으로 컨텍스트 축소
• omk skill pack / install / sync — 큐레이트된 Kimi 스킬 팩 관리
• omk agent — 16개 내장 역할 등록소 및 YAML 진단
• DAG evidence gates — command-pass 등 증거 기반 게이트 지원
• MCP doctor — MCP 진단 및 JSON-RPC 핸드셰이크 테스트
• 안전 기본값 — approval_policy = "auto", yolo_mode = false
• npm 10+ 지원 — doctor의 npm global bin 탐지 개선
• Smoke test 강화 — doctor.errors 검증으로 예기치 않은 실패 감지

🆕 v0.3.0 Highlights

• omk parallel <goal> (alpha) — coordinator → worker fan-out → reviewer 패턴으로 병렬 에이전트 팀 구성, 실시간 ETA 추적
• omk hud 대시보드 — System Usage / Kimi Usage 게이지, Project Status, TODO & Changed Files 사이드바를 포함한 실시간 터미널 대시보드
• TTY 인터랙티브 메뉴 — omk 단독 실행 시 HUD + @inquirer/prompts 메뉴 자동 실행
• --run-id 실행 재개 — 이전 실행 상태를 .omk/runs/에서 복원하여 장기 작업도 안전하게 이어감
• SendDMail 체크포인트 + Snippets — 리팩토링 전 D-Mail 체크포인트 저장 및 .omk/snippets/ 코드 블록 재사용
• OAuth Usage Gauges — OMK_KIMI_STATUS_GAUGES=1로 5시간/주간 할당량 시각적 게이지 활성화
• 16GB-friendly Runtime — 메모리 자동 감지 후 lite 프로파일 전환, 저사양 노트북/WSL 지원
• 역할 기반 앙상블 — coder/planner/architect/reviewer/QA/explorer 가중 후보 + 쿼럼 집계
• 로컬 그래프 메모리 — .omk/memory/graph-state.json 온톨로지 그래프 + mindmap/GraphQL-lite
• 내장 LSP — omk lsp typescript로 TypeScript language server 바로 실행
• 품질 게이트 강화 — npm run check/test/lint/build를 CI와 릴리스 체크에 연동
• 신규 PNG 에셋 — HUD, 상태줄 게이지, 대시보드 스크린샷 등 5종 추가

Install

npm install -g @oh-my-kimi/cli

Requirements: Node.js >= 20, Git, python3, Kimi CLI (v1.39.0+)

Quick Start

omk init
omk doctor
omk chat

Kimi-native context

oh-my-kimi agents use an Okabe-compatible base agent that inherits default and adds SendDMail, so D-Mail is available for checkpoint rollback and context recovery. Use it before risky refactors, long-running handoffs, or /compact; durable facts still go to project-local ontology graph memory.

Project-local graph memory

OMK stores project/session memory in .omk/memory/graph-state.json by default, decomposes notes into ontology nodes (Goal, Decision, Task, Risk, Command, File, Evidence, Concept), and exposes omk_memory_mindmap plus omk_graph_query for GraphQL-lite access. External Neo4j remains optional.

The interactive wrapper also augments Kimi’s native context: status line with a masked OAuth account plus 5-hour and weekly usage/quota. See docs/kimi-oauth-usage-status.md.

Preview

Live Cockpit (omk hud)

Live HUD (omk hud)

Kimi Status Line with Usage Gauges

OMK augments Kimi’s native context: status line with masked OAuth account + 5h/weekly quota. Set OMK_KIMI_STATUS_GAUGES=1 for visual bar gauges.

$ omk doctor
OK Node.js           v22.14.0
OK Git               2.49.0
OK Python            3.13.2
OK tmux              3.5a
OK Kimi CLI          v1.39.0
OK Scaffold          .omk/, .kimi/skills/ found

$ omk parallel "refactor auth module"
Parallel Execution
Run ID:   2025-05-01T12-34-56
Goal:     refactor auth module
Workers:  3
✔ Parallel DAG run complete

$ omk team  # Experimental — tmux layout scaffold only
Team Runtime starting...
   [architect]  Creating plan.md...
   [coder]      Implementation in progress...
   [reviewer]   Code review done
   [qa]         Tests passed

CLI Commands

Note: run, parallel, verify, summary, sync, runs, and goal are alpha features. Expect breaking changes.

Stable

Alpha

Experimental

Agent Registry

OMK ships with 16 built-in agent roles. Each role is a YAML file in .omk/agents/roles/ that extends the Okabe-compatible base and defines OMK_ROLE, excluded tools, and specialized prompts.

Stable — recommended for production DAG nodes:

Experimental — use with caution or for specialized tasks:

🏗️ 아키텍처

graph TD
    User["👤 User / omk CLI"] --> Router["🎮 Command Router"]

    Router --> Stable["🟢 init / doctor / update / menu"]
    Router --> Alpha["🟡 plan / run / parallel / goal / verify / summary / sync / runs / review"]
    Router --> Hud["📺 omk hud / cockpit"]
    Router --> Chat["💬 omk chat / lsp / design / star"]
    Router --> Experimental["🔴 team / merge / specify / agent / skill"]

    Alpha --> Orchestrator["⚡ Parallel Orchestrator"]
    Orchestrator --> Coordinator["🎛️ coordinator agent"]
    Orchestrator --> Workers["👷 worker agents\n(coder / explorer / tester)"]
    Orchestrator --> Reviewer["🔍 reviewer + quorum aggregation"]

    Orchestrator --> DAG["📊 DAG Scheduler + Task Graph"]
    DAG --> Executor["🚀 Executor + ETA Estimator"]
    DAG --> Evidence["🧪 Evidence Gate"]
    DAG --> Ensemble["🎭 Role-aware Ensemble"]

    Executor --> KimiLayer["🤖 Kimi Native Layer"]
    KimiLayer --> Wire["📡 Wire Client"]
    KimiLayer --> Runner["🏃 Kimi Runner + Bug Filter"]
    KimiLayer --> Cockpit["🖥️ Chat Cockpit / Banner"]
    KimiLayer --> StatusLine["📈 Usage / OAuth / Status Line"]

    Executor --> Memory["🧠 Memory & Context"]
    Memory --> LocalGraph["🗂️ Local Graph Memory"]
    Memory --> Neo4j["🌐 Neo4j (optional)"]
    Memory --> State["💾 Run State (.omk/runs/)"]
    Memory --> Snippets["📎 Snippets + D-Mail Checkpoints"]

    Orchestrator --> LiveUI["📺 Live HUD / Parallel Renderer"]

    KimiLayer --> Extensions["🔌 Extensions"]
    Extensions --> MCP["🔧 MCP Servers"]
    Extensions --> LSP["🔨 LSP TypeScript"]
    Extensions --> Skills["🎯 Project + Global Skills"]

    Orchestrator --> Safety["🛡️ Safety & Quality"]
    Safety --> Guard["🚧 Guard Hooks"]
    Safety --> QG["✅ Quality Gates"]
    Safety --> Secret["🔒 Secret Scan"]

🛡️ 안전

기본 훅은 파괴적 명령과 비밀 유출을 차단합니다:

• .omk/config.toml의 기본 approval policy는 auto이며, yolo_mode = false입니다.
• pre-shell-guard.sh — rm -rf /, sudo, git push --force 등 차단
• protect-secrets.sh — .env 편집 및 비밀 유출 차단
• post-format.sh — 수정된 파일 자동 포맷
• stop-verify.sh — 종료 시 최종 검증

🔌 내장 LSP

omk lsp --print-config
omk lsp --check
omk lsp typescript

omk init은 .omk/lsp.json을 생성하고, TypeScript/JavaScript 프로젝트에서 사용할 수 있는 번들 typescript-language-server 실행 경로를 제공합니다.

📄 라이선스

MIT

English

✅ Stable Release v1.1.0 — Turn Kimi Code CLI into a meme-tier multi-agent coding team. This is a Kimi-native wrapper — not a generic AI tool. DESIGN.md-aware UI, live quality gates, parallel HUD, AGENTS.md compatible.

Features

🆕 v1.0.0 Highlights (Stable)

• omk chat — Interactive Kimi session with orchestrated path, exit banner (Run ID, resume, workers, MCP, skills), and cockpit/tmux support
• Chat-dedicated first-run star prompt (OMK_STAR_PROMPT) with cockpit-child deduplication
• Performance — Parallel I/O optimization across cockpit, doctor, hud, ensemble, dag, run, and MCP server
• omk cockpit — Real-time compact dashboard with parallel TODO/agent rendering, git changes, and history
• omk hud — Full terminal dashboard with goal scoring, ETA estimation, and state-error recovery hints
• Safety & Quality — Strict lint, typecheck, 234 tests, smoke test, package audit, and secret scan all green
• Orchestration — DAG scheduler with retry, skip-on-failure, fallback roles, evidence gates, and ensemble candidates
• Memory — Local graph memory (default), optional Neo4j, and optional Kuzu backends

🆕 v0.4.0 Highlights

• omk specify — GitHub Spec Kit integration (init, workflow, preset, version)
• omk dag from-spec [spec-dir] — Convert spec Kit tasks.md into OMK DAG JSON with dependency inference and role-based routing
• omk parallel --from-spec <dir> — Load spec-based DAG and execute via existing parallel executor
• omk feature / bugfix / refactor / review — Workflow presets with --spec-kit support
• omk summary / omk summary-show — Generate summary.md + report.md for the latest run
• omk index / omk index-show — Build project index for context reduction
• omk skill pack / install / sync — Curated Kimi skill pack management
• omk agent — 16 built-in role registry and YAML diagnostics
• DAG evidence gates — command-pass and other evidence-based gate support
• MCP doctor — MCP diagnostics with executable checks and JSON-RPC handshake tests
• Safe defaults — approval_policy = "auto", yolo_mode = false
• npm 10+ support — Improved npm global bin detection in doctor
• Hardened smoke tests — Validates doctor.errors to catch unexpected failures

🆕 v0.3.0 Highlights

• omk parallel <goal> (alpha) — Run coordinator → worker fan-out → reviewer with live ETA tracking and 1.5 s UI refresh
• omk hud dashboard — Real-time terminal dashboard with System / Kimi Usage gauges, Project Status, TODO & Changed Files sidebar
• TTY interactive menu — Bare omk launches HUD + @inquirer/prompts menu for zero-config onboarding
• --run-id resume — Restore any previous run from .omk/runs/ persisted state
• SendDMail checkpoints + Snippets — Save D-Mail checkpoints before risky refactors and reuse code blocks via .omk/snippets/
• OAuth Usage Gauges — Visual bar gauges for 5 h/weekly quota via OMK_KIMI_STATUS_GAUGES=1
• 16 GB-friendly runtime — Auto-detects memory and switches to lite profile for low-spec laptops / WSL
• Role-aware ensemble — Weighted candidate scoring + quorum aggregation across coder/planner/architect/reviewer/QA/explorer
• Local graph memory — Ontology graph in .omk/memory/graph-state.json with mindmap and GraphQL-lite recall
• Built-in LSP — omk lsp typescript bundles TypeScript language server out of the box
• Quality gates wired to CI — npm run check / test / lint / build enforced in CI and release checks
• New PNG assets — 5 screenshots added: HUD, status-line gauges, dashboard, and more

Install

npm install -g @oh-my-kimi/cli

Requirements: Node.js >= 20, Git, python3, Kimi CLI (v1.39.0+)

Quick Start

omk init
omk doctor
omk chat

Preview

Live HUD (omk hud)

Kimi Status Line with Usage Gauges

OMK augments Kimi’s native context: status line with masked OAuth account + 5h/weekly quota. Set OMK_KIMI_STATUS_GAUGES=1 for visual bar gauges.

$ omk doctor
OK Node.js           v22.14.0
OK Git               2.49.0
OK Python            3.13.2
OK tmux              3.5a
OK Kimi CLI          v1.39.0
OK Scaffold          .omk/, .kimi/skills/ found

$ omk parallel "refactor auth module"
Parallel Execution
Run ID:   2025-05-01T12-34-56
Goal:     refactor auth module
Workers:  3
✔ Parallel DAG run complete

$ omk team
Team Runtime starting...
   [architect]  Creating plan.md...
   [coder]      Implementation in progress...
   [reviewer]   Code review done
   [qa]         Tests passed

CLI Commands

Note: run, parallel, verify, summary, sync, runs, and goal are alpha features. Expect breaking changes.

Stable

Alpha

Experimental

🏗️ Architecture

graph TD
    User["👤 User / omk CLI"] --> Router["🎮 Command Router"]

    Router --> Stable["🟢 init / doctor / update / menu"]
    Router --> Alpha["🟡 plan / run / parallel / goal / verify / summary / sync / runs / review"]
    Router --> Hud["📺 omk hud / cockpit"]
    Router --> Chat["💬 omk chat / lsp / design / star"]
    Router --> Experimental["🔴 team / merge / specify / agent / skill"]

    Alpha --> Orchestrator["⚡ Parallel Orchestrator"]
    Orchestrator --> Coordinator["🎛️ coordinator agent"]
    Orchestrator --> Workers["👷 worker agents\n(coder / explorer / tester)"]
    Orchestrator --> Reviewer["🔍 reviewer + quorum aggregation"]

    Orchestrator --> DAG["📊 DAG Scheduler + Task Graph"]
    DAG --> Executor["🚀 Executor + ETA Estimator"]
    DAG --> Evidence["🧪 Evidence Gate"]
    DAG --> Ensemble["🎭 Role-aware Ensemble"]

    Executor --> KimiLayer["🤖 Kimi Native Layer"]
    KimiLayer --> Wire["📡 Wire Client"]
    KimiLayer --> Runner["🏃 Kimi Runner + Bug Filter"]
    KimiLayer --> Cockpit["🖥️ Chat Cockpit / Banner"]
    KimiLayer --> StatusLine["📈 Usage / OAuth / Status Line"]

    Executor --> Memory["🧠 Memory & Context"]
    Memory --> LocalGraph["🗂️ Local Graph Memory"]
    Memory --> Neo4j["🌐 Neo4j (optional)"]
    Memory --> State["💾 Run State (.omk/runs/)"]
    Memory --> Snippets["📎 Snippets + D-Mail Checkpoints"]

    Orchestrator --> LiveUI["📺 Live HUD / Parallel Renderer"]

    KimiLayer --> Extensions["🔌 Extensions"]
    Extensions --> MCP["🔧 MCP Servers"]
    Extensions --> LSP["🔨 LSP TypeScript"]
    Extensions --> Skills["🎯 Project + Global Skills"]

    Orchestrator --> Safety["🛡️ Safety & Quality"]
    Safety --> Guard["🚧 Guard Hooks"]
    Safety --> QG["✅ Quality Gates"]
    Safety --> Secret["🔒 Secret Scan"]

🛡️ Safety

Default hooks block destructive commands and secret leakage:

• pre-shell-guard.sh — Blocks rm -rf /, sudo, git push --force, etc.
• protect-secrets.sh — Blocks .env edits and secret leakage
• post-format.sh — Auto-formats modified files
• stop-verify.sh — Final verification on stop

📄 License

MIT

Chinese

✅ Stable Release v1.1.0 — 将 Kimi Code CLI 转变为一个基于 worktree 的编码团队。支持 DESIGN.md 感知 UI 生成、AGENTS.md 兼容性、实时质量门禁以及并行 HUD。

Features

🆕 v1.0.0 更新亮点 (Stable)

• omk chat — 带编排路径、退出横幅（Run ID、恢复、workers、MCP、skills）及 cockpit/tmux 支持的交互式 Kimi 会话
• Chat 专用首次运行 star prompt (OMK_STAR_PROMPT) — cockpit 子进程去重
• 性能 — 对 cockpit、doctor、hud、ensemble、dag、run 及 MCP 服务器进行并行 I/O 优化
• omk cockpit — 实时紧凑仪表盘，支持并行 TODO/Agent 渲染、git 变更与历史记录
• omk hud — 完整终端仪表盘，支持目标评分、ETA 估算与状态错误恢复提示
• 安全与质量 — 严格的 lint、typecheck、234 项测试、smoke test、包审计、密钥扫描全部通过
• 编排 — 支持重试、失败跳过、回退角色、证据门控与候选集成的 DAG 调度器
• 记忆 — 本地图记忆（默认）、可选 Neo4j、可选 Kuzu 后端

🆕 v0.4.0 更新亮点

• omk specify — GitHub Spec Kit 连动（init、workflow、preset、version）
• omk dag from-spec [spec-dir] — 将 Spec Kit tasks.md 转换为 OMK DAG JSON（依赖推断 + 角色路由）
• omk parallel --from-spec <dir> — 通过并行执行器加载并执行 Spec 驱动 DAG
• omk feature / bugfix / refactor / review — 支持 --spec-kit 的工作流预设
• omk summary / omk summary-show — 生成执行摘要及 report.md
• omk index / omk index-show — 项目索引构建，压缩上下文
• omk skill pack / install / sync — 精选 Kimi 技能包管理
• omk agent — 16 个内置角色注册表及 YAML 诊断
• DAG evidence gates — 支持 command-pass 等证据门控
• MCP doctor — MCP 诊断及 JSON-RPC 握手测试
• 安全默认值 — approval_policy = "auto"，yolo_mode = false
• npm 10+ 支持 — 改进 doctor 的 npm global bin 探测
• Smoke test 强化 — 验证 doctor.errors 以捕获意外失败

🆕 v0.3.0 更新亮点

• omk parallel <goal> (alpha) — 协调器 → 多 Worker 分发 → Reviewer 闭环，支持实时 ETA 追踪与 1.5 秒 UI 刷新
• omk hud 仪表盘 — 实时终端仪表盘：系统/Kimi 资源 gauges、项目状态、TODO & 变更文件侧边栏
• TTY 交互式入口 — 直接执行 omk 即可唤起 HUD + 交互式菜单，零配置上
• --run-id 运行恢复 — 从 .omk/runs/ 持久化状态恢复任意历史运行
• SendDMail 检查点 + Snippets — 危险重构前保存 D-Mail 检查点，通过 .omk/snippets/ 复用代码块
• OAuth 用量可视化 — OMK_KIMI_STATUS_GAUGES=1 实时展示 API 配额与重置倒计时
• 16GB 友好运行时 — 自动检测内存并切换轻量资源画像，低内存设备流畅运行
• 角色感知型编排 — 加权候选者 + 仲裁投票机制覆盖 coder/planner/architect/reviewer/QA/explorer
• 本地图记忆 — .omk/memory/graph-state.json 本体图谱，支持 mindmap / GraphQL-lite 查询
• 内置 LSP — omk lsp typescript 开箱即用 TypeScript 语言服务
• CI 质量门禁 — npm run check/test/lint/build 全链路接入 CI 与发布检查
• 新增 PNG 截图 — HUD、状态栏 gauges、仪表盘等 5 张界面截图补充

Install

npm install -g @oh-my-kimi/cli

要求： Node.js >= 20、Git、python3、Kimi CLI (v1.39.0+)

Quick Start

omk init
omk doctor
omk chat

Preview

Live HUD (omk hud)

Kimi Status Line with Usage Gauges

OMK augments Kimi’s native context: status line with masked OAuth account + 5h/weekly quota. Set OMK_KIMI_STATUS_GAUGES=1 for visual bar gauges.

$ omk doctor
OK Node.js           v22.14.0
OK Git               2.49.0
OK Python            3.13.2
OK tmux              3.5a
OK Kimi CLI          v1.39.0
OK Scaffold          .omk/, .kimi/skills/ found

$ omk parallel "refactor auth module"
Parallel Execution
Run ID:   2025-05-01T12-34-56
Goal:     refactor auth module
Workers:  3
✔ Parallel DAG run complete

$ omk team  # Experimental — tmux layout scaffold only
Team Runtime 启动中...
   [architect]  创建 plan.md...
   [coder]      实现进行中...
   [reviewer]   代码审查完成
   [qa]         测试通过

CLI Commands

Note: run, parallel, verify, summary, sync, runs, and goal are alpha features. Expect breaking changes.

Stable

Alpha

Experimental

🏗️ 架构

graph TD
    User["👤 User / omk CLI"] --> Router["🎮 Command Router"]

    Router --> Stable["🟢 init / doctor / update / menu"]
    Router --> Alpha["🟡 plan / run / parallel / goal / verify / summary / sync / runs / review"]
    Router --> Hud["📺 omk hud / cockpit"]
    Router --> Chat["💬 omk chat / lsp / design / star"]
    Router --> Experimental["🔴 team / merge / specify / agent / skill"]

    Alpha --> Orchestrator["⚡ Parallel Orchestrator"]
    Orchestrator --> Coordinator["🎛️ coordinator agent"]
    Orchestrator --> Workers["👷 worker agents\n(coder / explorer / tester)"]
    Orchestrator --> Reviewer["🔍 reviewer + quorum aggregation"]

    Orchestrator --> DAG["📊 DAG Scheduler + Task Graph"]
    DAG --> Executor["🚀 Executor + ETA Estimator"]
    DAG --> Evidence["🧪 Evidence Gate"]
    DAG --> Ensemble["🎭 Role-aware Ensemble"]

    Executor --> KimiLayer["🤖 Kimi Native Layer"]
    KimiLayer --> Wire["📡 Wire Client"]
    KimiLayer --> Runner["🏃 Kimi Runner + Bug Filter"]
    KimiLayer --> Cockpit["🖥️ Chat Cockpit / Banner"]
    KimiLayer --> StatusLine["📈 Usage / OAuth / Status Line"]

    Executor --> Memory["🧠 Memory & Context"]
    Memory --> LocalGraph["🗂️ Local Graph Memory"]
    Memory --> Neo4j["🌐 Neo4j (optional)"]
    Memory --> State["💾 Run State (.omk/runs/)"]
    Memory --> Snippets["📎 Snippets + D-Mail Checkpoints"]

    Orchestrator --> LiveUI["📺 Live HUD / Parallel Renderer"]

    KimiLayer --> Extensions["🔌 Extensions"]
    Extensions --> MCP["🔧 MCP Servers"]
    Extensions --> LSP["🔨 LSP TypeScript"]
    Extensions --> Skills["🎯 Project + Global Skills"]

    Orchestrator --> Safety["🛡️ Safety & Quality"]
    Safety --> Guard["🚧 Guard Hooks"]
    Safety --> QG["✅ Quality Gates"]
    Safety --> Secret["🔒 Secret Scan"]

🛡️ 安全

默认钩子阻止破坏性命令和密钥泄漏：

• pre-shell-guard.sh — 阻止 rm -rf /、sudo、git push --force 等
• protect-secrets.sh — 阻止 .env 编辑及密钥泄漏
• post-format.sh — 自动格式化修改的文件
• stop-verify.sh — 停止时的最终验证

📄 许可证

MIT

Japanese

✅ Stable Release v1.1.0 — Kimi Code CLI を worktree ベースのコーディングチームに変換します。DESIGN.md 対応の UI 生成、AGENTS.md 互換性、ライブ品質ゲート、並列 HUD を提供します。

Features

🆕 v1.0.0 の主な更新 (Stable)

• omk chat — オーケストレーションパス、退出バナー（Run ID、再開、workers、MCP、skills）、cockpit/tmux 対応のインタラクティブ Kimi セッション
• Chat 専用 first-run star prompt (OMK_STAR_PROMPT) — cockpit 子プロセスの重複排除
• パフォーマンス — cockpit、doctor、hud、ensemble、dag、run、MCP サーバー全体に並列 I/O 最適化を適用
• omk cockpit — 並列 TODO/エージェントレンダリング、git 変更履歴、ヒストリーを含むリアルタイムコンパクトダッシュボード
• omk hud — 目標スコアリング、ETA 推定、状態エラー回復ヒントを含むフルターミナルダッシュボード
• 安全性と品質 — 厳格な lint、typecheck、234 テスト、smoke test、パッケージ監査、シークレットスキャンすべて合格
• オーケストレーション — リトライ、失敗時スキップ、フォールバックロール、エビデンスゲート、アンサンブル候補を備えた DAG スケジューラ
• メモリ — ローカルグラフメモリ（デフォルト）、オプションの Neo4j、オプションの Kuzu バックエンド

🆕 v0.4.0 の主な更新

• omk specify — GitHub Spec Kit 連携（init、workflow、preset、version）
• omk dag from-spec [spec-dir] — Spec Kit tasks.md を OMK DAG JSON に変換（依存推論 + ロールルーティング）
• omk parallel --from-spec <dir> — Spec ベース DAG を並列実行エンジンで実行
• omk feature / bugfix / refactor / review — --spec-kit 対応ワークフロープリセット
• omk summary / omk summary-show — 実行サマリーと report.md の生成
• omk index / omk index-show — プロジェクトインデックス構築でコンテキスト削減
• omk skill pack / install / sync — 厳選された Kimi スキルパック管理
• omk agent — 16 種類のビルトインロールレジストリと YAML 診断
• DAG evidence gates — command-pass などエビデンスベースゲート対応
• MCP doctor — MCP 診断と JSON-RPC ハンドシェイクテスト
• 安全なデフォルト — approval_policy = "auto"、yolo_mode = false
• npm 10+ 対応 — doctor の npm global bin 検出を改善
• Smoke test 強化 — doctor.errors の検証で予期しない失敗を検知

🆕 v0.3.0 の主な更新

• omk parallel <goal> (alpha) — コーディネーター → Worker 分散 → Reviewer 集約。リアルタイム ETA 追跡と 1.5 秒間隔の UI 更新
• omk hud ダッシュボード — システム／Kimi のメーター、プロジェクト状態、TODO & 変更ファイルサイドバーを含むリアルタイムターミナルダッシュボード
• TTY インタラクティブメニュー — omk 単体実行で HUD + 対話型プロンプトを自動起動。設定不要ですぐに使える
• --run-id 実行再開 — .omk/runs/ の永続化状態から任意の過去実行を再開
• SendDMail チェックポイント + Snippets — 危険なリファクタ前に D-Mail チェックポイントを保存し、.omk/snippets/ でコードブロックを再利用
• OAuth 使用量ゲージ — OMK_KIMI_STATUS_GAUGES=1 で API クォータとリセット残時間をステータスバーにリアルタイム表示
• 16GB メモリ対応ランタイム — 搭載メモリを自動検出し軽量プロファイルに切り替え。低スペック環境でも快適に動作
• 役割認識型アンサンブル — 重み付き候補＋クォーラム投票を coder/planner/architect/reviewer/QA/explorer で実現
• ローカルグラフメモリ — .omk/memory/graph-state.json オントロジーで知識を構造化。mindmap / GraphQL-lite 対応
• 組み込み LSP — omk lsp typescript で TypeScript 言語サーバーを即座に利用可能
• CI 品質ゲート — npm run check/test/lint/build を CI とリリースチェックに統合
• 新規スクリーンショット — HUD、ステータスバー gauges、ダッシュボードなど 5 点の UI スクリーンショットを追加

Install

npm install -g @oh-my-kimi/cli

要件: Node.js >= 20、Git、python3、Kimi CLI (v1.39.0+)

Quick Start

omk init
omk doctor
omk chat

Preview

Live HUD (omk hud)

Kimi Status Line with Usage Gauges

OMK augments Kimi’s native context: status line with masked OAuth account + 5h/weekly quota. Set OMK_KIMI_STATUS_GAUGES=1 for visual bar gauges.

$ omk doctor
OK Node.js           v22.14.0
OK Git               2.49.0
OK Python            3.13.2
OK tmux              3.5a
OK Kimi CLI          v1.39.0
OK Scaffold          .omk/, .kimi/skills/ found

$ omk parallel "refactor auth module"
Parallel Execution
Run ID:   2025-05-01T12-34-56
Goal:     refactor auth module
Workers:  3
✔ Parallel DAG run complete

$ omk team  # Experimental — tmux layout scaffold only
Team Runtime 開始中...
   [architect]  plan.md 作成中...
   [coder]      実装進行中...
   [reviewer]   コードレビュー完了
   [qa]         テスト通過

CLI Commands

Note: run, parallel, verify, summary, sync, runs, and goal are alpha features. Expect breaking changes.

Stable

Alpha

Experimental

🏗️ アーキテクチャ

graph TD
    User["👤 User / omk CLI"] --> Router["🎮 Command Router"]

    Router --> Stable["🟢 init / doctor / update / menu"]
    Router --> Alpha["🟡 plan / run / parallel / goal / verify / summary / sync / runs / review"]
    Router --> Hud["📺 omk hud / cockpit"]
    Router --> Chat["💬 omk chat / lsp / design / star"]
    Router --> Experimental["🔴 team / merge / specify / agent / skill"]

    Alpha --> Orchestrator["⚡ Parallel Orchestrator"]
    Orchestrator --> Coordinator["🎛️ coordinator agent"]
    Orchestrator --> Workers["👷 worker agents\n(coder / explorer / tester)"]
    Orchestrator --> Reviewer["🔍 reviewer + quorum aggregation"]

    Orchestrator --> DAG["📊 DAG Scheduler + Task Graph"]
    DAG --> Executor["🚀 Executor + ETA Estimator"]
    DAG --> Evidence["🧪 Evidence Gate"]
    DAG --> Ensemble["🎭 Role-aware Ensemble"]

    Executor --> KimiLayer["🤖 Kimi Native Layer"]
    KimiLayer --> Wire["📡 Wire Client"]
    KimiLayer --> Runner["🏃 Kimi Runner + Bug Filter"]
    KimiLayer --> Cockpit["🖥️ Chat Cockpit / Banner"]
    KimiLayer --> StatusLine["📈 Usage / OAuth / Status Line"]

    Executor --> Memory["🧠 Memory & Context"]
    Memory --> LocalGraph["🗂️ Local Graph Memory"]
    Memory --> Neo4j["🌐 Neo4j (optional)"]
    Memory --> State["💾 Run State (.omk/runs/)"]
    Memory --> Snippets["📎 Snippets + D-Mail Checkpoints"]

    Orchestrator --> LiveUI["📺 Live HUD / Parallel Renderer"]

    KimiLayer --> Extensions["🔌 Extensions"]
    Extensions --> MCP["🔧 MCP Servers"]
    Extensions --> LSP["🔨 LSP TypeScript"]
    Extensions --> Skills["🎯 Project + Global Skills"]

    Orchestrator --> Safety["🛡️ Safety & Quality"]
    Safety --> Guard["🚧 Guard Hooks"]
    Safety --> QG["✅ Quality Gates"]
    Safety --> Secret["🔒 Secret Scan"]

🛡️ セーフティ

デフォルトのフックは破壊的コマンドとシークレットの漏洩をブロックします：

• pre-shell-guard.sh — rm -rf /、sudo、git push --force などをブロック
• protect-secrets.sh — .env の編集とシークレットの漏洩をブロック
• post-format.sh — 変更ファイルの自動フォーマット
• stop-verify.sh — 停止時の最終検証

📄 ライセンス

MIT

Customization

⭐ First-run GitHub Star Prompt

After your first successful interactive OMK CLI command (init, doctor, hud, plan, parallel (alpha), run (alpha)), OMK may ask: "Support oh-my-kimi on GitHub?"

• Only appears in interactive TTY sessions; never in CI, non-TTY, chat, lsp, --help, or --version.
• Yes → explicitly requests a star via authenticated gh CLI.
• No → records preference and never asks again.
• Disable entirely: OMK_STAR_PROMPT=0.
• Manual retry: omk star. Check state: omk star --status.
• No GitHub tokens are stored. Browser automation is not used.

📊 Command Maturity

OMK uses a three-tier maturity model defined in MATURITY.md:

• Stable — init, doctor, chat, hud, lsp
• Alpha — parallel, run, review --ci, summary, verify, goal, sync
• Experimental — team, merge, specify, dag from-spec, skill, agent, workflow presets (feature / bugfix / refactor)

--json output is supported on doctor and verify (alpha) for CI consumption.

🎨 Custom Welcome Banner Image

You can override Kimi CLI's default ASCII banner with your own image:

1. Place your image in the project root (e.g. kimichan.png).
2. Add to .omk/config.toml:

[theme]
logo_image = "kimichan.png"

• Supports PNG, JPEG, and GIF.
• Relative paths are resolved from the project root; absolute paths (including Windows C:\... or M:\...) work too.
• In iTerm, Kitty, WezTerm, and other graphics-capable terminals, the image renders in full resolution.
• In standard terminals, it falls back to high-quality ANSI block art via terminal-image.
• If the image is missing or rendering fails, the built-in ASCII art is used automatically.

Acknowledgements

This project stands on the shoulders of giants. Every line of code here is possible because of the relentless dedication of open-source contributors around the world. With deepest respect and gratitude:

Core Platform & AI

• Kimi Code CLI by Moonshot AI — The foundation that makes everything possible. Without Kimi K2.6 and its brilliant native agent runtime, oh-my-kimi would not exist. Thank you for pushing the boundary of AI-native coding.
• Google DESIGN.md — For establishing a design-specification standard that bridges the gap between design intent and generated UI. A north star for structured frontend workflows.

Language & Runtime

• TypeScript by Microsoft — For bringing sanity, safety, and stellar IDE experience to JavaScript at scale. The type system is the unsung hero of every refactor in this codebase.
• Node.js by the OpenJS Foundation — For the runtime that powers CLI tools, async I/O, and the entire npm ecosystem. Still the most versatile server-side JavaScript runtime on the planet.

CLI & Developer Experience

• Commander.js by TJ Holowaychuk and contributors — The gold standard for building elegant, self-documenting command-line interfaces in Node.js.
• @inquirer/prompts by Simon Boudrias — For beautiful, accessible, and keyboard-friendly interactive prompts. The TTY menu experience in omk is built on this.
• execa by Sindre Sorhus — For making child-process execution predictable, promise-friendly, and cross-platform. Every shell-out in OMK goes through this.
• tsx by Anthony Fu — For zero-config TypeScript execution during development. npm run dev simply works, and that magic matters.

Data Validation & Parsing

• Zod by Colin McDonnell — For TypeScript-first schema validation that feels like part of the language. Runtime safety without sacrificing developer ergonomics.
• yaml by Eemeli Aro — For robust YAML parsing and stringifying. Agent configs, memory files, and CI definitions all rely on this.

Filesystem & Terminal

• fs-extra by JP Richardson — For the filesystem utilities Node.js should have had from day one. Copy, move, ensureDir — all battle-tested.
• terminal-image by Sindre Sorhus — For rendering images inside terminal emulators. The custom banner feature in OMK owes its magic to this.
• node-pty by Microsoft — For pseudo-terminal bindings that make interactive shell sessions feel native inside Node.js.
• tmux by Nicholas Marriott and contributors — The timeless terminal multiplexer. Team-runtime window layouts and long-lived agent sessions wouldn't be the same without it.

Language Server & Graph Database

• typescript-language-server by TypeFox — For bundling a standards-compliant LSP that gives coding agents the same intelligence as VS Code.
• neo4j-driver by Neo4j, Inc. — For making graph-database connectivity a first-class citizen in JavaScript. Optional in OMK, but powerful when enabled.

Version Control & Collaboration

• Git by Linus Torvalds and the Git community — For the distributed version control system that enables worktrees, branches, and every merge strategy in OMK.
• GitHub — For the platform that hosts this project, runs our CI, and connects maintainers with contributors across the globe.

Inspiration & Community

• OpenCode and the broader agentic-coding community — For proving that AI-native development workflows are not just possible, but inevitable. Your early experiments with autonomous coding agents lit the path.
• Creators of oh-my-opencode, oh-my-claude, and oh-my-codex — For showing that every major AI coding assistant deserves its own ergonomic harness. Your pioneering work inspired the architecture and philosophy behind oh-my-kimi.
• The Kimi engineering team at Moonshot AI — For building not just a model, but a complete native agent runtime with Okabe context management, D-Mail checkpoints, SendDMail recovery, and subagent orchestration. You redefined what a coding assistant can be.

"오픈소스는 코드가 아니라 사람들의 연대입니다. — Open source is not code; it is solidarity among people."