---

Table of Contents

• Quick Start
• Open Design localhost workflow
• OMK Ontology Graph
• 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. Optional: enable DeepSeek hybrid routing
printf '%s' "$DEEPSEEK_API_KEY" | omk deepseek api
omk deepseek doctor --soft

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

Open Design localhost workflow (OMK + Kimi)

Use this when you want OMK/Kimi to work through a local visual design surface instead of only a terminal prompt. OMK launches nexu-io/Open Design on localhost and registers an OMK CLI adapter, then you use the web UI to generate or iterate design artifacts with OMK CLI selected as the local agent.

How it works

1. OMK clones or reuses Open Design under .omk/open-design.
2. OMK installs the Open Design pnpm workspace with Corepack.
3. OMK starts the local daemon and web app:
   • Web UI: http://localhost:5175
   • Daemon: http://localhost:7457
4. In the Open Design UI, select OMK CLI if it is not selected automatically. OMK injects a local bridge that avoids the old Kimi ACP smoke-test timeout.
5. Describe the screen, landing page, deck, or prototype you want. Use DESIGN.md as the visual source of truth.
6. Bring the output back to OMK for implementation/review with /omk-flow-design-to-code, /omk-multimodal-ui-review, or the normal quality gate.

Korean quick note: “로컬호스트로 불러서 OMK가 붙는다”는 뜻은 OMK가 Open Design daemon/web을 로컬에서 띄우고, Open Design UI에서 OMK CLI를 선택해 디자인 산출물을 만들도록 연결한다는 의미입니다. 최종 코드화는 OMK의 DESIGN.md-aware 스킬과 품질 게이트가 이어받습니다.

License note

Open Design is upstream Apache-2.0 licensed (.omk/open-design/LICENSE after launch). OMK remains MIT; the omk design open-design command launches a local upstream checkout and does not relicense Open Design. Keep the upstream Apache-2.0 license/notice intact when redistributing Open Design source or modifications.

Commands

# Requirements for Open Design itself
node -v          # Open Design requires Node.js 24.x
corepack --version
git --version

# Preview the launch plan without cloning/installing/starting anything
omk design open-design --print-only

# First real launch: clone, install, start daemon + web, then open localhost
omk design open-design --open

# In Kimi chat, the slash skill does the same thing
/open-design

Useful options:

# Use different ports if 5175 / 7457 are busy
omk design open-design --web-port 5176 --daemon-port 7458 --open

# Reuse or update the checkout
omk design open-design --dir .omk/open-design --update --open

# Keep logs in the foreground while debugging startup
omk design open-design --foreground

# Check Open Design status/logs manually
cd .omk/open-design
corepack pnpm tools-dev status
corepack pnpm tools-dev check web
corepack pnpm tools-dev logs
corepack pnpm tools-dev stop web

On WSL, --open prefers wslview when installed and otherwise falls back to cmd.exe /c start, so the localhost URL opens in the Windows browser instead of failing through Linux xdg-open.

If an existing project does not have the slash skill yet, run:

omk skill sync

OMK Ontology Graph

Use this when the project feels too large to reason about from file names alone. OMK renders its project-local memory into an interactive graph so you can see which files, risks, decisions, goals, and evidence are connected before asking Kimi to change code.

What it shows

• Nodes: files, decisions, risks, goals, evidence, commands, and generated summaries.
• Edges: relationships such as PART_OF, DEPENDS_ON, TOUCHES_FILE, EVIDENCED_BY, and RELATES_TO.
• Filters: type or label filters for quickly narrowing a dense graph to the area Kimi should inspect.
• Counts: node/edge totals so you can tell whether the current memory snapshot is populated.

Commands

# Build and open the interactive graph from .omk/memory/graph-state.json
omk graph view --open

# In Kimi chat, the slash skill triggers the same inspection workflow
/graph-view

Korean quick note: 위 스크린샷처럼 “노드/엣지로 보는 OMK 기억 지도”를 띄운 뒤, Kimi에게 “이 리스크맵 기준으로 관련 파일만 고쳐줘”처럼 지시하면 불필요한 전체 레포 탐색을 줄일 수 있습니다.

Examples

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

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

Example	Prompt → Output	Status
One-prompt landing page	Next.js + Tailwind landing page from a single sentence	🎬 Video available
Neon Courier 2D	Browser 2D runner game in TypeScript	🎬 Video available
Neon Courier FPS	Three.js first-person prototype	🎬 Video available

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: 1.1.6

What's New in v1.1.6

Area	GitHub-visible update	Why it matters
Open Design	omk design open-design now patches the local Open Design checkout with an OMK CLI adapter, OMK_BIN setting, Kimicat icon, and root/deep-link route fix	Localhost design testing no longer hits the Kimi ACP 45 s smoke-test timeout; /api/test/connection returns ok through OMK in about 4 s
CLI Bridge	New omk open-design-agent connection point with fast smoke response and Kimi print-mode handoff for real prompts	Open Design has a stable local CLI endpoint that keeps OMK as the integration boundary
MCP Reliability	omk mcp test and omk-project return tool-level JSON-RPC errors instead of opaque json-rpc id 3: Internal error crashes	MCP debugging becomes actionable and safe for local/remote connection tests
Generated Hooks	omk init now ships session context, awesome-agent-skills advisory routing, precompact checkpointing, subagent completion audit, release guard, secret guard, formatter, and stop verification hooks	New projects start with stronger workflow routing and deploy-proof final reports
Skill Surface	OMK Core skill pack puts open-design first and includes graph-view plus DeepSeek controls	/skills exposes the most useful local design and graph workflows first
DeepSeek Hybrid	Goal progress and file-affecting nodes can use DeepSeek advisory while Kimi keeps write/merge authority	API keys are useful immediately without handing edits to the advisory model
README Assets	Refreshed Open Design localhost, hooks, skills, skill-pack, HUD, cockpit, and status-line screenshots under readmeasset/	GitHub/npm pages now show the current v1.1.6 product surface
Automation	omk cron — list, run, logs, enable, disable for scheduled DAG workflows, with validated job names and persisted run logs	Run repeatable DAG jobs without an external cron daemon
Timeouts	--timeout-preset for omk run / omk parallel, per-node timeoutPreset, and custom [timeouts.<name>] config	Keep quick tasks fast while allowing long-running agent work safely
Safety & Quality	omk init keeps global MCP secrets in user scope and preserves custom project .kimi/mcp.json; generated docs ignore is narrowed	Prevent accidental token leaks and avoid hiding authored documentation
Memory	Local graph memory remains default; embedded Kuzu is supported; stale Neo4j config no longer warns at startup	Cleaner ontology memory startup with no Neo4j credential noise
Memory	omk graph view renders .omk/memory/graph-state.json into an interactive HTML ontology graph, with /graph-view slash-command support	Inspect goals, decisions, risks, files, evidence, and derived relationships visually
Chat Harness	omk chat — Interactive Kimi session with orchestrated path, exit banner (Run ID, resume, workers, MCP, skills), and cockpit/tmux support	Turn Kimi CLI into a persistent, resumable chat session with full OMK context
Chat Harness	Chat-dedicated first-run star prompt (OMK_STAR_PROMPT) with cockpit-child deduplication	Polished onboarding without duplicate prompts in tmux splits
Performance	Parallel I/O optimization across cockpit, doctor, hud, ensemble, dag, run, and MCP server	Faster dashboard refresh and lower latency on every command
UI/UX	omk cockpit — Real-time compact dashboard with parallel TODO/agent rendering, git changes, and history	Monitor your multi-agent run from a tmux side panel
UI/UX	omk hud — Full terminal dashboard with goal scoring, ETA estimation, and state-error recovery hints	Understand run health at a glance and know the next action
Safety & Quality	Strict lint, typecheck, full npm test, smoke test, package audit, and secret scan gates	Production-grade reliability for daily use
Orchestration	DAG scheduler with retry, skip-on-failure, fallback roles, evidence gates, and ensemble candidates	Robust multi-agent execution with failure recovery
DeepSeek Hybrid	omk deepseek api stores the official API key locally, automatically enables hybrid routing, and uses deterministic Flash/Pro workers with Kimi as writer/merger	Add low-risk DeepSeek review/QA/advisory help without giving up Kimi authority
Memory	Local graph memory (default) and embedded Kuzu backends	Choose the right graph store for your project size

What's New in v0.4.0

Area	GitHub-visible update	Why it matters
Experimental	omk specify — GitHub Spec Kit integration (init, workflow, preset, version)	Connect structured planning to Kimi-native DAG execution
Core Engine	omk dag from-spec [dir] — Convert spec Kit tasks.md into OMK DAG with dependency inference and role-based routing	Turn written specs into executable parallel pipelines
Core Engine	omk parallel --from-spec <dir> — Load spec-based DAG and execute via existing parallel executor	Reuse the same spec for planning and execution
Core Engine	omk feature / bugfix / refactor / review workflow presets with --spec-kit	One-command entry points for common development workflows
UI/UX	omk summary / summary-show — Generate summary.md + report.md for the latest run	Understand what happened after a long agent run
UI/UX	omk index / index-show — Build project index (package manager, git status, file tree) for context reduction	Faster agent context building without manual repo exploration
Experimental	omk skill pack / install / sync — Curated Kimi skill pack management	Share and version agent skills across projects
Experimental	omk agent list / show / create / doctor — Agent registry and YAML diagnostics	Manage 16 built-in roles and validate agent definitions
Safety & Quality	Default approval_policy = "auto", yolo_mode = false	Safe-by-default for open-source users
Safety & Quality	doctor npm 10+ compatible; smoke test validates doctor.errors	First-install diagnostics work on modern Node/npm and catch real failures

v0.3.0 release notes (click to expand)

Area	GitHub-visible update	Why it matters
Core Engine	omk parallel <goal> (alpha) — coordinator → worker fan-out → reviewer with live ETA tracking	Spin up a multi-agent team from a single goal with real-time progress
Core Engine	Enhanced DAG engine with priority, cost, routing, failurePolicy, and evidence gates per node	DAG orchestration with retries, fallbacks, and I/O validation (alpha)
Core Engine	Role-aware ensemble — coder/planner/architect/reviewer/QA/explorer with weighted candidates + quorum aggregation	Improves agent-call quality while keeping max_parallel = 1 by default
Core Engine	omk run --run-id <id> (alpha) resumes persisted run state	Long-running agent tasks survive restarts and context switches
Core Engine	SendDMail checkpoint helpers + .omk/snippets/ reusable storage	Safer refactors and reusable code blocks across agent sessions
UI/UX	omk hud — live dashboard with System Usage, Kimi Usage gauges, Project Status, Latest Run, TODO & Changed Files sidebar	Real-time visibility into your agent fleet without external monitoring tools
UI/UX	Bare omk TTY entry point — HUD + interactive @inquirer/prompts menu	Zero-config entry point for new users; no more "what do I type first?"
UI/UX	ParallelLiveRenderer refreshes every 1.5 s with run state transitions	See workers start, finish, fail, and retry in real time
UI/UX	OMK_KIMI_STATUS_GAUGES=1 enables visual bar gauges for 5 h/weekly quota	Know your Kimi API budget at a glance
UI/UX	OMK_STAR_PROMPT guided GitHub star experience on first CLI use	Community growth without being intrusive; respects CI and --help
Memory & Intelligence	Local graph memory — .omk/memory/graph-state.json with ontology mindmap and GraphQL-lite recall	Local-first memory works without external database setup
Memory & Intelligence	omk lsp typescript exposes the bundled TypeScript language server	Helps coding agents and editors share the same language intelligence
Memory & Intelligence	I18n utilities added for multi-language agent workflows	Foundation for localized agent prompts and CLI output
Safety & Quality	stop-verify.sh comprehensive verification + eslint + hardened path validation	Even in yolo mode, destructive commands and credential exposure are blocked
Safety & Quality	runtime.resource_profile = "auto" selects lite profile on 16 GB machines	Keeps OMK usable on 16 GB laptops and WSL environments
Safety & Quality	npm run check, npm test, npm run lint, npm run build wired into CI	GitHub contributors can verify changes before PRs
Assets	refreshed PNG screenshots: Open Design localhost, generated hooks, generated skills, skill packs, HUD, cockpit, and status-line gauges	Rich visual documentation for the GitHub landing page

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.
• Current PNG assets (open-design-localhost.png, omk-v1.1.6-generated-hooks.png, omk-v1.1.6-generated-skills.png, HUD, cockpit, and status-line captures) are included in the repo.

README asset refresh (v1.1.6)

The README screenshots are regenerated from the current local package and Open Design localhost flow. They are intentionally static PNGs so GitHub and npm render them quickly.

Ontology Graph Viewer

See OMK Ontology Graph for the interactive memory graph workflow and screenshot.

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 v1.1.6 — Kimi Code CLI를 worktree 기반 코딩 팀으로 변환하세요. DESIGN.md 기반 UI 생성, AGENTS.md 호환성, 실시간 품질 게이트, 병렬 HUD를 제공합니다.

Features

Feature	Description
Kimi K2.6 Optimized	Kimi K2.6에 특화된 워크플로우와 컨텍스트 관리
DeepSeek Hybrid Routing	omk deepseek api로 공식 API 키를 저장하면 하이브리드 라우팅이 자동 enable 됩니다. DeepSeek Flash/Pro는 리뷰·QA·자문 역할을 맡고, 실제 작성/머지는 Kimi가 유지합니다.
Okabe + D-Mail	Kimi Code의 Okabe 스마트 컨텍스트 관리와 SendDMail 체크포인트 기본 활용
Worktree-based Parallel Team	Git worktree로 에이전트별 격리된 작업 공간 제공
DESIGN.md Integration	Google DESIGN.md 표준 기반 UI 생성
Multi-Agent Compatible	AGENTS.md / GEMINI.md / CLAUDE.md 동시 지원
Quality Gates	완료 전 자동 lint, typecheck, test, build 검증
Built-in LSP	omk lsp typescript로 번들 TypeScript language server 실행
Parallel HUD	omk hud / omk cockpit — 병렬 에이전트 실행 실시간 모니터링 (System Usage, Kimi Usage, Project Status, Latest Run, TODO / Changed Files 사이드바)
MCP Integration	다양한 MCP 서버와의 원활한 연동
Local Graph Memory	프로젝트/세션별 기억을 .omk/memory/graph-state.json 온톨로지 그래프로 저장하고 mindmap/GraphQL-lite 제공
OAuth Usage Badge	Kimi context: 상태줄 옆에 masked 계정, 5h/weekly quota 표시; OMK_KIMI_STATUS_GAUGES=1로 시각적 게이지 활성화
Approval Policy	기본값은 approval_policy = "auto" (안전 모드); 필요시 yolo로 전환 가능
Safety Hooks	yolo mode에서도 파괴적 명령어 및 비밀 유출 방지 기본 제공

🆕 v1.1.6 Highlights (Stable)

• Open Design + OMK CLI — omk design open-design --open registers OMK CLI locally and avoids the 45 s Kimi ACP smoke-test timeout
• MCP JSON-RPC stability — json-rpc id 3: Internal error paths now surface actionable tool-level errors
• Hooks/Skills refreshed — /open-design, /graph-view, awesome-agent-skills advisory routing, release guard, and stop verification are included in generated projects
• README assets refreshed — v1.1.6 package captures plus current localhost screenshots are committed under readmeasset/
• 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, 전체 npm test, smoke test, 패키지 감사, 시크릿 스캔 게이트 통과
• 오케스트레이션 — 재시도, 실패 시 건너뛰기, 폴백 역할, 증거 게이트, 앙상블 후보가 포함된 DAG 스케줄러
• DeepSeek Hybrid — 공식 API 키 입력 시 하이브리드 기능 자동 활성화; Flash/Pro 60/40 라우팅과 파일 변경 노드의 Pro 자문을 지원
• 메모리 — 로컬 그래프 메모리(기본값), 내장 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
• README 에셋 — Open Design localhost, 생성 hooks/skills, skill packs, HUD, 상태줄 게이지, 대시보드 최신화

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. Embedded Kuzu remains available for Cypher-style graph queries.

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

Command	Description
omk init	Scaffold .omk/, .kimi/skills/, .agents/skills/, docs, hooks, agents
omk doctor	Check Node, Kimi CLI, Git, python3, tmux, scaffold
omk doctor --soft	Soft mode: do not fail on missing tools — useful for smoke tests and CI
omk chat	Interactive Kimi with agent/config/MCP auto-detection
omk plan <goal>	Plan-only mode
omk hud	Live dashboard with system usage, Kimi quota, project status, run tracking
omk lsp [server]	Built-in LSP launcher; default server is TypeScript
omk star	GitHub star helper; manual retry and status check
omk design init	Create DESIGN.md with frontmatter
omk design list	List local/remote DESIGN.md files
omk design apply <name>	Convert DESIGN.md into code
omk google stitch-install	Install Google Stitch skills
omk update	Check or run OMK and Kimi CLI updates
omk menu	Interactive OMK main menu via @inquirer/prompts

Alpha

Command	Description
omk run <flow> <goal> (alpha)	Flow-based task execution
omk parallel <goal> (alpha)	Parallel DAG execution (coordinator → workers → reviewer)
omk review (alpha)	Code review + security review of current changes
omk review --ci (alpha)	CI mode: local checks only, no Kimi API calls
omk review --soft (alpha)	Soft mode: always exit 0 even if review fails
omk verify (alpha)	Evidence gate verification for completed runs
omk summary (alpha)	Run summary and report generation
omk sync (alpha)	Sync Kimi assets (hooks, MCP, skills, local graph memory)
omk sync --dry-run (alpha)	Preview sync without applying changes
omk sync --diff (alpha)	Show diff of what would change
omk sync --rollback (alpha)	Rollback last sync from manifest
omk runs (alpha)	List past OMK runs with status and dates
omk goal (alpha)	Codex-style goal management

Experimental

Command	Status	Notes
omk team	Layout only	tmux window layout scaffold
omk agent	Experimental	Agent registry and YAML diagnostics
omk skill	Experimental	Kimi skill pack manager
omk merge	Manual	Diff check + manual cherry-pick guidance
omk design lint	Stub	Validation not yet implemented
omk design diff	Stub	Diff not yet implemented
omk design export	Stub	Export not yet implemented

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:

Agent	Role	Best for
planner	Architecture / refactor planning	omk plan, omk run (alpha) plan-first flows
coder	Scoped implementation	Feature dev, bugfix, typed languages
reviewer	Adversarial code review	Pre-merge review, security audit
qa	Lint / typecheck / test / build	omk-quality-gate enforcement
security	Security review	Dependency audit, secret scan, RBAC

Experimental — use with caution or for specialized tasks:

Agent	Role	Best for
coordinator	Multi-agent coordination	omk parallel (alpha) fan-out orchestration
architect	System design	High-level module design
explorer	Repository discovery	Unfamiliar codebase mapping
tester	Test generation	Unit / integration test authoring
docs	Documentation	README, API docs, design docs
merger	Branch / PR merge	Conflict resolution, cherry-pick
release	Release flow	Version bump, changelog, tag
integrator	Cross-service integration	API glue, adapter code
interviewer	Interactive prompting	User requirement clarification
researcher	Deep research	Web search, doc reading, comparison
vision-debugger	UI / visual debugging	Screenshot analysis, CSS fixes

🏗️ 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 --> Kuzu["🧠 Kuzu (embedded)"]
    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.
• session-context.sh — 시작 시 Open Design, ontology graph, 배포 상태 확인 규칙 주입
• awesome-agent-skills-router.sh — 프롬프트를 OMK 스킬/워크플로로 advisory 라우팅
• precompact-checkpoint.sh — 컨텍스트 압축 전 목표/파일/검증/블로커 체크포인트 안내
• subagent-stop-audit.sh — 서브에이전트 종료 후 리더 검토/품질 게이트 요구
• pre-shell-guard.sh — rm -rf /, sudo, git push --force, 검증 없는 release/publish 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/배포 과대보고 방지 체크리스트

🔌 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

Open Design 연동은 upstream nexu-io/Open Design을 사용하며, 해당 프로젝트는 Apache-2.0 라이선스입니다. OMK 본체는 계속 MIT 라이선스입니다.

---

English

✅ Stable Release v1.1.6 — 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

Feature	Description
Kimi K2.6 Optimized	Workflows and context management tailored for Kimi K2.6
DeepSeek Hybrid Routing	omk deepseek api stores the official API key and automatically enables hybrid routing. DeepSeek Flash/Pro handles review, QA, and advisory work while Kimi stays responsible for writing and merge authority.
Okabe + D-Mail	Uses Kimi Code Okabe smart context management and SendDMail checkpoint recovery by default
Worktree-based Parallel Team	Git worktree provides isolated workspaces per agent
DESIGN.md Integration	UI generation based on Google DESIGN.md standard
Multi-Agent Compatible	Simultaneous support for AGENTS.md / GEMINI.md / CLAUDE.md
Quality Gates	Automated lint, typecheck, test, build verification before completion
Parallel HUD	omk hud / omk cockpit — Real-time monitoring of parallel agent execution (System Usage, Kimi Usage gauges, Project Status, Latest Run, TODO / Changed Files sidebar)
MCP Integration	Seamless connection with various MCP servers
Local Graph Memory	Stores project/session memory in .omk/memory/graph-state.json as an ontology graph with mindmap/GraphQL-lite tools
Parallel DAG	omk parallel <goal> (alpha) runs coordinator → worker fan-out → reviewer with live UI and ETA tracking
Safety Hooks	Default protection against destructive commands and secret leakage

🆕 v1.1.6 Highlights (Stable)

• Open Design + OMK CLI — omk design open-design --open registers OMK CLI locally and avoids the 45 s Kimi ACP smoke-test timeout
• MCP JSON-RPC stability — opaque internal errors now become actionable tool-level failures
• Hooks/Skills refreshed — generated projects include advisory routing, release guard, stop verification, /open-design, and /graph-view
• README assets refreshed — v1.1.6 package captures plus current localhost screenshots are committed under readmeasset/
• 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, full npm test, smoke test, package audit, and secret scan gates
• Orchestration — DAG scheduler with retry, skip-on-failure, fallback roles, evidence gates, and ensemble candidates
• DeepSeek Hybrid — Official API key setup automatically enables hybrid mode with deterministic Flash/Pro 60/40 routing and Pro advisory on file-affecting nodes
• Memory — Local graph memory (default) and embedded 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
• README assets — Open Design localhost, generated hooks/skills, skill packs, 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

Command	Description
omk init	Scaffold .omk/, .kimi/skills/, .agents/skills/, docs, hooks, agents
omk doctor	Check Node, Kimi CLI, Git, python3, tmux, scaffold
omk doctor --soft	Soft mode: do not fail on missing tools — useful for smoke tests and CI
omk chat	Interactive Kimi with agent/config/MCP auto-detection
omk plan <goal>	Plan-only mode
omk hud	Live dashboard with system usage, Kimi quota, project status, run tracking
omk lsp [server]	Built-in LSP launcher; default server is TypeScript
omk star	GitHub star helper; manual retry and status check
omk design init	Create DESIGN.md with frontmatter
omk design list	List local/remote DESIGN.md files
omk design apply <name>	Convert DESIGN.md into code
omk google stitch-install	Install Google Stitch skills

Alpha

Command	Description
omk run <flow> <goal> (alpha)	Flow-based task execution
omk parallel <goal> (alpha)	Parallel DAG execution (coordinator → workers → reviewer)
omk review (alpha)	Code review + security review of current changes
omk review --ci (alpha)	CI mode: local checks only, no Kimi API calls
omk review --soft (alpha)	Soft mode: always exit 0 even if review fails
omk verify (alpha)	Evidence gate verification for completed runs
omk summary (alpha)	Run summary and report generation
omk sync (alpha)	Sync Kimi assets (hooks, MCP, skills, local graph memory)
omk sync --dry-run (alpha)	Preview sync without applying changes
omk sync --diff (alpha)	Show diff of what would change
omk sync --rollback (alpha)	Rollback last sync from manifest
omk runs (alpha)	List past OMK runs with status and dates
omk goal (alpha)	Codex-style goal management

Experimental

Command	Status	Notes
omk team	Layout only	tmux window layout scaffold
omk agent	Experimental	Agent registry and YAML diagnostics
omk skill	Experimental	Kimi skill pack manager
omk merge	Manual	Diff check + manual cherry-pick guidance
omk design lint	Stub	Validation not yet implemented
omk design diff	Stub	Diff not yet implemented
omk design export	Stub	Export not yet implemented

🏗️ 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 --> Kuzu["🧠 Kuzu (embedded)"]
    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:

• session-context.sh — Injects startup reminders for Open Design, ontology graph, and deployment claims
• awesome-agent-skills-router.sh — Advisory-routes prompts to installed OMK skills/workflows
• precompact-checkpoint.sh — Reminds agents to checkpoint goal, files, verification, and blockers before compaction
• subagent-stop-audit.sh — Requires leader review and quality gates after subagent completion
• pre-shell-guard.sh — Blocks rm -rf /, sudo, git push --force, unverified release/publish commands, etc.
• protect-secrets.sh — Blocks .env edits and secret leakage
• post-format.sh — Auto-formats modified files
• stop-verify.sh — Final verification checklist and anti-overclaim reminder on stop

📄 License

MIT

Open Design integration uses nexu-io/Open Design, which is licensed under Apache-2.0 upstream. OMK itself remains MIT licensed.

---

Chinese

✅ Stable Release v1.1.6 — 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

Feature	Description
Kimi K2.6 优化	专为 Kimi K2.6 定制的工作流与上下文管理
Okabe + D-Mail	默认使用 Kimi Code Okabe 智能上下文管理和 SendDMail 检查点恢复
基于 Worktree 的并行团队	Git worktree 为每个 Agent 提供隔离工作空间
DESIGN.md 集成	基于 Google DESIGN.md 标准的 UI 生成
多 Agent 兼容	同时支持 AGENTS.md / GEMINI.md / CLAUDE.md
质量门禁	完成前自动执行 lint、typecheck、test、build 验证
并行 HUD	omk hud / omk cockpit — 并行 Agent 执行实时监控（系统用量、Kimi 配额条、项目状态、最新运行、TODO / 变更文件侧边栏）
MCP 集成	与多种 MCP 服务器无缝连接
Local Graph Memory	将项目/会话记忆存入 .omk/memory/graph-state.json 本地本体图，并提供 mindmap/GraphQL-lite
并行 DAG	omk parallel <goal> (alpha) 执行 coordinator → worker 扇出 → reviewer，带实时 UI 与 ETA 追踪
安全钩子	默认防止破坏性命令与密钥泄漏

🆕 v1.1.6 更新亮点 (Stable)

• Open Design + OMK CLI — localhost 设计桥接现在使用 OMK CLI，避免 45 秒 Kimi ACP smoke-test timeout
• MCP JSON-RPC 稳定性 — internal error 会以可处理的 tool-level failure 呈现
• Hooks/Skills 刷新 — 生成项目包含 release guard、stop verification、/open-design、/graph-view
• README assets 刷新 — v1.1.6 package captures 已加入 readmeasset/
• 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、完整 npm test、smoke test、包审计、密钥扫描门禁
• 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）和嵌入式 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

Command	Description
omk init	创建 .omk/、.kimi/skills/、.agents/skills/、docs、hooks、agents 脚手架
omk doctor	检查 Node、Kimi CLI、Git、python3、tmux、脚手架
omk doctor --soft	软模式：缺少工具时不失败 — 适用于 smoke 测试和 CI
omk chat	支持代理/配置/MCP 自动检测的交互式 Kimi
omk plan <goal>	仅计划模式
omk hud	实时仪表盘：系统用量、Kimi 配额、项目状态、运行追踪
omk lsp [server]	内置 LSP 启动器；默认服务器为 TypeScript
omk star	GitHub star 助手；手动重试与状态检查
omk design init	创建带 frontmatter 的 DESIGN.md
omk design list	列出本地/远程 DESIGN.md
omk design apply <name>	将 DESIGN.md 转换为代码
omk google stitch-install	安装 Google Stitch 技能
omk update	检查或运行 OMK 和 Kimi CLI 更新
omk menu	通过 @inquirer/prompts 显示交互式 OMK 主菜单

Alpha

Command	Description
omk run <flow> <goal> (alpha)	基于流程的任务执行
omk parallel <goal> (alpha)	并行 DAG 执行（coordinator → workers → reviewer）
omk review (alpha)	当前变更的代码审查 + 安全审查
omk review --ci (alpha)	CI 模式：仅本地检查，不调用 Kimi API
omk review --soft (alpha)	软模式：审查失败也返回 exit 0
omk verify (alpha)	已完成运行的证据门验证
omk summary (alpha)	运行摘要与报告生成
omk sync (alpha)	同步 Kimi 资源（hooks、MCP、skills、本地图记忆）
omk sync --dry-run (alpha)	预览变更，不实际应用
omk sync --diff (alpha)	显示将要变更的 diff
omk sync --rollback (alpha)	从 manifest 回滚上次同步
omk runs (alpha)	列出过去的 OMK 运行及其状态和日期
omk goal (alpha)	Codex 风格的目标管理

Experimental

Command	Status	Notes
omk team	仅布局	tmux 窗口布局脚手架
omk agent	实验性	Agent 注册表及 YAML 诊断
omk skill	实验性	Kimi 技能包管理器
omk merge	手动	Diff 检查 + 手动 cherry-pick 指导
omk design lint	占位	验证尚未实现
omk design diff	占位	Diff 尚未实现
omk design export	占位	导出尚未实现

🏗️ 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 --> Kuzu["🧠 Kuzu (embedded)"]
    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：

• session-context.sh — 启动时注入 Open Design、ontology graph、部署声明检查提醒
• awesome-agent-skills-router.sh — 将提示词 advisory 路由到已安装的 OMK skills/workflows
• precompact-checkpoint.sh — 压缩上下文前提醒记录目标、文件、验证和阻塞项
• subagent-stop-audit.sh — 子代理结束后要求 leader 复核并运行质量门禁
• pre-shell-guard.sh — 阻止 rm -rf /、sudo、git push --force、未验证 release/publish 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 — 停止时的最终验证清单并防止夸大部署状态

📄 许可证

MIT

Open Design 集成使用上游 nexu-io/Open Design，其许可证为 Apache-2.0；OMK 本身仍使用 MIT 许可证。

---

Japanese

✅ Stable Release v1.1.6 — 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

Feature	Description
Kimi K2.6 対応	Kimi K2.6 に特化したワークフローとコンテキスト管理
Okabe + D-Mail	Kimi Code Okabe のスマートコンテキスト管理と SendDMail チェックポイント復旧を標準利用
Worktree ベース並列チーム	Git worktree でエージェントごとに分離された作業空間を提供
DESIGN.md 連携	Google DESIGN.md 標準に基づく UI 生成
マルチエージェント互換	AGENTS.md / GEMINI.md / CLAUDE.md を同時サポート
品質ゲート	完了前に自動 lint、typecheck、test、build を検証
並列 HUD	omk hud / omk cockpit — 並列エージェント実行のリアルタイム監視（システム使用量、Kimi クォータゲージ、プロジェクト状態、最新実行、TODO/変更ファイルサイドバー）
MCP 統合	様々な MCP サーバーとのシームレスな連携
Local Graph Memory	プロジェクト/セッション記憶を .omk/memory/graph-state.json のローカル ontology graph に保存し、mindmap/GraphQL-lite を提供
並列 DAG	omk parallel <goal> (alpha) は coordinator → worker ファンアウト → reviewer を実行。ライブ UI と ETA 追跡付き
安全フック	破壊的コマンドとシークレット漏洩をデフォルトで防止

🆕 v1.1.6 の主な更新 (Stable)

• Open Design + OMK CLI — localhost デザインブリッジが OMK CLI を使用し、45 秒の Kimi ACP smoke-test timeout を回避
• MCP JSON-RPC stability — internal error を actionable な tool-level failure として表示
• Hooks/Skills refreshed — release guard、stop verification、/open-design、/graph-view を生成プロジェクトに同梱
• README assets refreshed — v1.1.6 package captures を readmeasset/ に追加
• 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、完全な npm test、smoke test、パッケージ監査、シークレットスキャンのゲート
• 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）と組み込み 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

Command	Description
omk init	.omk/、.kimi/skills/、.agents/skills/、docs、hooks、agents のスキャフォールドを作成
omk doctor	Node、Kimi CLI、Git、python3、tmux、スキャフォールドを診断
omk doctor --soft	ソフトモード：不足ツールでも失敗しない — smoke テストや CI に適しています
omk chat	エージェント/設定/MCP 自動検出対応の対話型 Kimi
omk plan <goal>	計画専用モード
omk hud	リアルタイムダッシュボード：システム使用量、Kimi クォータ、プロジェクト状態、実行追跡
omk lsp [server]	組み込み LSP ランチャー；デフォルトサーバーは TypeScript
omk star	GitHub star ヘルパー；手動再試行と状態確認
omk design init	frontmatter 付き DESIGN.md を作成
omk design list	ローカル/リモート DESIGN.md を一覧表示
omk design apply <name>	DESIGN.md をコードに変換適用
omk google stitch-install	Google Stitch スキルをインストール
omk update	OMK と Kimi CLI の更新を確認または実行
omk menu	@inquirer/prompts 経由でインタラクティブ OMK メニューを表示

Alpha

Command	Description
omk run <flow> <goal> (alpha)	フローベースのタスク実行
omk parallel <goal> (alpha)	並列 DAG 実行（coordinator → workers → reviewer）
omk review (alpha)	現在の変更のコードレビュー + セキュリティレビュー
omk review --ci (alpha)	CI モード：ローカル検証のみ、Kimi API 呼び出しなし
omk review --soft (alpha)	ソフトモード：レビュー失敗時も exit 0 を維持
omk verify (alpha)	完了済み実行のエビデンスゲート検証
omk summary (alpha)	実行サマリーとレポート生成
omk sync (alpha)	Kimi アセットを同期（hooks、MCP、skills、ローカルグラフメモリ）
omk sync --dry-run (alpha)	変更を適用せずプレビュー
omk sync --diff (alpha)	変更内容の diff を表示
omk sync --rollback (alpha)	manifest から前回の同期をロールバック
omk runs (alpha)	過去の OMK 実行を状態と日付で一覧表示
omk goal (alpha)	Codex スタイルのゴール管理

Experimental

Command	Status	Notes
omk team	レイアウトのみ	tmux ウィンドウ レイアウト スキャフォールド
omk agent	実験的	エージェントレジストリと YAML 診断
omk skill	実験的	Kimi スキルパックマネージャー
omk merge	手動	Diff 確認 + 手動 cherry-pick ガイダンス
omk design lint	スタブ	検証は未実装
omk design diff	スタブ	Diff は未実装
omk design export	スタブ	エクスポートは未実装

🏗️ 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 --> Kuzu["🧠 Kuzu (embedded)"]
    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：

• session-context.sh — 起動時に Open Design、ontology graph、デプロイ状態確認の注意を注入
• awesome-agent-skills-router.sh — プロンプトをインストール済み OMK skills/workflows へ advisory ルーティング
• precompact-checkpoint.sh — コンテキスト圧縮前に目標、ファイル、検証、ブロッカーの記録を促す
• subagent-stop-audit.sh — サブエージェント終了後の leader レビューと品質ゲートを要求
• pre-shell-guard.sh — rm -rf /、sudo、git push --force、未検証の release/publish 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 — 停止時の最終検証チェックリストと過大報告防止

📄 ライセンス

MIT

Open Design 連携は upstream の nexu-io/Open Design を使用し、同プロジェクトは Apache-2.0 ライセンスです。OMK 本体は引き続き MIT ライセンスです。

---

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.

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."