Token导航 LogoToken导航

Agent Skill · 开发

bmad(bmad 命令行)

bmad 用于处理 GitHub 仓库、Issue、Pull Request 和代码协作信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要围绕仓库状态、代码变更或协作事项进行整理时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

bmad 是开发类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证信息还不明确,第三方扫描有未通过项,按安装量在开发类的 21,746 个 Skill 里排第 11,219。

总安装 1,882周安装 80GitHub Stars 11下载 659核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:bmad(bmad 命令行)
来源仓库:https://github.com/akillness/oh-my-skills
仓库路径:skills/bmad
安装命令:
npx skills add https://github.com/akillness/oh-my-skills --skill bmad
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/akillness/oh-my-skills --skill bmad

这个 Skill 是做什么的

开发类 Skill 面向写代码、改代码和排查问题,一般会读写仓库文件,部分还会执行构建或测试命令。

bmad 作为 BMAD/BMM 的前置路由层,负责识别用户数据包并推荐下一步关键产出。它将运行时、评审与执行细节外派,聚焦于减少当前阶段的不确定性。适用于结构化开发流程管理,需配合 intake-packets-and-route-outs.md 使用。使用前应熟悉核心路由与状态管理规范,避免错误跳转或循环依赖。

bmad 属于开发类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/bmad

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/akillness/oh-my-skills --skill bmad;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

bmad 怎么安装?

在终端执行 npx skills add https://github.com/akillness/oh-my-skills --skill bmad。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

bmad 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

bmad 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

bmad 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

bmad 用的人多吗?

公开数据是累计安装 1,882 次,近一周安装 80 次,下载 659 次,GitHub 11 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

bmad 安全吗?

第三方扫描有未通过项,风险较高。除非你确认过源码,否则不建议直接安装到常用环境。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理开发类任务、又不想每次都手写提示词的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

开发类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
skill-vetter

技能审查员 · OpenClaw

8,779,018986操作浏览器MIT-0通过
ontology

本体论 · OpenClaw

6,636,960576只读MIT-0通过
entra-app-registration

内部应用程序注册 · Codex、Claude

6,631,872764接触敏感数据未明确通过
entra-app-registration

内部应用程序注册 · Codex、Claude

3,757,680201接触敏感数据未明确通过
nano-banana-pro

纳米香蕉专业版 · OpenClaw

3,526,898364接触敏感数据MIT-0有提醒
bmad

bmad 命令行 · 本页

1,88211需要联网未明确未通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

BMAD Packet-First Router

Use bmad as the common-layer BMAD/BMM front door.

The job is not to dump every phase command or runtime setup rule. The job is to:

  1. identify the packet the user already has,
  2. choose the project level and likely current phase,
  3. recommend the single next artifact or gate that reduces ambiguity now,
  4. route runtime-, review-, execution-, or game-specific detail outward.

Read references/intake-packets-and-route-outs.md first for the fastest routing model. Use references/core-routing.md, references/status-and-review.md, and references/runtime-and-module-boundaries.md when you need more depth.

When to use this skill

  • The user wants to start or resume a BMAD/BMM-style workflow from mixed project state
  • The user asks what phase comes next after idea notes, a PRD, a tech spec, architecture work, or partial implementation
  • The user has an existing repo, issue, review note, or brownfield status and needs the next artifact chosen truthfully
  • The user wants a vendor-neutral BMAD front door before dropping into Claude-, Codex-, or Gemini-specific runtime overlays
  • The user needs help deciding whether the next move is ideation, planning, architecture, approval, execution slicing, or runtime handoff

When not to use this skill

  • The main need is open-ended concept shaping before project framing exists → use bmad-idea
  • The main need is a game-production packet (GDD, playtest notes, build issues, launch beats) → use bmad-gds
  • The main need is artifact approval, annotation, or plan review → use plannotator
  • The main need is execution-ready backlog slicing after the next artifact is already known → use task-planning
  • The main need is runtime-specific orchestration setup for Claude Code → use omc
  • The main need is Codex-first workflow runtime behavior → use omx
  • The main need is Gemini / Antigravity portable harness setup → use ohmg

Instructions

Step 1: Normalize the intake as one primary packet

Before talking about phases, choose the single best packet that describes the current ask:

bmad_packet:
  packet_type:
    - concept-packet
    - planning-packet
    - architecture-packet
    - review-gate-packet
    - execution-packet
    - runtime-handoff-packet
    - brownfield-resume-packet
  evidence_in_hand:
    - idea-notes
    - product-brief
    - prd
    - tech-spec
    - architecture-draft
    - review-feedback
    - active-repo
    - issue-or-bug-link
    - sprint-plan
    - milestone-or-launch-pressure
  project_type: web-app | backend-api | mobile-app | game | library | automation | unknown
  project_level: 0 | 1 | 2 | 3 | 4 | unknown
  runtime_context: claude | codex | gemini | opencode | mixed | unknown
  main_constraint: ambiguity | review-blocker | execution-readiness | runtime-setup | brownfield-drift | unknown

Rules:

  • Pick one primary packet first, even if the user mentions several artifacts.
  • If the packet is obviously pre-planning, say so and route to bmad-idea early.
  • If the packet is mostly game-production coordination, route to bmad-gds early.
  • If the request starts from an existing repo, issue, or partial implementation, prefer brownfield-resume-packet over pretending the project is greenfield.

Step 2: Choose project level before artifact depth

Use this scale:

  • Level 0 — single atomic change, bug fix, config tweak
  • Level 1 — small feature, limited files, low coordination cost
  • Level 2 — medium feature set, cross-cutting behavior, clear planning + architecture need
  • Level 3 — complex integration, multiple subsystems, API/data contracts, multi-sprint delivery
  • Level 4 — major overhaul, platform migration, enterprise-scale coordination

Do not force architecture or heavyweight planning onto level 0–1 work unless the visible packet truly requires it.

Step 3: Infer the likely current phase from the packet

Use the packet to infer the current BMAD phase, not the other way around:

Packet typeLikely phaseDefault next artifact or gate
concept-packetanalysisproduct-brief or route to bmad-idea
planning-packetplanningprd or tech-spec
architecture-packetsolutioningarchitecture or architecture review gate
review-gate-packetboundary between phasesplannotator review before advancing
execution-packetimplementationsprint-plan, story packet, or route to task-planning
runtime-handoff-packetimplementation-readyruntime route-out to omc, omx, or ohmg
brownfield-resume-packetmixed / unknownworkflow-status update, repo-state summary, then the smallest truthful next artifact

Guiding rules:

  • Prefer the smallest truthful next artifact.
  • If an artifact already exists but has not been reviewed, the next move is often the review gate, not another new document.
  • If the request is really a runtime/setup ask, keep bmad short and route outward.

Step 4: Choose one next artifact, not a parallel stack

Pick the single next artifact or gate that best reduces ambiguity now:

Current statePreferred next move
vague idea, opportunity, user problemproduct-brief
requirements emerging, implementation shape still unclearprd or tech-spec
planning exists, system shape is still fuzzyarchitecture
architecture exists but approval is missingreview gate
approved architecture exists, execution slicing is weaksprint-plan or story packet
active implementation exists but status is unclearworkflow-status update + next-story recommendation
runtime is the real blocker after approvalruntime handoff to omc / omx / ohmg

Step 5: Route specialist work explicitly

If the front-door decision is made, route the rest clearly:

  • bmad-idea → pre-planning concept framing before formal BMAD artifacts exist
  • plannotator → review / approval gate for PRD, architecture, sprint plan, or plan diffs
  • task-planning → execution-ready slicing after the next artifact is approved or obvious
  • omc → Claude-first runtime orchestration, hooks, team mode, stop callbacks
  • omx → Codex-first orchestration, AGENTS.md injection, workflow skills, tmux team runtime
  • ohmg → Gemini / Antigravity portable harness, .agents/ source of truth, generated runtime views
  • bmad-gds → game-specific production routing

Leave a short reason for the route-out. Do not just name the neighboring skill.

Step 6: Keep state and review visible

When a project already uses BMAD state files or helper scripts, keep them in the loop:

  • inspect status/config artifacts before recommending a phase leap
  • if a required artifact exists but lacks approval, route to plannotator
  • prefer script/status awareness over re-deriving the workflow from scratch
  • use brownfield evidence honestly instead of pretending the project is still at idea stage

Step 7: Produce a compact BMAD routing brief

Always return this structure:

# BMAD Routing Brief

## Scope
- Packet type: ...
- Project type: ...
- Project level: 0 | 1 | 2 | 3 | 4
- Runtime context: ...
- Confidence: high | medium | low

## Current phase
- analysis | planning | solutioning | implementation | mixed | unknown
- Why: ...

## Recommended next move
- product-brief | prd | tech-spec | architecture | review gate | sprint-plan | story packet | workflow-status update | runtime handoff

## Why this is next
- 2-4 bullets grounded in the packet

## Route-outs
- Skill / module: ...
- Why: ...
- What to pass forward: ...

## What not to do yet
- 1-3 bullets preventing premature detail or the wrong lane

Step 8: Push detail into references

Do not inflate the main response with every helper script, command family, or runtime rule. Use the references when needed:

Output format

Return a short BMAD routing brief.

Required qualities:

  • choose one primary packet before choosing the next artifact
  • make the project-level assumption explicit
  • keep review-gate visibility before phase advancement
  • keep runtime/setup detail separated from the common BMAD layer
  • keep the result under roughly 400-700 words unless the user asks for a full workflow packet

Examples

Example 1: brownfield repo, unclear next step

Input

Use bmad. We already have a repo and some product notes, but planning drifted and I don't know whether we need a PRD, architecture, or just sprint planning.

Output sketch

  • Packet type: brownfield-resume-packet
  • Likely phase: mixed
  • Recommended next move: workflow-status update, then whichever artifact is actually missing
  • Route-outs only after the missing artifact is identified

Example 2: architecture exists, execution weak

Input

We already wrote the architecture doc for our API migration. What should we do next in BMAD?

Output sketch

  • Packet type: architecture-packet
  • Current phase: solutioning
  • Recommended next move: review gate if architecture is unreviewed, otherwise sprint-plan
  • Route-out: plannotator first, task-planning second

Example 3: runtime-specific follow-up

Input

I want BMAD for Codex CLI. What should I actually use?

Output sketch

  • Keep bmad as the packet/phase router
  • Packet type: runtime-handoff-packet
  • Route Codex runtime specifics to omx
  • Do not turn bmad into a Codex setup guide

Best practices

  1. Act like a packet-first router — start from the artifact or evidence the user already has.
  2. Choose level before depth — level 0–1 and level 2–4 should not produce the same paperwork.
  3. Keep review visible — if approval is the blocker, say so before inventing a new artifact.
  4. Keep runtime overlays separate — omc, omx, and ohmg own vendor/runtime specifics.
  5. Prefer one clear handoff — one next move plus one or two route-outs beats a giant workflow dump.
  6. Use references for deep detail — helper scripts, status rules, and runtime boundaries belong in support docs.

References

平台分布

Codex

31.27%
按下载量换算206

Claude

31.32%
按下载量换算206

Cursor

20.19%
按下载量换算133

Gemini CLI

8.65%
按下载量换算57

安全审计

Gen Agent Trust Hub

未通过

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills