Skip to content
ReviewRamble

The weekly brief

AI, reviewed, in your inbox.

  • New reviews of tools, models, MCP servers and repos, with the verdict up front
  • Re-checked prices and ratings when something changes
  • New analysis pieces as they are published

Models, agents, MCP servers and the tools actually worth your time.

One email a week. Unsubscribe with the link in any brief. How we use your email

Boss Skill

Agent skill that chains nine BMAD-style roles from PRD to deployment, with hard per-role checklists and every stage written to disk.

Reviewed by ReviewRamble editors · Updated

View on GitHub Open source

Share this page:XLinkedInFacebookRedditWhatsAppEmail

Our rating

3.4/5

Boss Skill is BMAD on autopilot: thorough paperwork and real test gates, paid for with a long, token-hungry run that you have to remember to stop.

What is Boss Skill?

Boss Skill turns one coding agent into a nine-role team modelled on the BMAD method: PM, Architect, UI Designer, Tech Lead, Scrum Master, Frontend, Backend, QA and DevOps. Each role is a subagent prompt with a table of non-negotiable requirements, and each stage writes a file under `.boss/<feature>/`: prd.md, architecture.md, ui-spec.md, tech-review.md, plan.md, tasks.md, qa-report.md and deploy-report.md. From whichever stage you start, it runs every later stage through to ship unless you say stop or pass `--stop-after`.

It installs as a Claude Code or Codex plugin or through `npx skills add echoVic/boss-skill`, and a separate npm bundle, dsh-boss-mode, adds it to DeepSeek Harness. The skill's own instructions are written in Chinese and its documents default to Chinese; the GitHub README is in English with six translations.

Is Boss Skill worth it?

Boss is a lighter and more automatic take on BMAD. Where BMAD Method has you walk through its planning documents, Boss chains the stages by default: `/boss` goes from PRD to deployment in one request, and `/boss:dev` starts at code and still runs QA and ship. Real commands decide three gates: typecheck, lint and a dependency audit after development; tests with coverage of at least 70% before deployment; and, for web projects, a Lighthouse score of 80 or more and API P99 latency under 500 ms.

Cost is the question the docs leave open. No token or price figures are published, and the role prompts pin no model, so every stage runs on whatever model your session uses. A full chain dispatches eight stages, each loading a role prompt of 121 to 331 lines plus the earlier artifacts, before code is written test-first. The README's own advice is to start with `/boss:review`, which writes only tech-review.md and plan.md, and to skip the pipeline when you do not need a traceable `.boss/` folder; the skill itself declines tasks a person would finish in under 30 minutes.

The ship stage can deploy. The DevOps prompt looks for Vercel, Railway, Render, Fly.io, Cloudflare or Netlify configuration and runs the matching deploy skill. Production deploys, migrations and DNS changes need your confirmation, but that rule lives in a prompt rather than in code, so `--stop-after qa` is the safer default until you trust it.

The repository is partway through a rewrite. A refactor on 27 October 2026 removed the CLI, the hooks and the event-sourced runtime, yet the Claude Code plugin manifest still lists a hooks file the repository no longer contains, and PRIVACY.md still describes the removed event log. The privacy claim itself holds in our reading of the current files: no telemetry, no outbound requests and no credentials read. The manifest says version 5.0.4, the last dated changelog entry is 5.0.3 on 29 September 2026, and the repository has 619 stars with almost all commits from one maintainer.

Where it earns its place

  • A solo developer starting a small web app

    Go from an idea to a tested first version with the planning on file

    Run `/boss Build a local personal todo app --skip-ui --stop-after qa` and let it write the PRD, architecture, tasks and test-first code, stopping before deployment. Expect a long session on your agent's model; no cost figures are published.

  • A tech lead inheriting an unfamiliar codebase

    Get a written review and plan without changing any code

    `/boss:review` writes tech-review.md and plan.md and touches nothing else, the cheapest entry the README suggests. Add a stop instruction, or the chain carries on into tasks, development and QA.

  • A Chinese-speaking product team on DeepSeek Harness

    Keep a traceable record of each feature from PRD to QA

    Install `dsh-boss-mode` and pick the Boss Expert Squad preset; artifacts land in `.boss/<feature>/` in Chinese by default, each opening with a summary so later roles read the conclusion first.

Before you commit

Setup
Claude Code: `/plugin marketplace add echoVic/boss-skill`, then `/plugin install boss@boss-skill`. Codex: `codex plugin marketplace add echoVic/boss-skill`. Other agents: `npx skills add echoVic/boss-skill`. DeepSeek Harness: `dsh plugin --profile desktop add dsh-boss-mode`. Working on the repository itself needs Node.js 22.18 or later.
Integrations
Claude Code, Codex, OpenClaw, Antigravity, Hermes and DeepSeek Harness; deploy skills for Vercel, Railway, Render, Fly.io, Cloudflare and Netlify; your project's own test, lint and Lighthouse tooling for the gates.
Data and privacy
Local by default: PRIVACY.md states no telemetry, no outbound requests and no reading of API keys, and artifacts stay in `.boss/` inside your repository. Your agent still sends prompts and code to its model provider. MIT licence.
Cost at scale
No paid tier. The cost is model usage across up to eight stages per feature on your session's model; no token or price figures are published, so measuring one `/boss:review` run before committing to full chains is sensible.
Limits
Not meant for single-file fixes or tasks under 30 minutes. Documents default to Chinese. The plugin manifest references a hooks file removed in the 27 October 2026 refactor. Three open pull requests, all dependency updates.

What’s good

  • Start at any stage; later stages run automatically
  • Measured gates: coverage, lint, Lighthouse and P99 latency
  • No telemetry or outbound requests
  • Installs in Claude Code, Codex, the skills CLI and DeepSeek Harness

What’s not

  • No published token or cost figures
  • Documents default to Chinese
  • Deploy confirmation enforced only by a prompt
  • Manifest and privacy docs lag the October refactor

Best for

Developers who want BMAD-style PRDs, architecture notes and QA records produced automatically on new features, and can afford long agent runs.

Boss Skill alternatives

Boss Skill: common questions

Is Boss Skill worth it?
Boss Skill is BMAD on autopilot: thorough paperwork and real test gates, paid for with a long, token-hungry run that you have to remember to stop. We rate it 3.4 out of 5.
How much does Boss Skill cost?
Free and open source (MIT). Model usage is billed by your coding agent's provider.
What are the downsides of Boss Skill?
No published token or cost figures; documents default to Chinese; deploy confirmation enforced only by a prompt; manifest and privacy docs lag the October refactor.
Who is Boss Skill best for?
Developers who want BMAD-style PRDs, architecture notes and QA records produced automatically on new features, and can afford long agent runs.
How much does Boss Skill cost for a team or in production?
No paid tier. The cost is model usage across up to eight stages per feature on your session's model; no token or price figures are published, so measuring one `/boss:review` run before committing to full chains is sensible.
Is Boss Skill safe for company data?
Local by default: PRIVACY.md states no telemetry, no outbound requests and no reading of API keys, and artifacts stay in `.boss/` inside your repository. Your agent still sends prompts and code to its model provider. MIT licence.
What are the best Boss Skill alternatives?
The closest alternatives we have reviewed are BMAD Method, Superpowers and Spec Kit.

Answers checked Nov 18, 2026.