bwai-cli

boilerplates-with-ai-skills

npx a stack starter that ships with curated, scanned agent skills — wired for Claude, Cursor, Codex, and Copilot, with NVIDIA SkillSpector gating and a skills.lock provenance file.

Every generated project includes the Omni-Skills startup workflow bench: $startup-goal, $founding-engineer, $qa-lead, $cto, and $product-manager are ready the moment you scaffold.

Recommended flow: idea → advisor → scaffold

• Install the advisor (once, globally — also installs startup-goal):

npx bwai-cli install-skill bwai-advisor --global

• Shape the idea in any agent chat (paste an idea or attach a .md):

$bwai-advisor  I want to build …
$bwai-advisor  Read @my-idea.md and recommend a boilerplate

• Run the command the advisor prints:

bwai new nextjs-app ./my-app --agents claude,cursor

• Continue inside the project with $founding-engineer / $qa-lead.

Already know the boilerplate? Skip to step 3. Check setup anytime: bwai doctor.

After global install, bwai-cli and the shorter alias bwai are the same CLI.

Quick start

Install advisor	npx bwai-cli install-skill bwai-advisor --global
List starters	npx bwai-cli list-boilerplates
Check setup	npx bwai-cli doctor
Scaffold	npx bwai-cli new nextjs-app ./app --agents claude,cursor
Scan skills	bwai scan-project ./app --threshold 50

Seven boilerplates: nextjs-app, nextjs-ai-app, express-api, fastify-api, python-service, node-service, react-native-app.

What you get on scaffold

my-app/
  src/ or app/              # runnable template for your stack
  .bwai/skills/             # canonical curated skills
    startup-goal/           # orchestrate a goal across role subagents
    founding-engineer/      # implement, test, debug, verify
    qa-lead/                # acceptance checks + release risk
    cto/                    # architecture + technical direction
    product-manager/        # PRDs, issue slicing, roadmap
    …                       # stack-specific skills (TDD, code-review, etc.)
  .bwai/plugin/             # Agent Plugins 1.0 package (skills + typed MCP)
  .claude/skills/ …         # mirrored for each --agents target
  skills.lock               # SHA-256 + scan status per skill
  .github/workflows/skill-scan.yml  # SkillSpector gate on push/PR
  .github/copilot/settings.json     # optional Copilot enabledPlugins
  workflows/bwai-delivery/  # delivery workflow bundle

Each skill is a spec-compliant SKILL.md. Skills + MCP also ship as a portable Agent Plugins package under .bwai/plugin/ — see docs/agent-plugins.md. Trust (SkillSpector + skills.lock) stays bwai’s layer; the plugin format does not define provenance. Role skills are vendored from Omni-Skills and pinned in registry/skills-index.json. Run bwai sync-upstream to pull updates.

Usage

bwai install-skill bwai-advisor --global   # also installs startup-goal
bwai list-boilerplates
bwai doctor                                # includes global advisor check
bwai new node-service ./my-app --agents claude,cursor
bwai export-plugin nextjs-app ./bwai.nextjs-app   # portable Agent Plugins package
bwai scan-project ./my-app --threshold 50
bwai scan-catalog --threshold 30 --require-scanner
bwai search-skills "code review"
bwai promote my-skill --from ./path/to/skill --target shared --require-scanner
bwai sync-upstream          # pull latest omni-skills content
bwai sync-skills

The safety gate

bwai scan-project runs SkillSpector over each installed skill, updates skills.lock, writes safety-reports/ (JSON + SARIF), and exits 1 when any skill exceeds the threshold.

uv tool install git+https://github.com/NVIDIA/skillspector.git
bwai scan-project --threshold 50 --require-scanner

Without SkillSpector locally, scans record skipped unless you pass --require-scanner.

Repository layout

src/                     # bwai-cli (TypeScript)
shared/skills/           # catalog skills (source: "shared")
shared/workflows/        # GetSuperpower bundles (e.g. bwai-delivery)
boilerplates/<name>/     # boilerplate.json, template/, skills/, optional workflow/
registry/skills-index.json  # upstream pins + scan status
docs/
tests/

See docs/ARCHITECTURE.md for boilerplates vs skills vs workflows.

Adding a boilerplate

Create boilerplates/<name>/ with boilerplate.json, template/ (ship .gitignore as gitignore), optional local skills/ and workflow/. Reference shared skills with { "name": "code-review", "source": "shared" }. Bundled skills must pass SkillSpector in CI.

Contributors: clone, npm install, npm run build, npm test — see CONTRIBUTING.md.

Landing page: https://boilerplates-with-ai-skills.vercel.app

Credits

• Omni-Skills by devos-ing — startup role workflow bench bundled in every project.
• NVIDIA SkillSpector — safety scanner that gates every bundled skill.

License

MIT