Mandu Deployment

Mandu Deployment Skill

Mandu 앱을 프로덕션 환경에 안전하고 효율적으로 배포하기 위한 가이드입니다.

Agent Workflow Contract

This skill is a Domain addendum. It must not replace mandu-agent-workflow. Use it only after mandu.agent.plan selects the deploy domain.

Canonical workflow step: plan -> apply -> verify.

Preferred MCP tools:

Step	Tools
plan	mandu.agent.plan, mandu.deploy.plan
apply	mandu.agent.apply, mandu.deploy.compile
verify	mandu.agent.verify, mandu.deploy.preview
repair	mandu.agent.repair

Allowed file edits:

• .mandu/deploy.intent.json
• Provider artifacts named in the plan, such as render.yaml, Dockerfile, docker-compose.yml, fly.toml, vercel.json, or netlify.toml
• CI workflow files only when deploy automation is explicitly requested

Verification command:

mandu agent verify --changed --json --write

Common failures:

• Executing a provider deploy before a dry-run or preview step
• Writing secrets into tracked config
• Changing provider artifacts without route/deploy intent verification

Repair path:

mandu agent repair --from .mandu/agent-verify.json --json

핵심 원칙

1. Bun 네이티브: Bun 런타임과 번들러를 최대한 활용
2. 환경 분리: 개발/스테이징/프로덕션 환경 명확히 구분
3. 자동화: CI/CD를 통한 일관된 배포 프로세스
4. 보안 우선: 민감 정보는 환경 변수로 관리

Provider Artifact Examples

Use these examples only after mandu.agent.plan selects the deploy domain and mandu.agent.apply has produced the intended action order.

Render 배포 (권장)

# render.yaml
services:
  - type: web
    name: mandu-app
    runtime: node
    buildCommand: bun install && bun run build
    startCommand: bun run start
    envVars:
      - key: NODE_ENV
        value: production
      - key: BUN_ENV
        value: production

Docker 배포

FROM oven/bun:1.3.14

WORKDIR /app
COPY package.json bun.lockb ./
RUN bun install --frozen-lockfile --production

COPY . .
RUN bun run build

EXPOSE 3000
CMD ["bun", "run", "start"]