Minimal useful starter

A starter contract should be obvious, stable, and runnable.

Starter contractyaml

version: 1metadata:  ota:    minimum_version: "1.6.26"project:  name: ota-web  type: applicationtoolchains:  node:    version: ">=22"    package_managers:      pnpm: "10"    fulfillment:      source: corepack      mode: runtasks:  setup:    prepare:      kind: dependency_hydration      medium: package_dependencies      source:        kind: node_package_manager        cwd: .        manager: pnpm        mode: install    requirements:      toolchains: [ node ]    effects:      writes: [ node_modules ]      network: true      network_kind: dependency_hydration  build:    depends_on: [ setup ]    command:      exe: pnpm      args: [ build ]    requirements:      toolchains: [ node ]    effects:      writes: [ dist ]  test:    depends_on: [ build ]    command:      exe: pnpm      args: [ test ]    requirements:      toolchains: [ node ]  deploy:    depends_on: [ test ]    command:      exe: pnpm      args: [ deploy ]    effects:      writes: [ release ]    safe_for_agent: falseagent:  entrypoint: setup  default_task: test  safe_tasks: [ setup, build, test ]  refusal_canaries:    - task: deploy  protected_paths: [ ota.yaml, .env, .env.local ]  verify_after_changes: [ build, test ]

What makes it useful

it names the repo
it declares the core runtime and tools
it defines safe tasks and verification tasks
it gives agents a deterministic entrypoint

When the first task is a service

If the starter contract includes a long-running app server, docs preview, or worker process, prefer launch.kind: command instead of hiding that start path inside raw run shell text.

launch owns process start. runtime plus surfaces owns the reachable URL, listener, and readiness truth.

use launch.kind: command for long-running service processes
use runtime.kind: service with surfaces or listeners for the endpoint and readiness contract
use aggregate.tasks for named verification entrypoints that only run other tasks
keep run for finite shell tasks, pipelines, or real escape-hatch cases

Structured service taskyaml

tasks:  dev:    launch:      kind: command      exe: bundle      args: [exec, rails, server, -b, 0.0.0.0, -p, "3000"]    runtime:      kind: service      surfaces:        - api