What the agent does
The agent reads the scripts before it writes a workflow, because an AdonisJS app already says how it wants to be checked. In an app made from the api kit in adonisjs/starter-kits, package.json has lint (eslint .), typecheck (tsc --noEmit), test (node ace test) and build (node ace build), with @adonisjs/core 7, @adonisjs/lucid on better-sqlite3, @adonisjs/auth with access tokens, @japa/runner, @japa/api-client and TypeScript 6. adonisrc.ts declares two Japa suites, unit under tests/unit and functional under tests/functional with a 30 second timeout, and tests/bootstrap.ts already loads the apiClient, sessionApiClient, authApiClient and dbAssertions plugins and starts the HTTP server for the functional suite. What the kit does not ship is a workflow, a lockfile or a single test file: the monorepo's own .github/workflows/ci.yml lints and typechecks the hypermedia, inertia-react and inertia-vue kits on Node 24 and never touches api.
Two files decide how the test job is wired. config/database.ts keeps SQLite at app.tmpPath('db.sqlite3'), and .gitignore ignores everything in tmp/ except .gitkeep, so a fresh checkout has no database and the two migrations under database/migrations/ (users, then access tokens) have to run before a functional test can sign anyone up. .env.test sets only SESSION_DRIVER=memory; the rest comes from .env, which is ignored too, so the workflow copies .env.example to .env before node ace test. The agent adds testUtils.db().migrate() to runnerHooks.setup in tests/bootstrap.ts, the place the kit leaves empty for exactly this, so the test run migrates its own SQLite file and tears it down.
With that in place the agent writes .github/workflows/ci.yml: push and pull_request on the default branch, a concurrency group per ref, permissions: contents: read, and actions/setup-node with Node 24 and the cache keyed to whichever lockfile the app has. The kit ships none, so the agent reads the one the reader's checkout committed (package-lock.json, pnpm-lock.yaml or yarn.lock, which .yarnrc.yml's nodeLinker: node-modules and pnpm-workspace.yaml's onlyBuiltDependencies both prepare for) and picks npm ci, pnpm install --frozen-lockfile or yarn install --immutable to match. better-sqlite3 compiles a native module on install; package.json approves its script under allowScripts and pnpm-workspace.yaml under allowBuilds, so the install needs no extra flag. The jobs then run lint, typecheck, test and build in that order, the build last because it proves the production bundle compiles.
A workflow that runs zero tests proves little, so the agent writes tests/functional/auth.spec.ts against the routes in start/routes.ts: POST /api/v1/auth/signup creates an account, POST /api/v1/auth/login answers with the user and a token from User.accessTokens.create, and GET /api/v1/account/profile behind the auth middleware returns the profile when that token is sent as a bearer and 401 when it is not. The test uses the client from @japa/api-client and assert from @japa/assert, both already registered in tests/bootstrap.ts. The agent runs the four scripts in its sandbox on Node 24 the way the workflow will, and actionlint on the YAML, before opening the pull request.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Set up GitHub Actions CI for this AdonisJS app. Read package.json, the lockfile, adonisrc.ts, tests/bootstrap.ts, bin/test.ts, config/database.ts, config/auth.ts, start/routes.ts, start/kernel.ts, .env.example, .env.test, .gitignore, eslint.config.js and tsconfig.json first. Add a workflow that installs with the package manager the lockfile belongs to, on Node 24, and runs the lint, typecheck, test and build scripts. Make the test run self-contained: copy .env.example to .env in the job, and migrate the SQLite database in the Japa setup hook so functional tests start from the migrations. Add one functional test that signs up, logs in with the access token guard and reads the authenticated profile route, plus the 401 case. Run all four scripts in the sandbox and actionlint on the workflow, then open a pull request that shows their output.",
"resources": [
{
"url": "https://github.com/your-org/your-app"
}
]
}- Get an API key. In the console, open Settings, then API keys, and create a key for your org. Send it in the Authorization header as a Bearer token.
- Send the request. POST the body in request.json to https://api.phoenix.so/o/{org}/sessions, with your repository as the resources URL. The run starts in its own cloud sandbox.
- Read what came back. Follow the run on its live stream or wait for the run.completed webhook, then read the result and its trajectory.
What comes back
What comes back is a pull request with .github/workflows/ci.yml, a new tests/functional/auth.spec.ts, and a one-line change in tests/bootstrap.ts that migrates the test database in runnerHooks.setup. The description shows the eslint, tsc --noEmit, node ace test and node ace build output from the sandbox, names the lockfile it found and the install command it chose, and lists the three assertions in the new test.
The run's record keeps every command, including the first node ace test before the migrate hook existed, which fails on a missing users table, and the passing run after it, so a reviewer can see why the bootstrap change is there.
Before merging, a reviewer checks the branch names in the trigger, decides whether the app should run tests on a database other than SQLite in CI, and whether APP_KEY from .env.example is acceptable for a test job or should come from a repository secret. The agent keeps SQLite, because that is what config/database.ts configures, and says so.
Run it on every repository with no workflow
A team with several AdonisJS services sends one POST /o/{org}/sessions per repository from the terminal or a script, with this prompt and the repository in resources, and a request_id that is a UUID v5 of the repository's full name so a retry returns the run already started. GET /o/{org}/github lists every repository the Rush GitHub App installation reaches, so the script can clone each and start a run only where .github/workflows/ is empty.
Each run reads its own package.json and lockfile, so a service on pnpm gets pnpm install --frozen-lockfile, one on PostgreSQL with pg in config/database.ts gets a services: block with a health check instead of SQLite, and one that already has unit tests gets the functional test added beside them. Each opens its own pull request.
The use case this job follows, with its automation and trigger: Hand off a long task.
Why run it on Rush?
A CI file copied from another project fails on the parts that are specific to this one: a native better-sqlite3 build the install never ran, a tmp/db.sqlite3 that does not exist on a clean runner, an .env that only a developer's machine has, or a cache keyed to the wrong lockfile. In a Rush sandbox the agent installs from the real lockfile, runs node ace test and reads the failure about the missing table before it decides where the migration belongs.
The commands and their output stay with the run, and the pull request cites them, so the reviewer sees the lint, typecheck, test and build results rather than a green badge that proves only that YAML parsed. Several services at once cost the same attention as one, because each run has its own sandbox, its own log and its own pull request.
Frequently asked questions
Why migrate in tests/bootstrap.ts rather than in the workflow?
Because then the tests set themselves up everywhere, not only on GitHub. runnerHooks.setup in tests/bootstrap.ts is the hook the kit leaves empty for this; testUtils.db().migrate() runs the migrations before the suites and rolls them back after. A developer running node ace test locally gets the same fresh database the runner gets.
Does it work with pnpm or yarn?
Yes. The kit ships .yarnrc.yml and pnpm-workspace.yaml as well as npm's allowScripts, but no lockfile, so the agent reads the one your repository committed and writes the matching install step and cache key. If there is no lockfile at all it says so in the pull request and uses npm install, which is less reproducible.
Will it change my existing tests or routes?
No. The new spec file lives beside whatever is in tests/functional/, the routes in start/routes.ts are read and not edited, and the only change to an existing file is the migrate call in the bootstrap's setup hook. If a functional test already signs up and logs in, the agent extends it instead of adding a duplicate.
Why run the build in CI?
node ace build compiles the TypeScript into build/, which is what npm start runs in production. tsc --noEmit catches type errors, but only the build catches an import that resolves in development through hot-hook and not in the compiled output.
More AdonisJS jobs
The same job on other stacks
- Add GitHub Actions CI to an Airflow Project (Apache Airflow)
- Set Up GitHub Actions CI for an AWS CDK App (AWS CDK)
- Set Up GitHub Actions CI for a Bazel Repository (Bazel)