What the agent does
The agent starts by reading how the API is put together, because an audit log has to sit where every mutation already passes. In a Fastify API shaped like fastify/demo, package.json pins fastify ^5.10.0 with @fastify/autoload, @fastify/session, @fastify/type-provider-typebox, knex on mysql2, and postgrator for the numbered SQL files in migrations/. src/app.ts autoloads src/plugins/external, then src/plugins/app, then src/routes with autoHooks on, and src/routes/api/autohooks.ts rejects any unauthenticated request outside /api/auth/login, so request.session.user is set on every route the log cares about. src/routes/api/tasks/index.ts holds the mutations: POST /, PATCH /:id, DELETE /:id behind request.isAdmin, POST /:id/assign behind request.isModerator, and the upload and image routes.
It adds migrations/006.do.audit_log.sql and its undo pair, following the 002.do.tasks.sql shape: an audit_log table with id, actor_id referencing users(id), entity and entity_id, action, a changes JSON column, and created_at defaulting to CURRENT_TIMESTAMP. The repository goes in src/plugins/app/audit/audit-repository.ts as a fastify-plugin that depends on knex, like tasks-repository.ts, with record() and paginate() methods. The schema for a log row goes in src/schemas/audit.ts in TypeBox, so the response is validated and shows up in Swagger under its own tag.
Rather than editing each handler, the agent writes the recording in one place. In src/routes/api/tasks/index.ts the POST, PATCH, DELETE and assign routes gain an onSend hook scoped to that plugin, which fires only when the reply status is below 400 and calls auditRepository.record() with request.session.user.id, the task id from request.params or the created id in the payload, the method mapped to create, update, delete or assign, and the request body as changes. The read-only routes are left alone. A new src/routes/api/audit/index.ts exposes GET /api/audit with page and limit query parameters and the preHandler: (request, reply) => request.isAdmin(reply) pattern the delete route already uses.
Tests follow test/routes/api/tasks/tasks.test.ts: build() from test/helper.ts, app.injectWithLogin('admin@example.com', …) for the session cookie, and rows inserted with app.knex. The new test/routes/api/audit/audit.test.ts creates a task through the API, patches it, checks one create and one update row exist with the right actor_id, checks that a moderator gets 403 from GET /api/audit, and checks a failed PATCH on a missing id writes nothing. The agent runs npm run db:migrate, npm run lint, npm run build -- --noEmit and npm test against the mysql:8.4 container from docker-compose.yml, the same steps .github/workflows/ci.yml runs on Node 22 and 24.
Start it with one API call
POST https://api.phoenix.so/o/{org}/sessions with this body:
{
"agent": "claude",
"input": "Add an audit log to this Fastify API and open a pull request. Read package.json, src/app.ts, src/routes/api/autohooks.ts, src/routes/api/tasks/index.ts, src/plugins/app/tasks/tasks-repository.ts, src/plugins/app/authorization.ts, src/schemas/tasks.ts, migrations/002.do.tasks.sql, test/helper.ts, test/routes/api/tasks/tasks.test.ts, docker-compose.yml and .github/workflows/ci.yml first. Add migrations/006.do.audit_log.sql and 006.undo.audit_log.sql creating audit_log with id, actor_id referencing users(id), entity, entity_id, action, a JSON changes column and created_at. Add src/plugins/app/audit/audit-repository.ts as a fastify-plugin depending on knex, with record() and paginate(). Add src/schemas/audit.ts in TypeBox. In src/routes/api/tasks/index.ts, record create, update, delete and assign through one onSend hook that only writes on 2xx responses, using request.session.user.id as the actor. Add src/routes/api/audit/index.ts with GET /api/audit, paginated, admin only via request.isAdmin. Add test/routes/api/audit/audit.test.ts using build() and injectWithLogin from test/helper.ts: a task created then patched leaves one create and one update row with the right actor_id, a moderator gets 403 from GET /api/audit, and a failed PATCH writes nothing. Run npm run db:migrate, npm run lint, npm run build -- --noEmit and npm test against the docker-compose MySQL, and paste the output in the pull request. Do not change the existing route schemas or the login flow.",
"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 two migration files, audit-repository.ts under src/plugins/app/audit/, src/schemas/audit.ts, the hook in src/routes/api/tasks/index.ts, the admin-only src/routes/api/audit/index.ts, and test/routes/api/audit/audit.test.ts. The description lists the columns, which routes write which action, and the output of npm run lint, the tsc --noEmit build and npm test with the c8 coverage table from the sandbox.
The repository's CI runs the same npm run build -- --noEmit, npm run lint and npm test on the pull request against the mysql:8.4 service on Node 22 and 24, so a migration that fails under Postgrator or a type the TypeBox schema does not accept shows up as a red check rather than after merge.
A reviewer checks that the hook writes only on success, that the actor comes from the session and never from the body, that GET /api/audit is behind isAdmin and paginated with the same page and limit names the tasks list uses, and that the undo migration drops the table cleanly. Then they decide whether changes should also hold the previous values, which the prompt leaves out on purpose.
Run it on every service that handles customer data
A team with several Fastify services sends one POST /o/{org}/sessions per repository with this prompt and that repository in resources. Each run reads its own routes and session plugin, so a service that authenticates with a JWT header instead of @fastify/session gets the actor from its own request decorator, and each opens its own pull request.
Give the API key a webhook_url and Rush POSTs run.completed to it when each turn ends, with a session_url to read the pull request from, so a script can list the open audit-log pull requests in one place.
The use case this job follows, with its automation and trigger: Hand off a long task.
Why run it on Rush?
An audit log is easy to describe and tedious to prove: the proof is the row that appears after a real PATCH and the row that does not appear after a 404. In a Rush sandbox the agent brings up the mysql:8.4 container from docker-compose.yml, runs the Postgrator migration, drives the API through app.inject with a real session cookie, and keeps every command and its output in the run's record, so the pull request shows the rows, not a claim about them.
The work passes the repository's own gates before anyone reads it: neostandard lint, the TypeScript build, and the node:test suite under c8. The same request body works unchanged on the next Fastify service, and each run opens its own pull request on its own branch.
Frequently asked questions
Why an onSend hook instead of a call in each handler?
Because the routes already share one plugin file, a hook scoped to it covers POST, PATCH, DELETE and assign without touching each handler's body, and it can check reply.statusCode so a 404 or a validation 400 writes nothing. The read-only GET routes stay out of the log because the hook maps only the mutating methods.
Does it work with MySQL and the Postgrator migrations?
Yes. The demo stores migrations as numbered .do and .undo SQL files run by scripts/migrate.ts through the mysql driver, so the agent adds 006.do.audit_log.sql and 006.undo.audit_log.sql in that format and uses MySQL's JSON type for changes.
Who can read the log?
Only users with the admin role, through the same request.isAdmin(reply) preHandler that guards DELETE /api/tasks/:id. Moderators and plain users get 403, and the test asserts it.
Will the existing tests still pass?
The hook only adds a row after a successful mutation, so the task tests in test/routes/api/tasks/tasks.test.ts see the same responses. The agent runs the full npm test, which seeds the database first, and pastes the result in the pull request.
More Fastify jobs
The same job on other stacks
- Add an Audit Log to a Filament Admin Panel (Filament)
- Add an Audit Log to a Mongoose and Express API (Mongoose)
- Add an Audit Log to a Prisma Project (Prisma)