For coding agents

Setup a coding agent can follow

This page is written so a coding agent can read it and set the platform up in a repository. The same content is served as plain text at /llms.txt and as a skill file.

The agent runs npx aitp init, reads the specs in tests/*.test.yaml, writes new specs in the same format, runs npx aitp run and reads the report at .aitp/last-run.json.

/llms.txt /llms-full.txt Download skill file

Install path

Step 1
Install and create the config
npx aitp init creates aitp.config.yaml and a tests/ folder with one example spec.
terminal
```
$npx aitp init$npx aitp run
```

Step 2

Check the config file

Set baseUrl to the environment under test. Keep heal.review so that text and order changes wait for a person.

aitp.config.yaml

1project: my-app2baseUrl: https://staging.example.com3browsers: [chromium]4retention: 30d5heal:6  auto: locator, layout     # what is healed without a human7  review: text, order       # what waits for review8ci:9  failOnHealedPending: true

Step 3

Write a minimal spec

One action per step, visible text in quotes, and at least one assert: line stating the expected outcome. Field reference: spec reference.

tests/login.test.yaml

1name: Log in with email2url: https://staging.example.com3steps:4  - Click "Log in"5  - Type EMAIL into the email field6  - Type PASSWORD into the password field7  - Click "Continue"8  - assert: The dashboard shows the account name

Step 4
Run and read the report
Exit code 0 pass, 1 fail, 2 healed steps pending review. The agent reads .aitp/last-run.json to see which step failed and why.
terminal
```
$npx aitp run# report written to .aitp/last-run.json
```

Step 5

Add the CI step

Store AITP_TOKEN and test credentials as CI secrets. The run reports a check per test on the pull request.

CI step

1- name: Run E2E2  run: npx aitp run --tag smoke --cloud3  env:4    AITP_TOKEN: ${{ secrets.AITP_TOKEN }}5    TEST_EMAIL: ${{ secrets.TEST_EMAIL }}6    TEST_PASSWORD: ${{ secrets.TEST_PASSWORD }}

Rules for the agent

Use only the fields, commands and flags listed in the docs. Do not invent others.
Write the spec in the same pull request as the feature, so the reviewer sees the check result next to the change.
Do not approve healed steps on behalf of a person. Report them; npx aitp heal --review is for a human reviewer.
When a test fails, read the classification in the report. Fix the spec only for test-issue; report app-bug and environment to a person.

Questions

Which coding agents can use this?

Any agent that can run shell commands and edit files in the repository, for example Claude Code, Cursor or Copilot in agent mode. The setup uses only npx and YAML files.

Is there an MCP server?

Not in the current documentation. The agent works through the CLI and the report file.

How should an agent test code it wrote itself?

Write a spec for the flow the change touches, run it, and leave approval of healed steps and classification to a person. See the blog post on testing AI-generated code.

For teams shipping with coding agents

Request early access

Early access: prices are the planned list prices. Billing starts when the platform enters beta; early-access use is not charged. We reply within 24 hours on business days.

Request early access See how it works