Skip to main content
Proton v5

Project from scratch

With the connector, the assistant starts an automation project in Python or Java from scratch: the code already in the Proton structure and everything registered in Proton, without opening the Proton screens.

What you need​

  • An assistant that works on your machine, such as Claude Code or Codex: it clones the template, writes the code, runs it, and pushes.
  • Write access enabled in the organization and in the connection. See Permissions and security.
  • A git repository for the project (GitHub, GitLab, or Bitbucket) and an organization git integration registered in Proton, on the Integrations screen. The credential stays in Proton.
  • A runner available for the first run.

How to start​

The connector includes a ready-made prompt, criar_projeto_do_zero, with two inputs: the language (python or java) and the goal (what to automate). In Claude Code:

/mcp__proton__criar_projeto_do_zero

In other assistants, ask in your own words, for example: "I want to create a Java project from scratch, with Proton, to test the store checkout".

The prompt only appears when write access is enabled.

The path​

The assistant follows these steps and shows you its proposal before each decision (the repository, the integration, the systems, the names of components and parameters, the automation, and the runner):

  1. Template. Clones the language's template project into a new folder and removes from the sample whatever doesn't serve the goal.
  2. Repository. Asks where the repository is and makes the first push.
  3. Git integration. Confirms with you which of the organization's integrations to use.
  4. Repository and systems in Proton. Registers the repository and one system per platform under test (for example, the portal and the API), linked to it.
  5. Components. Proposes the steps, with input (in_) and output (out_) parameters and documentation, reusing parameters that already exist, and registers them in Proton.
  6. Code. Generates each component's file in the template's format and writes the logic. The generated file fails on purpose until the logic is written: a run never passes without doing anything.
  7. Automation and data. Creates the automation and the dataset, with the steps in order and the values. Passwords, tokens, and keys are encrypted, never in the code.
  8. Project check. Pushes to a working branch and runs check_project until no errors remain.
  9. Run. Puts the dataset in a group, chooses a runner with you, runs check_automation, and runs on the working branch. The first run of a repository on a runner downloads the code and prepares the environment, and can take a few minutes.
  10. Fixes. If it fails, reads the log, the evidence, and the run details, fixes the code, pushes, and runs again.
  11. Delivery. With everything passing, you decide whether to merge the working branch. Then, if you ask, the assistant schedules a recurring run.

Check before running​

Two tools tell you, before the run, what would make it fail. They also work for existing projects: ask "check the project for the Portal system" or "check whether the Sales order automation is ready to run".

check_project reads the repository through the git integration, without writing anything, and compares the code with the project contract and with what's registered in Proton:

  • the file the runner uses to recognize the project (pom.xml or pytest.ini) and the Proton entry point;
  • the declared Proton Cloud SDK and the minimum version;
  • the file for each registered component, and component files that aren't in Proton;
  • parameters the code reads or returns that the component doesn't have, including ones with different capitalization;
  • passwords or tokens committed to the repository.

check_automation checks whether the datasets are ready: each step with its component, system, repository, and integration; the input values; references to outputs and variables; unencrypted secrets; the environment; an available runner; and the dataset in an open cycle.

Both separate errors (what makes the run fail or give a wrong result) from warnings (what deserves a look).

What Proton doesn't do​

The connector never writes to your git repository: it's the assistant that writes the code and pushes it, on your machine. Merging the working branch into the default branch is always your decision.