What pruvo does with your code
pruvo is a GitHub App that reads a pull request and writes manual test cases from it. That means asking for access to your source. This page says precisely what it reads, where that goes, and what is kept.
What it can access
The App requests three permissions, and no others:
- Pull requests — read and write. Read to see the change; write only to post the results comment on the PR.
- Contents — read-only. Needed to fetch the diff and the changed files at the commit being reviewed.
- Metadata — read-only. Mandatory for every GitHub App.
There is no write access to your code, no access to Actions, secrets, packages or administration, and no ability to push, merge or change a branch. pruvo only reacts to pull request events on repositories you have explicitly added to the installation.
What is sent to the model
Generation runs on Anthropic’s API. For each pull request, four things are sent: the diff, the list of changed files, the PR title and the PR description.
Your whole repository is never sent — only the change. Anthropic does not train on data submitted through its API.
What is kept
The generated test cases, and enough context to show them against the right pull request: PR number, branch, commit SHA, title, the plain summary, and each case’s pass or fail status once QA signs off.
The diff is not stored. It is fetched from GitHub when a pull request event arrives, sent to the model, and dropped — there is no column for it anywhere in the database. The same is true of file contents. If you want to verify that rather than take it on trust, the schema is one file.
Deleting it
Deleting a scenario removes its cases. Deleting a project removes its scenarios. Deleting an organisation removes everything belonging to it. Uninstalling the GitHub App stops all access immediately: pruvo can no longer read anything, because the installation token it uses ceases to exist.