Libretto PR Agents

Libretto PR Agents watches your Playwright tests and, when one fails, investigates the live page to figure out what changed. It then opens a GitHub pull request with proposed code fixes. The goal is to automate the tedious maintenance work that comes with browser automation scripts, especially when upstream UIs change without warning.

The problem it tackles is familiar to anyone maintaining end-to-end tests. A button gets renamed, a selector path shifts, a form adds a new field, and your test suite breaks. You get a failure notification, dig through logs, compare the current page to what the test expected, write the fix, and push a PR. Libretto automates that entire chain. When a test fails, the agent examines the actual page state, figures out the discrepancy, and proposes a correction without you having to investigate manually.

It integrates with your existing Playwright setup without requiring changes to fixtures, retries, logging, or deployment. You add the libretto-playwright-debugger package, and the agent activates only at the failure boundary. Your existing error handling still governs the current run, meaning the agent doesn't try to recover mid-test. It just handles the aftermath by creating a PR for review.

The tool works with local browsers, self-hosted infrastructure, and cloud browser providers. You bring your own LLM provider API keys, so the AI analysis runs through whatever model you prefer. Libretto itself doesn't charge for the PR agent, though your model provider and browser infrastructure have their own costs.

The target audience is QA teams and developers who maintain Playwright-based browser automations at scale. If you have dozens or hundreds of tests that break occasionally due to UI changes, the manual triage adds up. Libretto shifts that burden to an automated agent that can handle the investigation and propose fixes while you review the PRs.

The Playwright debugger package is open source under MIT license. Documentation, the npm package, and a Discord community are available for support. Currently, Libretto focuses exclusively on Playwright. Adapters for Selenium and Puppeteer are not yet available.

Where Libretto differs from other test maintenance tools is the action it takes. Many tools help you debug failures or rerun flaky tests. Libretto goes further by creating the fix as a pull request, reducing the loop from diagnosis to correction to review. You still approve the change, but the grunt work of investigating and writing the patch is done.