Libretto turns failed Playwright runs into PRs

A failed Playwright script is usually where browser automation work stops. Libretto wants to make that the starting point instead: its new PR agent investigates the live page after a failure and opens a GitHub pull request with a proposed code fix.

The pitch is straightforward. Teams keep their existing Playwright automations and only add a repair loop for failures. In Libretto’s example, the agent spots that a selector is wrong and proposes changing:

• input[name=“username”] to input[name=“login”]

According to the demo, the agent confirmed the fix by inspecting the live page, then generated a PR with the update.

Recommended reading

Avito eyes free dating inside its app

Libretto says the agent does not replace a workflow’s existing failure handling for the current run. Retry logic, fallbacks, catches, and error handling still deal with the immediate failure. The PR agent is aimed at diagnosing what changed and proposing a code fix for future runs.

The company also says you do not need to use the Libretto runtime. Developers can add libretto-playwright-debugger to an existing Playwright project, initialize the debugger once, and call debugFailure() from the failure path. Existing runtime choices, browser providers, deployment setup, and workflow structure can stay the same.

Right now, support is limited to Playwright. Libretto says Selenium and Puppeteer are not supported yet because they would need separate adapters. For browser infrastructure, the company says the agent works with local, self-hosted, and hosted browsers as long as the automation has a live Playwright Page and keeps it open while debugFailure() runs.

Libretto says it does not charge for the PR agent itself. Users bring their own model provider API key and browser infrastructure, so any costs come from those providers instead. The Playwright debugger package is also open source under the MIT license in the Libretto repository.