An LLM reads your failed CI runs so you don't have to.

A Playwright reporter that classifies every test failure with an LLM and posts a short, human-readable summary to stdout, a GitHub PR comment, or Slack. A red build stops being a pile of stack traces and becomes four labelled buckets you can act on.

// playwright.config.ts
export default defineConfig({
  reporter: [['list'], ['playwright-ai-triage']],
});

One line in your config, ANTHROPIC_API_KEY in your CI env — that's the whole setup.

Failures a script can decide never reach the API at all: passed-on-retry, pure network signatures, and expired credentials are classified locally, for free. The model is reserved for the failures that actually need judgment.

• Self-hosted. There is no hosted platform behind the reporter. Your results are processed inside your own CI.
• Bring your own key. You supply an Anthropic API key; the per-run cost is printed honestly, and it lands around a cent.
• No vendor between you and your logs. Redacted failure text goes to the Anthropic API and to the outputs you switch on — plus your own endpoint, if you opt into the HTTP sink. Every one of those is a destination you control. Screenshots, videos, trace files, and your source code are never uploaded.
• It never fails your build. No key degrades to a plain summary; an API outage marks failures UNCLASSIFIED; any internal error is a warning and the run exits normally.

Directional, not a promise — the full list lives in the roadmap.

• Cross-shard merge — one combined summary instead of a section per shard.
• Bedrock, Vertex, and gateway baseURL support, for teams that reach Claude indirectly.
• GitLab merge-request comments — stdout and slack already work on any CI.

A hosted add-on with cross-run history and flakiness trends is on the later, if there's demand end of that list — not shipped, and not promised. The reporter stays self-hosted and free either way. If you'd use the hosted piece, the waitlist on flaketrace.com is the way to say so.

Docs and setup · Roadmap · Contributing · npm · flaketrace.com

Issues and PRs are welcome. If something on the roadmap matters to you, an issue saying so moves it up.

_{From the same maintainer: Agentic OS — governed AI agents for your repo, with a setup guide for QA engineers.}

_{Playwright is a trademark of Microsoft Corporation. This project is community-built and is not affiliated with or endorsed by Microsoft.}