Changelog

Friendlier first runs

• —GitHub blocks Actions from opening PRs by default — the bootstrap error now names the branch your suite was committed to and points at the exact setting to flip. Nothing is lost.
• —Documented the one-time repo setting in the README.

Bootstrap works on repos with zero test setup

• —The exact audience bootstrap mode is for — repos with no tests — usually has neither @playwright/test nor a config. GreenCI now installs both on the fly and writes a minimal playwright.config.ts with your base-url.
• —The bootstrap PR includes a one-time local-setup section when GreenCI had to install things for the run.

Bootstrap mode — from zero tests to a passing suite in one run

• —Describe your critical flows in plain English; GreenCI reads your real pages, writes a spec per journey, runs and heals them against your app, and opens a PR with the verified-passing suite.
• —Journeys behind login get Playwright storageState scaffolding with TODOs.
• —Destructive or costly steps (real payments, emails) stop at the confirmation step by design.

Run results in the dashboard

• —Every run now reports pass/fail counts, self-heals, and suspected app bugs (with verdict reasoning) to your GreenCI dashboard.
• —max-retries is clamped to the API's cap so extra retries can't error.
• —Generated filenames are normalized — no more accidental e2e/e2e/ nesting.

First stable release

• —AI test generation for pull requests: analyzes the diff, writes Playwright specs, runs them, and commits only what passes.
• —Honest self-healing: every heal is classified test-issue vs app-bug-suspected. Suspected real regressions are never healed over — they're flagged on the PR with evidence.
• —Bring your own LLM: run generation on your Anthropic or OpenAI key, directly from your runner — your code never touches the GreenCI API.
• —Cypress → Playwright migration mode.