volatileElementIds and noSelectorFailover can now be set in a Playwright project's metadata, so apps whose element IDs change on every render record stable selectors in every test.
Runs from the HTTP testing plugin now show each request and response, with highlighted JSON, polling summaries, and redacted secrets, instead of a browser-shaped layout.
Every shard of a CI run now publishes its fixes into one heal pull request, GitLab gets merge requests with no token setup, and stale heal pull requests close themselves.
The mobile plugin now matches web tests on failure triage, auto-heal, AI-step caching, and evidence, and can reuse one device session across tests to cut device-farm setup time.
With --auto-heal, an AI coding agent now diagnoses each CI failure, fixes the tests that are out of date, has Donobu re-run them, and opens a pull request with verified results.
Screenshot localization reviews now flag unit spacing, locale-specific punctuation, capitalization, and unnatural copy, and always explain findings in English.
Replayed clicks recover when a grid re-renders the target, never fire twice when a click opens a dialog, and no longer stall when a step runs right after a navigation.
page.ai.proofread() checks visible page text, a PDF, or any string for spelling, grammar, and brand-name errors, and page.extractPdfText() reads PDFs, including scans, with OCR.
When a sharded CI run is merged with playwright merge-reports, the Donobu HTML, Markdown, and Slack reports now show auto-healed tests as healed instead of failed.
page.checkLocalization() reviews what a visitor actually sees: it OCRs a screenshot, checks language, translation, and formatting, and shows each finding highlighted on the image.
Localization sweeps now ignore screen-reader-only text, detect languages more accurately, and can ask an AI model to separate real missed translations from names, part numbers, and reviews.
Adding a provider in Studio now pre-selects each provider's newest models, including GPT-6, Claude Opus 5.5 and Sonnet 5, and Gemini 3.8 Flash.
donobu explore runs the same read-only exploration as Studio's Explore button from a terminal, and donobu context ask answers questions about the app with citations to the pages it saw.
One command configures the Donobu registry, installs @donobu/test with your package manager, keeps .env out of git, and installs the agent skill for your coding agent.
Test ideas Studio proposes now carry preconditions, steps, expected results, and a priority, and Accept turns one into a test case on donobu.com that you can automate from the same screen.
When a reviewer classifies a failure on donobu.com, later runs that fail the same way inherit that verdict directly, with the reviewer's name and date, instead of asking the AI again.
A new donobu cases command family lists, creates, updates, and links test cases from the terminal, so a coding agent can keep the case inventory in sync while it writes tests.
A first-party plugin adds an HTTP target, request.call/assert/extract/generate/poll on Playwright's request fixture, request.ai, and an agent skill installed via donobu skills install.
runs show and results list --run accept a CI run URL or provider run ID, and runs list shows the CI run and commit behind each Donobu run.
While touring, explore records concrete facts like prices and plan names, then flags when the same fact reappears with a different value elsewhere, quoting both sides.
Entitled accounts can mirror solved page.ai cache entries to Donobu Cloud so teammates and CI replay without re-calling a model; local-first remains the default.
Each recorded step's selectors get a background AI review during the act path, then validated answers rewrite the recording at flow end. Heuristics stand if the review cannot complete.
Element discovery now includes Kendo and Syncfusion EJ2 controls, native date/time inputs set values atomically, and an opt-in Tab commit flushes masked and numeric widget fields before save.
donobu session start lets an external coding agent (or a human) drive a real browser one step at a time and leave behind a deterministic @donobu/test spec, with replay-until-failure and amend to close the debug loop.
Install a versioned Agent Skills file with donobu skills install so a coding agent working in your repo gets grounded guidance on writing and running Donobu tests instead of guessing at the API.
Locale sweeps now flag short untranslated strings like menu items and buttons, not just full sentences, and correctly handle sites that keep locales under a path prefix.
page.ai.within(subject) scopes an AI assert, extract, or locate to one region of the page, so a matching phrase elsewhere on the page can no longer cause a false pass.
When a failure is not a clean call, AI triage now runs a bounded investigation, gathering evidence such as prior AI-check results, tool call history, browser logs, and git history before it settles on an explanation.
Define environment variables once and reference them anywhere in a flow, including AI-driven locate and assert steps, now with autocomplete when you're typing a reference.
Gemini models sometimes describe an action in plain text instead of issuing a structured tool call; Donobu now detects and recovers those cases instead of stalling the run.
A new run mode where the agent proposes each action and waits for your approval before executing it, with the ability to switch between supervised and live execution mid-run.
The control panel can now trigger specific tools (like Audit and Assert) directly, and pass a tool-call directive into an in-progress flow.
Set DONOBU_RUN_ID once in CI and every flow produced by that run, across parallel shards and workers, gets grouped under a single identifiable test run.
The HTML test report now supports substring search, filtering by tag, failure mode, or triage status, one-click copy for tags and file paths, and higher-resolution screenshots.
AI failure-triage now runs and reports on its own (grounded in the actual end-state screenshot) instead of only being available as a side effect of an auto-heal attempt.
Sign in to Donobu Studio with your Donobu account and your test runs, flows, and videos sync to Donobu Cloud automatically, so the rest of your team can see results in one place.
Group related tests into named suites you can create, browse, and bulk-manage from the sidebar, with the same sorting, search, and pagination you already use on the main tests table.
Donobu now drives native mobile apps through a first-party Appium-based plugin. Upload an Android .apk or iOS .app build and author flows the same way you would for the web.
Fixed a bug where enabling self-heal in a repo with no playwright.config.ts crashed the re-run step outright, instead of attempting a fix.
The AI Models list now includes Claude Haiku and the newest GPT-5.2 models, giving teams a faster, lower-cost option for high-volume test runs.
