Changelog
What changed in SpecHawk, release by release.
Release 3214 November 2026
Repairs ask before moving a test's first page
- ImprovedWhen a repair would send a test to a different first page, it now asks you first and doesn't save the new page on its own. Before, a test could go green on the wrong page.
- FixedSecurity and privacy fixes.
Release 3113 November 2026
The Hawk counts failing tests like your dashboard
- FixedThe Hawk gives the same failing count as your dashboard, for each environment. When it has only seen the last few runs, it says so at the start of its answer and doesn't give a total.
- FixedThe Hawk no longer runs two sentences together, and no longer leaves stray asterisks around italic text.
- FixedSecurity and privacy fixes.
Release 3013 November 2026
A rebuilt Monitoring and smoother motion
- ImprovedMonitoring is rebuilt. Five tiles show what's watched, what's waiting, the last check, what failed in the past 7 days and competitor updates, with a card for every environment, filter chips and one feed of changes with thumbnails.
- ImprovedPanels, tabs, menus and slide-outs move more smoothly.
- ImprovedThe welcome email has the new design.
- FixedA repaired test goes back to needing verification until its new steps pass again.
- FixedSecurity and privacy fixes.
Release 2911 November 2026
The home page tells the story
- NewThe blog has posts with their pictures, and sources for the numbers in them.
- ImprovedThe home page has a short scene that shows the four steps: record once, vary, check and report.
Release 2810 November 2026
Dropdowns that match, and a clearer repair screen
- ImprovedIn Chrome, every dropdown, including the ones in forms, opens our own menu, and its label stays on one line.
- ImprovedA visual check with no AI note or How it looks score says why, and what to do next.
- ImprovedThe repair screen says Possible bug in your app, and lists proposals you can Apply or Keep.
- ImprovedThe home page opens with a new headline: Move fast. Break nothing.
- FixedIn a visual check group, Report one bug covers the same pages Approve does.
- FixedThe bulk Delete on Runs, and three other buttons, are readable.
- FixedSecurity and privacy fixes.
Release 279 November 2026
A new website, a new look and verified tests
- NewA test is Verified only after three clean passes, and a check that was already true at the start doesn't count.
- NewA new SpecHawk website, with a changelog.
- ImprovedTests we write for you can run to 80 steps, up from 25.
- ImprovedClaude is the default AI on our own keys. Until your workspace picks a model, everything uses Claude Haiku 5.5, and Agent UI checks use Claude Sonnet 5.5.
- ImprovedA map judges a press by what it sends, not by its words, so a harmless Payments tab is explored. Tabs are explored like links, and a native confirm dialog is cancelled and listed under Not tried.
- ImprovedA new look across the app, with new colors, type and layout in the website's language. Every page stays where it was.
- ImprovedPages load faster. Replies are compressed, dashboard numbers are counted in the database, and the Hawk chat loads after the page, not with it.
- ImprovedVisual check pages load far less picture data, with lazy loading and thumbnails.
- ImprovedThe dashboard chart counts your calendar days, not UTC.
- ImprovedThe side menu has a Monitoring link.
- ImprovedThe privacy page lists every cookie we set.
- FixedButtons with an icon beside the label are found by their words, and a failed step shows the button's name.
- FixedAgent UI checks update the call count live, and a map no longer reports a false "didn't show the new name".
- FixedThe top bar's pill lists every test in progress.
- FixedSecurity and privacy fixes.
Release 266 November 2026
Visual checks and Monitoring
- NewVisual checks. Each check takes pictures of your key pages, compares them with the baselines you approved, and holds what changed for you to approve or report.
- NewHow it looks. Visual checks give each screen a score as an opinion, and it never fails a check.
- NewMonitoring. One page shows every environment's last check, what's waiting for you and competitor updates, with a status filter and links you can share.
- NewA weekly report for a sponsor: what we tested, what broke, what got fixed and the time saved, on a page you can print.
- ImprovedOne visual problem that shows on many pages is one card. Approve or report covers those pages, and the card says what it approves and what it leaves.
- ImprovedA visual check names each picture for its note, so a change is never read backwards. Times read in your time zone, a first look at a page isn't counted as a change, and each page card shows its approved baseline.
- ImprovedAgent UI check results show your own words instead of [masked], hiding only saved values.
- FixedA first visual check can be started on a mapped environment.
- FixedSecurity and privacy fixes.
Release 255 November 2026
Faster pages and plainer words
- ImprovedThe dashboard and the environments page stop waiting on their numbers, and a link's page starts loading when you point at it.
- ImprovedThe app fits a phone screen, and fonts load from our own domain, not Google's.
- ImprovedEvery page names its browser tab.
- ImprovedDashboard tiles say their time window, so Pass rate and Failing read together.
- ImprovedRuns wrap long step lines, show events in plain words and have a Copy run id button.
- ImprovedA test page's API checks show each field's type and what the call first answered.
- FixedSecurity and privacy fixes.
Release 244 November 2026
Clearer messages and steadier runs
- ImprovedA run says Replayed or Agent for how it ran.
- ImprovedAgent UI check results count only the versions a check ran on and say what we couldn't check.
- ImprovedEvery page's breadcrumbs start at its menu section.
- ImprovedMessages for a failed request say what to do next, and a form that can't save shows its own sentence.
- ImprovedWords across the app say verify, and old saved reasons are rewritten to match.
- FixedA run page that loses its live connection reopens it and keeps waiting, and live pages keep refreshing until the run is done.
- FixedA stopped run counts as stopped, not as a failure, on the dashboard, in the Runs filter and in the Hawk's weekly summary.
- FixedA repair no longer fixes a test against a page that answered with a 404 or a 500.
- FixedScrolling a long page no longer drags the whole app up with it, and a send-again replay finds your chat's box.
- FixedSecurity and privacy fixes.
Release 2330 October 2026
Steadier sign-ins and exported tests
- ImprovedA sign-in that leaves a session cookie keeps it, so the steps after it stay signed in.
- ImprovedButtons inside web components are found by the text they show.
- ImprovedTests you export sign in on the page you saved.
- FixedSecurity and privacy fixes.
Release 2229 October 2026
Production hosts kept out of every job
- ImprovedThe production hosts you list are now kept out of every job, sign-in included.
- ImprovedSpecHawk never presses a control it couldn't read, and a map only presses the controls it checked first.
- ImprovedA map's time limit starts after sign-in, and map failures are explained in plain words.
- ImprovedSign-in status is more accurate. A cookie session that's still live isn't called expired, and a sign-in that didn't hold shows as signed out.
- ImprovedAsk the Hawk has a fair-use limit per person, and a question it turns down doesn't count against it.
- ImprovedUpdated privacy policy and terms.
- FixedSecurity and privacy fixes.
Release 2119 October 2026
Agent UI checks
- NewAgent UI checks. Record a conversation with your chat agent, and we check what it draws, version by version, with results in plain words and a certificate for the agent.
- NewThe activity record and the bell show what changed across your account.
- NewA welcome email when you sign up.
- ImprovedSchedules can run every 15 minutes.
- ImprovedSteps that type a password-like value show it masked, and a new test has "Verify it now".
- ImprovedA saved sign-in is checked before a job relies on it, and signed in again once if it has ended.
- ImprovedPublic pages load again on maps, and a page with only a password field isn't mistaken for a sign-in page.
- FixedStop reaches a run that's already going, a stopped run says so, and Stop never ends a finished run.
- FixedA 403 no longer signs you out.
- FixedSecurity and privacy fixes.
Release 2014 October 2026
Verify a test before you rely on it
- NewVerify a test. Run it once to show it works before you rely on it.
- NewA new default AI model, with six more open models to choose from.
- ImprovedEvery action waits until the page is ready, and controls inside shadow roots are found.
- ImprovedMore maps run at once, and a queued map says why it's waiting.
- ImprovedHelp when a site's bot protection stops a map.
- FixedThe test list starts again from the top when you change the Source filter.
Release 196 October 2026
Faster for big accounts, fair turns for everyone
- NewPlain-English steps, written with made-up test data.
- NewAgent certification: repeat runs, a risk rating and a certificate.
- ImprovedThe Security home, dashboard and suite list are faster for big accounts, and long lists page.
- ImprovedScheduled runs and short tests start on time while maps run, and no one account takes every slot.
- ImprovedRuns start as soon as tests are generated, and queued work says its place in line.
- ImprovedMaps find controls in Lit components, work across apps on several hosts, and handle hover menus.
- FixedA page that stops answering no longer holds up the map.
Release 184 October 2026
Jobs that recover on their own
- ImprovedJobs have time budgets, and a job that stalls recovers on its own.
- ImprovedEach environment's Security page leads with "Fix these first": the three fixes that give back the most points.
Release 172 October 2026
Findings that carry their evidence
- NewPage health shows console errors and failed requests on each page.
- NewPlain-English tests written from the map's pages and controls, page by page.
- NewEach environment gets a readable sign-in inbox address, with a copy button.
- NewYour site's own icon shows for each environment.
- ImprovedFindings carry their evidence: what we saw, what it means and how to fix it. Grades count problems, not sightings.
- ImprovedThe Security pages are in plain words, and dark mode is easy to read.
- ImprovedSteps target exactly one element, and the words that stop a map pressing a button like Delete now cover ten more languages.
Release 161 October 2026
Repairs that say what changed
- NewSign-ins that email a code work in production, with each environment's own inbox.
- NewRegenerate repairs a test. It replays it, fixes the failing step and says what changed.
- NewThings that open, close and change now animate.
- ImprovedAd-hoc run and Ticket to steps have examples from your map, steps you can edit, and a mic.
- ImprovedThe map starts from wherever sign-in landed, and shows the site's tech stack.
- ImprovedRun counts agree everywhere. Test Cases shows each environment's ten newest tests, with Show all for the rest.
- ImprovedA search box the whole site carries counts as one workflow, and a test never depends on today's date.
- ImprovedTwo-step sign-ins move on in any language.
- ImprovedThe top bar shows the true totals of tests and maps in progress.
Release 1530 September 2026
More at once, and repairs for new tests
- NewExport a saved test to Playwright, from the button on its page.
- ImprovedYour environment's test browsers and one map run at once, and a queued run says it's waiting for a free browser.
- ImprovedA new test that fails its first run, or a workflow when it's saved, gets repaired.
- ImprovedWith form submissions off, a form's test fills the form and checks its submit button is ready.
- ImprovedWorkflows that are the same journey count once.
- Improved"Needs a fix" and "Failing" mean the same thing everywhere, and a pass only on retry counts as flaky.
- ImprovedA map that stops on bot protection says so.
- ImprovedAd-hoc run, Ticket to steps and Schedules match the look of Environments and Runs.
- FixedSecurity and privacy fixes.
Release 1429 September 2026
Sign in with an emailed code
- NewSign in with an emailed code, end to end, with each environment's own inbox.
- NewList your production hosts, and maps and runs stay off them.
- ImprovedRuns follow what the map found, and the map page lists the pages it didn't map, with reasons.
- ImprovedTest Cases filters by status, environment and source, like Runs.
- ImprovedThe dashboard greets you, shows your 10 most recent runs with Hawk insights, and its map cards open the map.
- ImprovedThe Security page shows the latest scan only, one row per issue with how many pages it's on.
- ImprovedIdentical workflow steps show once on the environment page.
- ImprovedNew summary tiles at the top of Environments, Runs and Test Cases.
- ImprovedA site's bot-check page stops the run with a clear reason.
Release 1326 September 2026
Single sign-on, one-time codes and suites
- NewSign in with your provider's button. Single sign-on names the provider and presses it for you.
- NewSign-ins that send a one-time code, and sign-ins with a saved session cookie.
- NewSuites. Run a group of tests together and see the results grouped by outcome, failing step and reason.
- NewEach test case has a page with every run of it, newest first.
- NewThe jobs on one environment share one sign-in. If your app doesn't allow that, turn on "Sign in separately for each test".
- ImprovedIf a sign-in doesn't go through, we tell you why instead of guessing.
- ImprovedCookie banners and consent pop-ups no longer stop a map or a run.
- ImprovedA run says which step failed and why, and a run that passed on a retry isn't reported as a failure.
- ImprovedA repaired step is only saved to the test once the whole run passes, and a check is never changed.
- ImprovedThe map page shows each page's status, why a page gave no workflows, and why the map stopped.
- ImprovedA run that's waiting says what it's waiting for.
- ImprovedForm submissions are allowed per environment, and form workflows aren't ticked by "select all".
- ImprovedSecurity findings are counted and scored by type, with the pages as detail.
- FixedThe sidebar no longer covers the page on a phone.
Release 1217 September 2026
Runs sign in the way maps do
- ImprovedA run signs in the same way a map does.
Release 1116 September 2026
Signed in means signed in
- FixedA sign-in that ends on your provider's password page is no longer called signed in.
Release 1015 September 2026
The environment page, side by side
- ImprovedThe environment page lays out side by side, and "tick all" is back.
Release 94 September 2026
Test changes you approve
- NewChecking tests after a merge. Turn it on for an environment, and after a merge we suggest changes to the tests it touched. They wait for your approval, and an approval never undoes an edit you made.
Release 81 September 2026
Re-scan on merge
- NewWith GitHub connected, turn on a re-scan on every merge for an environment, and we match the change to the tests it touches.
- ImprovedA run that stopped partway says so.
Release 729 August 2026
Choose what a map may click
- NewA list of what a map is allowed to click.
- ImprovedThe theme follows your system setting.
Release 628 August 2026
Maps that read your code
- NewWith a repository linked, we read its routes to map your app more completely.
- NewA map tells you when it can't read your site's language to spot risky buttons by name, such as Delete.
- ImprovedA redesigned Settings page.
- FixedThe Jira form no longer gets cut off.
Release 526 August 2026
Connect GitHub
- NewConnect GitHub. Install the app and link a repository to an environment.
Release 416 August 2026
More accurate maps
- ImprovedMore accurate maps.
- ImprovedTest variants and findings are grouped, and you can see what changed from one run to the next.
Release 310 August 2026
The real demo
- ImprovedThe homepage shows the real product demo, in your theme, light or dark.
Release 29 August 2026
New default AI models
- ImprovedNew default AI models.
Release 15 August 2026
SpecHawk's first release
- NewMap your app. Our agents explore it and find the workflows that matter: creating, updating, searching and filtering.
- NewTest cases written for you from the map, with realistic test data.
- NewRun your tests, with a pass or fail picture for every step.
- NewAd-hoc run. Describe a test in plain words and run it.
- NewTicket to steps. Turn a Jira ticket into test steps.
- NewSchedules. Run tests daily, weekly, monthly or yearly.
- NewSign in or sign up with Google.