1. Highlight

    Featured this release

    PR reviews prove bug fixes with a before/after reproduction

    The review agent reproduces the original behavior on your base environment before showing it resolved on the preview.

    For bug-fix pull requests, the review agent derives the failure from the fix itself, drives your app into the state that triggers it, and runs the same check against both your base environment and the PR preview.

    The review comment now opens with a Reproduction section showing the before/after result with screenshots, so a passing review means the behavior was actually confirmed and fixed rather than just a happy path run.

    # Direct link

  2. Update 02 of 08

    Jump straight from your PR list to the review conversation

    A new icon on the pull request list takes you directly to the review conversation.

    Rows with an associated review conversation now show a link icon in the pull request table.

    Click it to go straight to the conversation instead of opening the PR, scrolling to reviews and clicking through. More on pull request testing.

    # Direct link

  3. Update 03 of 08

    Knowledge rules open in a shareable side sheet

    Rules under Settings → Knowledge are now a sortable table, and each one opens in a linkable side panel.

    Click any rule to open it in a side sheet with its own URL, so you can share a direct link to a specific rule instead of describing where to find it.

    Editing happens inline in the sheet, and rules with JSON content are pretty-printed in a monospace block instead of rendering as a wall of wrapped text. See knowledge.

    # Direct link

  4. Update 04 of 08

    PR reviews now test your dependency updates

    Package and lockfile bumps are now tested against the flows they actually touch.

    When a pull request is primarily a dependency update, the review agent identifies which changed packages affect runtime behavior and tests the flows that exercise them, falling back to a core smoke pass when no specific surface is affected.

    Purely build and dev-tooling bumps still get a clear "unable to review", so a real behavior change dressed up as a routine bump gets covered instead of waved through.

    # Direct link

  5. Update 05 of 08

    Preview downloaded files without leaving the run

    The Download Viewer has a dedicated Preview button, so you can check a file an agent downloaded during a test.

    Every completed download in the Download Viewer now shows a named "Preview" button alongside the file, opening its contents inline.

    Fast downloads open the preview immediately; slower ones show a clear in-progress state until the file is ready. See file downloads.

    # Direct link

  6. Update 06 of 08

    Steer a running chat with ⌘Enter

    Press ⌘Enter (Ctrl+Enter on Windows and Linux) to have a message picked up immediately instead of waiting in the send queue.

    Messages sent while the chat agent is mid-turn used to wait for the whole chain to finish. Regular Enter still queues that way.

    ⌘Enter marks a message for immediate delivery: the running agent picks it up at its next step and works it into the response. A "⌘ + Enter to send now" hint appears next to the send button while a turn is in progress, and force-sent messages show as "Sending now" in the queue.

    # Direct link

  7. Update 07 of 08

    Faster todo triage: priority sorting and clearer rejection reasons

    Todos default to urgent-first order, and rejecting one takes a picked reason instead of a blank text box.

    The Todos list now sorts by priority automatically, with newest-first as the tiebreaker, so the most urgent items surface without any manual sorting.

    Rejecting a todo starts with a reason picker – including an "Other" option for anything custom – instead of requiring free-form text every time, which makes rejected items easier to scan and compare later.

    # Direct link

  8. Update 08 of 08

    Tracer timeline shows the step numbers your assessment cites

    Assessment findings reference "Step 5" – the tracer timeline now shows that same number, so you can find it at a glance.

    Each row in the test-run tracer timeline is now labeled with the step number the assessment report uses, and clicking a "Step N" badge in the report jumps to the matching row.

    Collapsed step cards show the range they cover, for example "#4–6", until you expand them.

    # Direct link

Your team moves fast. Can your testing keep up?

QA.tech agents test your product autonomously, so moving fast never means shipping broken. See how it works in a 30-minute demo.

Get a demo