Changelog

• Keys typed under a prompt stay there. In the interactive CLI, keys typed while an approval prompt or a picker is open, a paste or Ctrl+V included, no longer turn up in your next message.
• The first run of testrelic init finishes. It no longer waits forever on Playwright's HTML report (PLAYWRIGHT_HTML_OPEN=never is set for that run). A first run still going after 10 minutes is stopped, with a note to run the command yourself to see where it waits. /init works the same way.
• A page in place of TestRelic is reported. When a proxy or antivirus page, a Wi-Fi sign-in page or the TestRelic web app answers instead of the API, a tool result is no longer counted as delivered. The CLI reports why, for example "the connection was intercepted before it reached TestRelic".
• Retried uploads keep their files. An upload retried after a server error still sends its screenshots, videos and traces.
• Performance runs are sent once. If a server error may have come after the run was saved, testrelic upload k6 and testrelic upload jmeter don't send it again; the error asks you to check the repo's performance runs first. A refusal that saved nothing (HTTP 429 or 503) is still retried.

Released as testrelic-cli 1.8.4.

In Ask AI → Remote, code folds into compact cards, and opening one shows the code and its changes in the Artifact pane. Web chats are unchanged.

• Long code folds into a card. A code block of more than three lines becomes a one-line card with its language or file name and line count, and Open shows it in the pane. Shorter blocks stay in the chat with line numbers, Wrap and Copy, plus, on a test file, Save & run, which writes that request into the composer.
• A card for every file the agent touched. Each file it saved or deleted on the device gets a card with its path, lines added and removed, and a create, update or delete badge. A git diff it ran gets a card too, without a badge.
• Code and Changes. The pane shows the code with syntax highlighting, and what changed as a diff with old and new line numbers. A save is compared with the last copy of that file the chat saw. When there isn't one, for example because the chat never read the file or the read stopped at Studio's 64 KB limit, the pane opens on the code and says why.
• Step through the chat's code with the previous and next arrows, and copy or download it. Ask about this code puts a question about it in the composer.
• Review a save before you approve it. When the device asks before saving a file, Review changes on the approval card opens the diff in the pane, where you can answer Yes, run it or No, skip this. To tell it what to do instead, answer on the card.

Studio and the CLI need no update. See Ask AI.

tr_diagnose_run could list a different number of failures from the run's own total, which it prints right beside them. It now matches.

• A test that passed on retry isn't a failure. Each test is counted once, by its final attempt, so a test that failed and then passed is flaky, as the run says, not failed.
• No failures go missing on large runs, and steps with no test attached are no longer counted as extra failures. A failure past the run's first 1,000 steps is counted with its message, but without a stack trace.
• The same list feeds the other tools that read a run's failures: tr_compare_runs, tr_search_failures, tr_ai_rca, tr_create_jira, tr_heal_run and tr_replay_failure.

Released as @testrelic/mcp 3.4.7.

A round of platform fixes.

• Sign-in, sign-up and password-reset pages work better with screen readers. Each page declares its language and has a real page heading, and links inside text are underlined instead of marked by colour alone.
• Large runs upload. A run of about 2,850 tests or more, such as a large CTRF report, failed to upload, and from about 2,600 tests the repo's list of tests stopped updating. k6 and JMeter uploads from CLI 0.100.0 to 0.109.0 (testrelic upload k6, testrelic upload jmeter) failed too. All of these work now.
• Signing in a device at the token limit. Each user can hold 10 MCP tokens. Signing in Studio or the CLI with all 10 in use could be refused. Now the sign-in revokes the least recently used token that Studio, the CLI or onboarding created for a machine, and that machine has to sign in again. Tokens you created and named yourself under Settings → MCP Tokens are never revoked. If only those are left, the device says the limit was reached and asks you to revoke one. Expired tokens no longer count towards the limit.
• Steps no longer appear twice in runs uploaded with the upload strategy set to both.
• Ask AI: Branch from here and feedback on an answer work as soon as it finishes, without reloading the page.
• Amplitude: errors include Amplitude's own reason instead of a bare "400", and keys Amplitude rejects now show the integration as not valid.
• API: a malformed ID in a /repos/:repoId/… or Ask AI conversation path returns 404 instead of a server error. Repo routes that answered 400 VALIDATION_ERROR for a malformed repo ID now answer 404 REPO_NOT_FOUND, the same as for an unknown one.
• Ask AI apps: connecting an app that isn't in the catalog now returns an error instead of creating a connection.

Two Studio tools that had stopped working are back.

• Bash terminals start. Since 0.91.0, a terminal running bash closed before showing a prompt ("Git Bash closed. It stopped with exit code 2"), and Restart failed the same way. That covered Git Bash on Windows and, on Linux, the default terminal wherever your shell is bash. Bash terminals now open at a prompt, with Studio's command blocks and failure summaries. PowerShell and zsh terminals weren't affected.
• DevTools connects. On recent Chromium versions, Open DevTools in the Live Browser showed its tabs with nothing in them. It now shows the page's DOM and styles, its console (including messages from clicks you make in the live view) and its network requests, docked to the bottom or the right.
• Undock works. Undock into a separate window closed the panel and opened nothing. It now opens DevTools in your default browser.

Released as testrelic-cli 1.8.3.

Ask AI → Remote now reads like Studio's own chat, and its composer has Studio's options. Web chats are unchanged.

• New Session takes New Chat's place on the Remote tab and opens the start-a-session dialog. The Chats tab keeps New Chat, and the collapsed sidebar's + follows the tab you're on.
• Only what was typed. A message sent from Studio or the CLI shows the person's own words. The instructions those apps add to it, such as the reply language, the project rules from TESTRELIC.md and Studio's notes, are hidden. Attached files and recorded steps show as chips under the message.
• Studio's layout. The header says where the agent works (for example "runs and edits on device in repo"). Each answer is headed TestRelic with how long it worked, and consecutive tool calls fold into one "N actions · click to expand" line.
• Approvals as in Studio: Yes, run it, No, skip this or No — tell it what to do instead, with the keys 1–3, y and n, the arrow keys, Enter and Esc once the card has focus.
• Plan first. The permission pill switches between the device's own approvals (for example "Studio's setting") and ⏸ Plan first: click it, press Shift+Tab or type /permission. Under Plan first, that turn gets only the device's read-only tools (reading files and folders, the docs, and git status, diff, log and branches), so it plans and changes nothing. Write tests freely and Full autopilot are still chosen on the device.
• The + menu attaches text files, up to 12,000 characters each, or references a test or run with @. / lists /permission, /model, /effort and /memory, plus report, dashboard, plan, deck, compare and alert prompts.
• Chat settings choose the model and reasoning effort for the messages you send, and Memory marks one message to be remembered. Remote sessions always answer with TestRelic Cloud.

Studio and the CLI need no update. See Ask AI.

The changes chip in the agent chat header, for example 2 files · 1 run · 1 commit, now opens its panel and counts every engine's work.

• Click it and the panel opens beside the header, instead of being cut off inside it.
• Every engine's edits count. Files written by Claude Code, Cursor and Codex are counted, not only the TestRelic agent's, so a Claude Code chat no longer stays at "No changes yet".
• Commits count from any shell, PowerShell included, and so does a commit you approved on a git card.
• Each file says where it stands now: still changed, with its +/− counts and a diff (a new file's diff shows its content); in one of this chat's commits, which opens in History; or "no changes now".
• The footer says what's left: how many files aren't committed yet, with Review & commit, or "Nothing from this chat is left to commit."

Released as testrelic-cli 1.8.2.

A Remote session, the Studio, CLI or runner conversation you follow in Ask AI → Remote, can now be shared with other members of your workspace.

• Share from the session header. The session's owner gets Share, next to the people who have access and who's watching now.
• Two levels of access. Add members by name or email and give each Can view or Can collaborate, or open the session to Everyone in your workspace, who can view it. Copy link gives a link that opens only for members with access.
• Viewers see the live session, its transcript and its artifacts. They can Request to collaborate, and the owner can Allow or Dismiss the request in the Share dialog.
• Collaborators can send messages. A collaborator's message runs on the owner's device with the owner's tools and permissions, and the transcript shows who sent it ("Sent by … from the web"). Collaborators can change only their own queued messages.
• What stays with the owner: approvals, Stop this turn, End session, and changes to the conversation itself, such as renaming or deleting it.
• Finding what's shared with you. A notification appears when someone shares a session with you, and the Remote tab has a Shared with me section. Ended shared sessions move to Archived. Remove from my list hides one without changing anyone's access. If you haven't seen a share in the app within 5 minutes, you get one email about it.
• Workspace members only. Only members of the same workspace can be given access, and someone who leaves the workspace loses it.

Studio and the CLI need no update. See Ask AI.