New
- Choose when agents commit in Settings → Git and editor. Keep a commit per checked slice, make one before Verify, or review the uncommitted diff and decide whether to commit it yourself.
- Answer a terminal question directly in the App. Routine choices take one click. Amendments and spending decisions ask for confirmation, and the next question still follows your terminal preference.
- Open files and folders in a custom editor. In Settings → Git and editor, choose Custom editor, enter its program and arguments, then save.
- Run a task check or record a slice commit in another checkout named in the approved Plan. Records → Checks shows where the check ran and which commit was recorded.
- Use models from a provider you configured in GitHub Copilot. Add the model as
provider/modelin Settings.
Fixed
- Verify shows owner decisions and evidence requests separately. It shows a native question as presented only after the agent’s matching picker appears.
- Antigravity appears after Claude Code and Codex in agent selectors. Model controls show the names and effort levels reported by Antigravity, while preserving the IDs used to launch it.
- Verify uses one committed range for the changed-file count, code viewer and reviewer brief. Fixes made after the registered slices appear in the review.
- Verify opens a fresh review after a recorded branch rewrite replaces a task commit.
- Verify keeps App questions open while work continues. The instructions tell agents to finish independent work before opening the terminal picker.
- A stopped planning session shows its status and a continuation action in the sidebar.
- Revising an approved Plan prompts for a fresh approval. Waiting for a question answer returns the answer, and the native picker keeps the full question, including cost and timing.
- After Plan approval, the App shows the Plan size, ways to start the build and a prompt you can copy into another session.
- Claude Code shows a permission request while the tool is waiting and restores the task count when work resumes.
- Artifact inspection shows the source path for a task draft. Screenshot captures report when a long page exceeds the image height limit.
- OpenCode gives Ollama up to 15 minutes to send response headers in the App and in your terminal. Your explicit OpenCode timeout takes precedence.
- OpenCode starts models that offer no effort levels. Models with effort levels keep your selection.
- When OpenCode runs outside the App, new messages and tool results appear in the open Conversation without reloading.
- Answer native Codex 0.171.0 questions when Codex runs in the App.
- The Convert to task dialog keeps its controls visible with a long Chat request.
- Antigravity shows setup and workspace trust in Terminal before the App sends a message. If a send goes unconfirmed, copy your text or open Terminal to continue.
- Resuming an Antigravity session keeps your selected model when the client has no newer selection to report.
- Finished outside Chats show Continue it here after the turn ends.
Upgrading
- If you use QualityLayer 12.x, run
qualitylayer update --version 14.3.3once to move to this release.