v0.37.0-preview.0
Gemini CLISummary
This release introduces several new capabilities, including experiment-gated support for Gemini Flash 3.1 Lite and a new skill for detecting duplicated code during PR reviews. Key fixes stabilize the browser agent's domain restrictions and resolve various CLI issues like flickering in copy mode and TTY hangs in headless environments. Additionally, the sandbox manager sees improvements with OS-specific path handling and dynamic expansion on Windows and Linux.
New Features
- Added support for experiment-gated use of gemini flash 3.1 lite.
- Introduced a new skill specifically designed to look for duplicated code during PR reviews.
- Implemented forbiddenPaths configuration for OS-specific sandbox managers.
- Enabled dynamic discovery of read-only tools within the browser agent.
- Added support for inline agentCardJson configuration for remote agents.
- Implemented Windows sandbox dynamic expansion (Phase 1 and 2.1).
- Added a 'ci' skill for automated failure replication.
- Enabled dynamic Linux sandbox expansion and worktree support.
- Added reliability harvester and 500/503 retry support for evaluations.
- Added development worker infrastructure for GCP integration.
Bug Fixes
- Fixed an issue where the copy mode experienced flickering and cursor resets.
- Fixed the browser agent bypassing domain restrictions by incorrectly detecting embedded URLs in query parameters.
- Fixed the browser agent's domain restriction system prompt lacking a proxy bypass constraint.
- Fixed the shell tool conditionally exposing additional_permissions.
- Fixed the escape key prematurely cancelling requests when in shell mode.
- Fixed the CLI showing the updating spinner even when auto-update was not in progress.
- Fixed the CLI hanging on headless environments by ensuring process.stdin is resumed before React Ink launches.
- Fixed the ask question dialog not taking the full window height in the CLI.
- Fixed an issue where chat logs and errors did not handle tail tool calls correctly.
- Fixed the CLI prioritizing non-primary name matches during slash command search.
- Fixed the plan mode incorrectly switching the model to a flash model after exiting plan mode.
- Fixed an issue where the agent loader produced poorly formatted errors when paths were empty.
- Fixed the UI cleanup of estimated string length calculations in the composer.
- Fixed an issue where the policy for write_file argsPattern in plan mode was too strict, preventing paths without a session ID.
- Fixed an issue where non-interactive ASK_USER denials were used instead of explicit policy rules.
- Fixed an issue where the browser input blocker was not kept active across navigations.
- Fixed an issue where the model routing for gemini 3.1 pro to customtools model was incorrect.
- Fixed an issue where console logs/info were shown in headless mode when they should have been skipped.
Improvements
- Centralized test agents into test-utils for reuse.
- Reverted the disabling of agents by default in configuration.
- Improved telemetry attribute keys and added timestamps for plan updates.
- Refined onboarding metrics to explicitly log duration and use the tier name.
- Reduced logging verbosity by adding toJSON to tools and invocations.
- Moved a flaky ctrl-c-exit test to a non-blocking test suite.
- Improved error types in telemetry by stripping leading underscores.
- Clarified policy requirements for general.plan.directory in the settings schema documentation.
- Split the tools table into category sections in the reference documentation.
- Fixed grammar in CONTRIBUTING documentation and numbering in sandbox documentation.
- Added a note about root privileges in sandbox documentation.
- Documented agent_card_json string literal options for remote agents.
- Fixed over-eager slash subcommand completion in the UI.
- Fixed UI issues related to preventing over-eager slash subcommand completion.