●Change8

v0.29.0-nightly.20260205.a3af4a8ca

Gemini CLI
View on GitHubDocumentation
21 features12 fixes10 improvementscliextensionsmcpthemestools

Summary

This release introduces significant planning capabilities, including the new `/plan` slash command and tools to enter and exit plan mode. Key updates also include automatic theme switching based on the terminal environment and enhanced security measures like requiring user consent for MCP server OAuth. Several bug fixes address regressions in navigation and improve stability across various modes.

New Features

  • Introduced tool name aliasing for backward compatibility in the core.
  • Enabled automatic theme switching based on the terminal background.
  • Updated the slash command from `review-frontend-and-fix` to `review-and-fix`.
  • Implemented the `plan` slash command.
  • Added the `enter_plan_mode` tool.
  • Added markdown rendering support to the `ask_user` tool.
  • Added support for MCP configuration via admin controls (part 1).
  • Required user consent before performing MCP server OAuth.
  • Moved user identity display to the header for better visibility.
  • Added pluggable authentication provider infrastructure for A2A.
  • Enabled `getUserTierName` in the configuration.
  • Added default execution limits for subagents.
  • Added support for draft-2020-12 JSON Schema with lenient fallback.
  • Added A2A auth config types.
  • Enabled the `.agents/skills` directory as an alias for skill discovery.
  • Added core logic and the `exit_plan_mode` tool definition for planning.
  • Added the `exit_plan_mode` UI and prompt.
  • Enabled the ability to disable loop detection via a setting.
  • Added logic to handle inconsistency in schedulers within plan mode.
  • Enabled custom deny messages in plan mode policies.
  • Increased the `ask_user` label limit to 16 characters.

Bug Fixes

  • Fixed a regression affecting up/down arrow navigation and added a corresponding test.
  • Prevented content leak in `MaxSizedBox` bottom overflow issues.
  • Fixed an issue where the agent would get stuck at interactive commands.
  • Fixed an issue where the agent was not respecting the user's `.gitignore` preference.
  • Fixed environment loading and authentication validation issues when running in ACP mode.
  • Fixed an issue where the `ask_user` tool was incorrectly used in non-interactive modes.
  • Fixed an issue where detailed error messages were not prioritized during code assist setup.
  • Fixed an issue where Google environment variables (`GOOGLE_GEMINI_BASE_URL` & `GOOGLE_VERTEX_BASE_URL`) were not propagated correctly in the sandbox.
  • Fixed handling of empty settings.
  • Fixed an issue where folder trust was not enforced for workspace settings, skills, and context.
  • Fixed an issue allowing restricted `.env` loading in untrusted sandboxed folders.
  • Fixed an issue where an error message was shown in the `/plan` command.

Improvements

  • Updated the placeholder used for the "Other" option in choice questions within plan mode.
  • Improved the `Ctrl+R` reverse search functionality.
  • Refactored the connection logic to helper functions (no-op refactoring in IDE context).
  • Improved test isolation and enabled subagent evaluations.
  • Set the default maximum attempts to 3 and standardized the variable used.
  • Reloaded skills automatically when extensions change.
  • Added telemetry tracking for the rewind feature.
  • Unified the workflow location in the system prompt to optimize caching.
  • Removed the hardcoded policy bypass for local subagents.
  • Removed the unused `autoAccept` setting.

Related Documentation

Gemini CLI Documentation

mcptoolsgetting started