Gemini CLI mcp
ComponentUpdates related to the mcp component of Gemini CLI.
All MCP Features
- Enabled showing plan content when using plan mode prompts.(v0.40.0-nightly.20260415.g06e7621b2)
- Enabled mouse clicking for cursor positioning within multi-line answers provided by AskUser.(v0.40.0-nightly.20260415.g06e7621b2)
- Enabled 'TerminalBuffer' mode to solve flicker in the UI.(v0.38.0)
- Added support for default values for environment variables in the CLI.(v0.38.0)
- Introduced a background memory service for skill extraction.(v0.38.0)
- Implemented context-aware persistent policy approvals.(v0.38.0)
- Enabled compact tool output by default in the CLI.(v0.38.0)
- Added a scrollbar for the input prompt.(v0.38.0)
- Implemented background process monitoring and inspection tools.(v0.38.0)
- Added an experimental setting: experimental.adk.agentSessionNoninteractiveEnabled.(v0.38.0)
- Enabled the 'Other' option for the yes/no question type.(v0.38.0)
- Introduced ContextCompressionService in the core.(v0.38.0)
- Scoped subagent workspace directories via AsyncLocalStorage.(v0.38.0)
- Implemented a high-signal PR regression check for evaluations.(v0.38.0)
- Migrated nonInteractiveCli to LegacyAgentSession.(v0.38.0)
- Added minimalist sandbox status labels in the CLI.(v0.38.0)
- Enabled browser agent metrics.(v0.38.0)
- Prioritized summary for topics in core and CLI.(v0.38.0)
- Changed default loadingPhrases to 'off' to hide tips in the CLI.(v0.38.0)
- Added the /memory inbox command for reviewing extracted skills.(v0.39.0-preview.0)
- Introduced user confirmation required for activating skills in Plan Mode.(v0.39.0-preview.0)
- Refined shell tool description display logic.(v0.39.0-preview.0)
- Refined tool output formatting for compact mode.(v0.39.0-preview.0)
- Implemented the useAgentStream hook in the CLI.(v0.39.0-preview.0)
- Wired up useAgentStream in AppContainer.(v0.39.0-preview.0)
- Migrated chat recording to JSONL streaming format.(v0.39.0-preview.0)
- Refactored the subagent tool to use a unified invoke_subagent tool.(v0.39.0-preview.0)
- Added Strategic Re-evaluation guidance to the system prompt.(v0.39.0-preview.0)
- Added a large memory regression test.(v0.39.0-preview.0)
- Enabled support for Ctrl+Shift+G input.(v0.39.0-preview.0)
- Added an optional debugRainbow setting for debugging UI.(v0.39.0-preview.0)
- Persisted subagent agentId in tool call records.(v0.39.0-preview.0)
- Added support for skill patching with /memory inbox integration.(v0.39.0-nightly.20260414.gdaf500623)
- Enabled tool-controlled display protocol implementation (Steps 2-3) for agents.(v0.39.0-nightly.20260414.gdaf500623)
- Supported authentication block configuration in MCP servers for agents.(v0.39.0-nightly.20260414.gdaf500623)
- Implemented silent fallback mechanism for Plan Mode model routing.(v0.39.0-nightly.20260414.gdaf500623)
- Introduced decoupled ContextManager and Sidecar architecture.(v0.39.0-nightly.20260414.gdaf500623)
- Added Strategic Re-evaluation guidance to the system prompt for improved reasoning.(v0.39.0-nightly.20260411.0957f7d3e)
- Enabled persistence of subagent agentId within tool call records.(v0.39.0-nightly.20260411.0957f7d3e)
- Added the "/help" command to the ACP interface.(v0.39.0-nightly.20260408.a39461718)
- Introduced Agent Protocol UI types and enabled them via an experimental flag.(v0.39.0-nightly.20260408.a39461718)
- Enabled using experiment flags to control default fetch timeouts.(v0.39.0-nightly.20260408.a39461718)
- Switched the command shortcut for an action from Ctrl+X to Ctrl+G.(v0.39.0-nightly.20260408.a39461718)
- Added the "/memory inbox" command for reviewing extracted skills in memory.(v0.39.0-nightly.20260408.a39461718)
- Added a memory usage integration test harness for testing purposes.(v0.39.0-nightly.20260408.a39461718)
- Added a flag for EPT size and increased the default size.(v0.39.0-nightly.20260408.a39461718)
- Added support for experiment-gated use of gemini flash 3.1 lite.(v0.37.0)
- Implemented forbiddenPaths configuration for OS-specific sandbox managers.(v0.37.0)
- Introduced a new skill specifically designed to look for duplicated code during PR reviews.(v0.37.0)
- Enabled support for inline agentCardJson configuration for remote agents.(v0.37.0)
- Added a 'ci' skill to facilitate automated failure replication.(v0.37.0)
- Implemented Windows sandbox dynamic expansion (Phase 1 and 2.1).(v0.37.0)
- Added dynamic discovery of read-only tools for the browser agent.(v0.37.0)
- Added reliability harvester and 500/503 retry support for evaluations.(v0.37.0)
- Added development worker infrastructure for GCP integration.(v0.37.0)
- Enabled dynamic Linux sandbox expansion and worktree support.(v0.37.0)
- Added maxActionsPerTask setting for the browser agent.(v0.37.0)
- Enabled 'TerminalBuffer' mode to solve flicker in the UI.(v0.38.0-preview.0)
- Added support for default values for environment variables in the CLI.(v0.38.0-preview.0)
- Introduced a background memory service for skill extraction.(v0.38.0-preview.0)
- Implemented context-aware persistent policy approvals.(v0.38.0-preview.0)
- Enabled compact tool output by default in the CLI.(v0.38.0-preview.0)
- Added a scrollbar for the input prompt.(v0.38.0-preview.0)
- Implemented background process monitoring and inspection tools.(v0.38.0-preview.0)
- Added an experimental setting: experimental.adk.agentSessionNoninteractiveEnabled.(v0.38.0-preview.0)
- Enabled 'Other' option for the yes/no question type.(v0.38.0-preview.0)
- Introduced ContextCompressionService in the core.(v0.38.0-preview.0)
- Scoped subagent workspace directories via AsyncLocalStorage.(v0.38.0-preview.0)
- Migrated nonInteractiveCli to LegacyAgentSession.(v0.38.0-preview.0)
- Implemented a high-signal PR regression check for evaluations.(v0.38.0-preview.0)
- Enabled web_fetch in plan mode explicitly when using ask_user.(v0.38.0-preview.0)
- Prioritized summary for topics in core and CLI.(v0.38.0-preview.0)
- Added minimalist sandbox status labels in the CLI.(v0.38.0-preview.0)
- Added browser agent metrics.(v0.38.0-preview.0)
- Changed default loadingPhrases to 'off' to hide tips in the CLI.(v0.38.0-preview.0)
- Introduced PR Guidance workflow for Evals.(v0.36.0-nightly.20260402.13ccc1645)
- Implemented full "GEMINI CLI" logo display for the logged-out state.(v0.36.0-nightly.20260402.13ccc1645)
- Added the behavioral-evals skill, including fixing and promoting guides.(v0.36.0-nightly.20260402.13ccc1645)
- Added LegacyAgentSession support.(v0.36.0-nightly.20260402.13ccc1645)
- Added TestMcpServerBuilder and support in TestRig for testing utilities.(v0.36.0-nightly.20260402.13ccc1645)
- Added multi-turn interactive evals for memoryManager.(v0.36.0-nightly.20260402.13ccc1645)
- Added behavioral evaluations for subagent routing.(v0.36.0-nightly.20260402.13ccc1645)
- Implemented a refreshed UX for the Composer layout.(v0.36.0-nightly.20260402.13ccc1645)
- Implemented dynamic macOS sandbox expansion and worktree support.(v0.36.0-nightly.20260402.13ccc1645)
- Enabled immediate user input during startup, preventing blocking.(v0.36.0-nightly.20260402.13ccc1645)
- Changed user-facing authentication type from oauth2 to oauth.(v0.36.0-nightly.20260402.13ccc1645)
- Added maxActionsPerTask setting for the browser agent.(v0.36.0-nightly.20260402.13ccc1645)
- Introduced multi-registry architecture and tool filtering for subagents.(v0.36.0)
- Added support for dynamic model Resolution to ModelConfigService.(v0.36.0)
- Integrated task tracker protocol into the core system prompt.(v0.36.0)
- Added agent acknowledgment command and enhanced registry discovery.(v0.36.0)
- Added 'blocked' status to tasks and todos.(v0.36.0)
- Enabled subagent local execution and tool isolation.(v0.36.0)
- Formatted multi-line banner warnings with a bold title in the UI.(v0.36.0)
- Removed the write todo list tool from UI tips.(v0.36.0)
- Added support for plan mode in non-interactive mode.(v0.36.0)
- Implemented strict macOS sandboxing using a Seatbelt allowlist.(v0.36.0)
- Enabled resilient subagent tool rejection with contextual feedback.(v0.36.0)
- Added content-utils.(v0.36.0)
- Added partial threading of AgentLoopContext.(v0.36.0)
- Injected memory and JIT context into subagents.(v0.36.0)
- Capped JIT context upward traversal at the git root.(v0.36.0)
- Added experimental memory manager agent to replace the save_memory tool.(v0.36.0)
- Added token usage metadata to the ACP `send` method's return value.(v0.36.0)
- Added ModelChain support to ModelConfigService and made ModelDialog dynamic.(v0.36.0)
- Added support for experiment-gated use of gemini flash 3.1 lite.(v0.37.0-preview.0)
- Introduced a new skill specifically designed to look for duplicated code during PR reviews.(v0.37.0-preview.0)
- Implemented forbiddenPaths configuration for OS-specific sandbox managers.(v0.37.0-preview.0)
- Enabled dynamic discovery of read-only tools within the browser agent.(v0.37.0-preview.0)
- Added support for inline agentCardJson configuration for remote agents.(v0.37.0-preview.0)
- Implemented Windows sandbox dynamic expansion (Phase 1 and 2.1).(v0.37.0-preview.0)
- Added a 'ci' skill for automated failure replication.(v0.37.0-preview.0)
- Enabled dynamic Linux sandbox expansion and worktree support.(v0.37.0-preview.0)
- Added reliability harvester and 500/503 retry support for evaluations.(v0.37.0-preview.0)
- Added development worker infrastructure for GCP integration.(v0.37.0-preview.0)
- Added customizable keyboard shortcuts via the CLI.(v0.35.0)
- Enabled support for literal character keybindings and extended Kitty protocol keys in the CLI.(v0.35.0)
- Introduced the ability to remove keybindings using a '-' prefix in the CLI.(v0.35.0)
- Added support for defining supplemental admin policies via the --admin-policy flag.(v0.35.0)
- Added Vim yank/paste functionality (y/p/P) using the unnamed register.(v0.35.0)
- Enabled support for custom base URLs via environment variables.(v0.35.0)
- Simplified subagent success UI and improved early termination display.(v0.35.0)
- Implemented the SandboxManager interface and config schema.(v0.35.0)
- Differentiated User-Agent for a2a-server and ACP clients.(v0.35.0)
- Added display for pending and confirming tool calls.(v0.35.0)
- Implemented an input blocker overlay in the browser during automation.(v0.35.0)
- Added Clearcut instrumentation for AI credits billing events via telemetry.(v0.35.0)
- Added a Google credentials provider for remote agents.(v0.35.0)
- Enabled safe tools to execute concurrently.(v0.35.0)
- Introduced multi-registry architecture and tool filtering for subagents.(v0.36.0-preview.0)
- Enabled dynamic model resolution support within the ModelConfigService.(v0.36.0-preview.0)
- Integrated task tracker protocol into the core system prompt.(v0.36.0-preview.0)
- Added agent acknowledgment command and enhanced registry discovery for agent-to-agent (A2A) interactions.(v0.36.0-preview.0)
- Added 'blocked' status capability for tasks and todos.(v0.36.0-preview.0)
- Enabled subagent local execution and tool isolation.(v0.36.0-preview.0)
- Formatted multi-line banner warnings with a bold title for better visibility.(v0.36.0-preview.0)
- Removed the 'write todo list' tool from UI tips.(v0.36.0-preview.0)
- Added support for plan mode in non-interactive command-line usage.(v0.36.0-preview.0)
- Implemented strict macOS sandboxing using a Seatbelt allowlist.(v0.36.0-preview.0)
- Added content-utils for core functionality.(v0.36.0-preview.0)
- Added partial threading of AgentLoopContext.(v0.36.0-preview.0)
- Enabled the 'Allow all server tools' session policy for the browser agent.(v0.36.0-preview.0)
- Injected memory and JIT context into subagents.(v0.36.0-preview.0)
- Capped JIT context upward traversal at the Git root directory.(v0.36.0-preview.0)
- Added an experimental memory manager agent to replace the save_memory tool.(v0.36.0-preview.0)
- Added token usage metadata to the return value of the ACP 'send' method.(v0.36.0-preview.0)
- Added ModelChain support to ModelConfigService and made ModelDialog dynamic.(v0.36.0-preview.0)
- Introduced local execution and tool isolation for subagents.(v0.36.0-nightly.20260321.fc03891a1)
- Enabled plan mode support in non-interactive command-line execution.(v0.36.0-nightly.20260321.fc03891a1)
- Implemented strict macOS sandboxing using a Seatbelt allowlist.(v0.36.0-nightly.20260321.fc03891a1)
- Added native Windows sandboxing support.(v0.36.0-nightly.20260321.fc03891a1)
- Introduced Git worktree support, enabling isolated parallel sessions.(v0.36.0-nightly.20260321.fc03891a1)
- Added telemetry tracking for sessions running within a Git worktree.(v0.36.0-nightly.20260321.fc03891a1)
- Added an experimental memory manager agent to replace the save_memory tool.(v0.36.0-nightly.20260321.fc03891a1)
- Added content-utils for core functionality.(v0.36.0-nightly.20260321.fc03891a1)
- Implemented partial threading of AgentLoopContext.(v0.36.0-nightly.20260321.fc03891a1)
- Injected memory and JIT context into subagents.(v0.36.0-nightly.20260321.fc03891a1)
- Capped JIT context upward traversal at the Git root directory.(v0.36.0-nightly.20260321.fc03891a1)
- Added support for admin-forced MCP server installations.(v0.36.0-nightly.20260321.fc03891a1)
- Conditionally excluded the ask_user tool when operating in ACP mode.(v0.36.0-nightly.20260321.fc03891a1)
- Introduced AgentSession and renamed stream events to agent events for clarity.(v0.36.0-nightly.20260321.fc03891a1)
- Added support for linking in the extension registry.(v0.36.0-nightly.20260321.fc03891a1)
- Added a --skip-settings flag to the extension install command.(v0.36.0-nightly.20260321.fc03891a1)
- Added token usage metadata to the ACP send method's return value.(v0.36.0-nightly.20260321.fc03891a1)
- Added a security prompt specifically for the browser agent.(v0.36.0-nightly.20260321.fc03891a1)
- Refined the User-Agent string for VS Code traffic into a unified format.(v0.36.0-nightly.20260321.fc03891a1)
- Enabled subagents functionality.(v0.36.0-nightly.20260318.e2658ccda)
- Introduced Topic-Action-Summary model for reducing verbosity in prompts.(v0.36.0-nightly.20260318.e2658ccda)
- Added cryptographic integrity verification for extension updates.(v0.36.0-nightly.20260318.e2658ccda)
- Enabled JIT context loading by default.(v0.36.0-nightly.20260318.e2658ccda)
- Added tracker policy support.(v0.36.0-nightly.20260318.e2658ccda)
- Added the ability to return TodoList display for tracker tools.(v0.36.0-nightly.20260318.e2658ccda)
- Added allowed domain restrictions for the browser agent.(v0.36.0-nightly.20260318.e2658ccda)
- Integrated SandboxManager to sandbox all process-spawning tools.(v0.36.0-nightly.20260318.e2658ccda)
- Added the `disableAlwaysAllow` setting to disable auto-approvals for security.(v0.36.0-nightly.20260318.e2658ccda)
- Increased sub-agent turn and time limits.(v0.36.0-nightly.20260318.e2658ccda)
- Instrumented file system tools for JIT context discovery.(v0.36.0-nightly.20260318.e2658ccda)
- Implemented Stage 2 security and consistency improvements for `web_fetch`.(v0.36.0-nightly.20260318.e2658ccda)
- Added chat resume footer functionality when a session is quit in the CLI.(v0.34.0)
- Enabled support for bold and other text styles in SVG snapshots.(v0.34.0)
- Introduced tracker CRUD tools and visualization capabilities in the core.(v0.34.0)
- Standardized semantic focus colors and enhanced history visibility in the UI.(v0.34.0)
- Improved @file autocomplete to prioritize filenames.(v0.34.0)
- Added experimental LXC container sandbox support in the sandbox.(v0.34.0)
- Added overall pass rate row to the evaluation nightly summary table.(v0.34.0)
- Included language information in telemetry data and fixed accepted lines computation.(v0.34.0)
- Added robustness tests, logging, and metrics for CodeAssistServer SSE parsing.(v0.34.0)
- Added issue assignee workflow.(v0.34.0)
- Implemented iterative loop detection and model feedback mechanisms.(v0.34.0)
- Disabled fast ack helper for hints in the core.(v0.34.0)
- Shipped esbuild bundle in the npm package.(v0.34.0)
- Added custom footer configuration via the `/footer` command.(v0.34.0)
- Added native gVisor (runsc) sandboxing support in the CLI.(v0.34.0)
- Implemented Clearcut logging for startup statistics.(v0.34.0)
- Added area/documentation to issue triage.(v0.34.0)
- Added customizable keyboard shortcuts via the CLI.(v0.35.0-preview.1)
- Enabled support for literal character keybindings and extended Kitty protocol keys in the CLI.(v0.35.0-preview.1)
- Added support for removing keybindings using a '-' prefix in the CLI.(v0.35.0-preview.1)
- Introduced the --admin-policy flag for specifying supplemental admin policies.(v0.35.0-preview.1)
- Added Vim yank/paste functionality (y/p/P) using the unnamed register.(v0.35.0-preview.1)
- Enabled support for custom base URLs via environment variables.(v0.35.0-preview.1)
- Simplified subagent success UI and improved early termination display.(v0.35.0-preview.1)
- Implemented the SandboxManager interface and configuration schema.(v0.35.0-preview.1)
- Differentiated User-Agent strings for a2a-server and ACP clients.(v0.35.0-preview.1)
- Added display for pending and confirming tool calls.(v0.35.0-preview.1)
- Implemented an input blocker overlay in the browser component during automation.(v0.35.0-preview.1)
- Added Clearcut instrumentation for AI credits billing events in telemetry.(v0.35.0-preview.1)
- Added a Google credentials provider for remote agents.(v0.35.0-preview.1)
- Enabled safe tools to execute concurrently.(v0.35.0-preview.1)
- Added specific tracking IDs for GitHub Actions, including PR, issue, and custom tracking IDs.(v0.35.0-nightly.20260313.bb060d7a9)
- Introduced OAuth2 Authorization Code authentication provider for A2A agents.(v0.35.0-nightly.20260313.bb060d7a9)
- Enabled visibility for the /tools list command in the TUI, aligning it with other subcommand patterns.(v0.35.0-nightly.20260313.bb060d7a9)
- Implemented support for auto-adding policies by default and scoped persistence.(v0.35.0-nightly.20260313.bb060d7a9)
- Added per-model token usage reporting to stream-json output.(v0.35.0-nightly.20260313.bb060d7a9)
- Removed the experimental badge for plan mode in the sidebar documentation.(v0.35.0-nightly.20260313.bb060d7a9)
- Enabled skill activation directly via slash commands.(v0.35.0-nightly.20260313.bb060d7a9)
- Implemented standardized normalization and streaming reassembly for A2A operations.(v0.35.0-nightly.20260313.bb060d7a9)
- Switched the A2A tool scheduler from callback-based to event-driven architecture.(v0.35.0-nightly.20260313.bb060d7a9)
- Implemented a speech-friendly response formatter for voice interactions.(v0.35.0-nightly.20260313.bb060d7a9)
- Added a pulsating blue border automation overlay for the browser agent.(v0.35.0-nightly.20260313.bb060d7a9)
- Added the ExtensionDetails dialog and support for installing extensions.(v0.35.0-nightly.20260313.bb060d7a9)
- Implemented background process logging and cleanup routines.(v0.35.0-nightly.20260313.bb060d7a9)
- Introduced customizable keyboard shortcuts for the CLI.(v0.35.0-nightly.20260313.bb060d7a9)
- Added support for literal character keybindings and extended Kitty protocol keys.(v0.35.0-nightly.20260313.bb060d7a9)
- Added missing Vim mode motions (including X, ~, r, f/F/t/T, df/dt, and related commands).(v0.35.0-nightly.20260313.bb060d7a9)
- Implemented retry logic for fetch notifications.(v0.35.0-nightly.20260313.bb060d7a9)
- Added the github-issue-creator skill.(v0.33.0)
- Implemented HTTP authentication support for A2A remote agents.(v0.33.0)
- Enabled support for authenticated A2A agent card discovery.(v0.33.0)
- Centralized read_file limits and updated the gemini-3 model description.(v0.33.0)
- Truncated large MCP tool output to improve performance.(v0.33.0)
- Introduced subagent activity UX enhancements.(v0.33.0)
- Added support for plan directory in the extension manifest.(v0.33.0)
- Enabled built-in research subagents in plan mode.(v0.33.0)
- Directly indicate the authentication required state for agents.(v0.33.0)
- Inverted the context window display to show usage.(v0.33.0)
- Implemented slash command handling in ACP for /memory, /init, /extensions, and /restore.(v0.33.0)
- Added support for annotating plans with feedback for iteration.(v0.33.0)
- Enabled 30-day default retention for chat history and removed the associated warning.(v0.33.0)
- Added behavioral evaluations for the ask_user tool.(v0.33.0)
- Added /unassign support.(v0.33.0)
- Added 'Install' as an option when an extension is selected.(v0.33.0)
- Added chat resume footer functionality when a session is quit via the CLI.(v0.34.0-preview.0)
- Enabled support for bold and other text styles in SVG snapshots.(v0.34.0-preview.0)
- Introduced tracker CRUD tools and visualization capabilities in the core.(v0.34.0-preview.0)
- Improved @file autocomplete to prioritize matching filenames.(v0.34.0-preview.0)
- Added experimental LXC container sandbox support.(v0.34.0-preview.0)
- Added overall pass rate row to the evaluation nightly summary table.(v0.34.0-preview.0)
- Included language information in telemetry data and fixed accepted lines computation.(v0.34.0-preview.0)
- Added robustness tests, logging, and metrics for CodeAssistServer SSE parsing.(v0.34.0-preview.0)
- Added issue assignee workflow support.(v0.34.0-preview.0)
- Implemented iterative loop detection and model feedback mechanisms.(v0.34.0-preview.0)
- Shipped an esbuild bundle within the npm package.(v0.34.0-preview.0)
- Added custom footer configuration support via the /footer command.(v0.34.0-preview.0)
- Implemented Clearcut logging for startup statistics.(v0.34.0-preview.0)
- Added area/documentation label support to issue triage.(v0.34.0-preview.0)
- Added native gVisor (runsc) sandboxing support.(v0.34.0-preview.0)
- Overhauled the thinking UI experience in the CLI.(v0.34.0-nightly.20260310.4653b126f)
- Inverted quota language to display usage as 'percent used'.(v0.34.0-nightly.20260310.4653b126f)
- Unified the user experience for the /chat and /resume commands.(v0.34.0-nightly.20260310.4653b126f)
- Exposed the /tools description as an explicit subcommand for better discoverability.(v0.34.0-nightly.20260310.4653b126f)
- Added the /compact alias for the existing /compress command.(v0.34.0-nightly.20260310.4653b126f)
- Enabled Plan Mode by default.(v0.34.0-nightly.20260310.4653b126f)
- Introduced the AgentLoopContext for enhanced core functionality.(v0.34.0-nightly.20260310.4653b126f)
- Implemented the /upgrade command.(v0.34.0-nightly.20260310.4653b126f)
- Enabled progress emission for the browser agent.(v0.34.0-nightly.20260310.4653b126f)
- Implemented robust IP validation and the safeFetch foundation for security.(v0.34.0-nightly.20260310.4653b126f)
- Improved the display of subagent results.(v0.34.0-nightly.20260310.4653b126f)
- Added support for subagent-specific policies in TOML.(v0.34.0-nightly.20260310.4653b126f)
- Redesigned the compact header.(v0.34.0-nightly.20260310.4653b126f)
- Added support for updating extension sources and names.(v0.34.0-nightly.20260310.4653b126f)
- Added concurrency safety guidance for subagent delegation.(v0.34.0-nightly.20260307.6c3a90645)
- Dynamically generated all keybinding hints for improved usability.(v0.34.0-nightly.20260307.6c3a90645)
- Implemented a unified KeychainService and migrated token storage.(v0.34.0-nightly.20260307.6c3a90645)
- Implemented a generic CacheService and optimized the setupUser process.(v0.34.0-nightly.20260307.6c3a90645)
- Implemented dot-prefixing to resolve slash command conflicts.(v0.34.0-nightly.20260307.6c3a90645)
- Hidden Gemma settings from display and marked them as experimental.(v0.34.0-nightly.20260307.6c3a90645)
- Enabled model steering capability within the workspace.(v0.30.0-nightly.20260226.f9f916e1d)
- Introduced the foundation for a task tracker system and service.(v0.30.0-nightly.20260226.f9f916e1d)
- Added support for RuntimeHook functions in extensions.(v0.30.0-nightly.20260226.f9f916e1d)
- Enabled users to run /terminal-setup with explicit yes/no confirmation.(v0.30.0-nightly.20260226.f9f916e1d)
- Implemented experimental browser agent functionality.(v0.30.0-nightly.20260226.f9f916e1d)
- Added the ability to summarize work after executing a plan.(v0.30.0-nightly.20260226.f9f916e1d)
- Introduced support for automatic model switching within Plan Mode.(v0.30.0-nightly.20260226.f9f916e1d)
- Enabled loading of extensions in parallel for faster startup.(v0.30.0-nightly.20260226.f9f916e1d)
- Added the GEMINI_CLI_IDE_PID environment variable to override IDE process detection.(v0.30.0-nightly.20260226.f9f916e1d)
- Introduced the Conseca security framework.(v0.30.0-nightly.20260226.f9f916e1d)
- Allowed expanding full details of an MCP tool upon approval.(v0.30.0-nightly.20260226.f9f916e1d)
- Implemented AfterTool tail tool calls.(v0.30.0-nightly.20260226.f9f916e1d)
- Added progress bar, throttling, and input validation for MCP tool progress reporting.(v0.30.0-nightly.20260226.f9f916e1d)
- Centralized plan mode tool visibility within the policy engine.(v0.30.0-nightly.20260226.f9f916e1d)
- Injected a message when a user manually exits Plan mode.(v0.30.0-nightly.20260226.f9f916e1d)
- Enforced folder trust requirements for installing local extensions.(v0.30.0-nightly.20260226.f9f916e1d)
- Added context breakdown to API response telemetry events.(v0.30.0-nightly.20260226.f9f916e1d)
- Exposed the model.name setting in the settings dialog for persistence across sessions.(v0.30.0-nightly.20260226.f9f916e1d)
- Added text wrapping capabilities to markdown tables.(v0.30.0)
- Added setting to autoconfigure memory usage in the dialog.(v0.30.0)
- Added setting to hide the shortcuts hint UI.(v0.30.0)
- Formalized a 5-phase sequential planning workflow.(v0.30.0)
- Introduced limits for search results.(v0.30.0)
- Enabled tool output masking by default.(v0.30.0)
- Enabled multi-line text answers in the ask-user tool.(v0.30.0)
- Documented and validated Plan Mode policy overrides.(v0.30.0)
- Allowed skills to be enabled in plan mode.(v0.30.0)
- Updated internal utility models to Gemini 3.(v0.30.0)
- Added a value-resolver for auth credential resolution.(v0.30.0)
- Deprecated --allowed-tools and excludeTools in favor of the policy engine.(v0.30.0)
- Refined the Plan Mode system prompt for agentic execution.(v0.30.0)
- Created metrics for usage of the AskUser tool.(v0.30.0)
- Added support for Ctrl-Z suspension.(v0.30.0)
- Added strict seatbelt profiles and removed unusable closed profiles.(v0.30.0)
- Added a prototype clean UI toggle and minimal-mode bleed-through.(v0.30.0)
- Added a generic searchable list to back settings and extensions.(v0.30.0)
- Aligned the AskUser tool color scheme with the UX specification.(v0.30.0)
- Added admin control.(v0.30.0)
- Added a hint to use `gemini --resume` upon exit.(v0.31.0-preview.0)
- Added Alt+D keyboard shortcut for forward word deletion in the CLI.(v0.31.0-preview.0)
- Added macOS run-event notifications for interactive use.(v0.31.0-preview.0)
- Added support for MCP progress updates.(v0.31.0-preview.0)
- Added experimental in-progress steering hints.(v0.31.0-preview.0)
- Included directories added via `/dir add` in @ autocomplete suggestions.(v0.31.0-preview.0)
- Admin settings now only apply if `adminControlsApplicable = true` and fetch errors are treated as fatal.(v0.31.0-preview.0)
- Centralized compatibility checks and added TrueColor detection.(v0.31.0-preview.0)
- Added support for setting the `experimental.useOSC52Copy` setting in the CLI.(v0.31.0-preview.0)
- Replaced the boolean setting for loading phrases with an enum setting.(v0.31.0-preview.0)
- Added support for configuring a custom plans storage directory.(v0.31.0-preview.0)
- Implemented project-level policy support.(v0.31.0-preview.0)
- Implemented parallel FC for read-only tools.(v0.31.0-preview.0)
- Added a `pr-address-comments` skill to assist with PR feedback.(v0.31.0-preview.0)
- Added support for policy chain configuration with Gemini 3.1.(v0.30.0-nightly.20260224.544df749a)
- Introduced experimental direct web fetch capability.(v0.30.0-nightly.20260224.544df749a)
- Enabled mapping of tool kinds to explicit ACP.ToolKind values.(v0.30.0-nightly.20260224.544df749a)
- Implemented support for MCP Server Wildcards in the Policy Engine.(v0.30.0-nightly.20260224.544df749a)
- Implemented Tool Annotation Matching within the Policy Engine.(v0.30.0-nightly.20260224.544df749a)
- Replaced the deprecated expected_replacements parameter with allow_multiple in the replace tool.(v0.30.0-nightly.20260224.544df749a)
- Added a new setting to configure the maximum number of retries (maxRetries).(v0.30.0-nightly.20260224.544df749a)
- Updated command reference and added a /stats command.(v0.30.0-nightly.20260224.544df749a)
- Added a hint to use the `gemini --resume` command upon exit.(v0.30.0-nightly.20260223.c537fd5ae)
- Added Alt+D keyboard shortcut for forward word deletion in the CLI.(v0.30.0-nightly.20260223.c537fd5ae)
- Added support for macOS run-event notifications (interactive mode only).(v0.30.0-nightly.20260223.c537fd5ae)
- Added support for MCP progress updates.(v0.30.0-nightly.20260223.c537fd5ae)
- Added experimental in-progress steering hints.(v0.30.0-nightly.20260223.c537fd5ae)
- Included directories added via `/dir add` in @ autocomplete suggestions.(v0.30.0-nightly.20260223.c537fd5ae)
- Enabled project-level policy support.(v0.30.0-nightly.20260223.c537fd5ae)
- Implemented parallel FC for read-only tools.(v0.30.0-nightly.20260223.c537fd5ae)
- Added experimental setting `experimental.useOSC52Copy` for clipboard operations.(v0.30.0-nightly.20260223.c537fd5ae)
- Replaced the boolean setting for loading phrases with an enum setting.(v0.30.0-nightly.20260223.c537fd5ae)
- Added support for configuring a custom storage directory for plan mode.(v0.30.0-nightly.20260223.c537fd5ae)
- Implemented project-level policy support.(v0.30.0-nightly.20260223.c537fd5ae)
- Introduced a session-based architecture via SDK refactoring.(v0.30.0-nightly.20260223.c537fd5ae)
- Added a new `pr-address-comments` skill to assist with PR feedback.(v0.30.0-nightly.20260223.c537fd5ae)
- Added a generic searchable list to back settings and extensions.(v0.30.0-nightly.20260218.ce84b3cb5)
- Enabled in-CLI extension management commands for the team.(v0.30.0-nightly.20260218.ce84b3cb5)
- Introduced initial package bootstrap for the SDK.(v0.30.0-nightly.20260218.ce84b3cb5)
- Implemented SessionContext for SDK tool calls.(v0.30.0-nightly.20260218.ce84b3cb5)
- Implemented dynamic system instructions for the SDK.(v0.30.0-nightly.20260218.ce84b3cb5)
- Enabled 30-day session retention by default.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added the --policy flag for user-defined policies.(v0.30.0-nightly.20260218.ce84b3cb5)
- Implemented support for custom skills in the SDK.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added the /commands reload command to refresh custom TOML commands.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added a loading state indicator to new agents notifications.(v0.30.0-nightly.20260218.ce84b3cb5)
- Refactored the model command to support set and manage subcommands.(v0.30.0-nightly.20260218.ce84b3cb5)
- Enabled support for custom reasoning models by default.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added Solarized Dark and Solarized Light themes.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added role-specific statistics to telemetry and UI.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added keychain availability and token storage metrics to telemetry.(v0.30.0-nightly.20260218.ce84b3cb5)
- Enabled Plan Mode to support project exploration without planning when active.(v0.30.0-nightly.20260218.ce84b3cb5)
- Added refresh/reload aliases to slash command subcommands.(v0.30.0-nightly.20260218.ce84b3cb5)
- Hidden plan write and edit operations when actively in Plan Mode.(v0.30.0-nightly.20260218.ce84b3cb5)
- Disabled workspace settings when starting GCLI in the home directory.(v0.30.0-nightly.20260218.ce84b3cb5)
- Handled invalid model names in useQuotaAndFallback, falling back to chat-base for unrecognized models.(v0.30.0-nightly.20260218.ce84b3cb5)
- Updated the approval mode cycle order.(v0.30.0-nightly.20260218.ce84b3cb5)
- Introduced the /plan slash command for planning workflows.(v0.29.0)
- Enabled configuration of the user tier name via `enableGetUserTierName` in config.(v0.29.0)
- Added default execution limits for subagents to control resource usage.(v0.29.0)
- Implemented linking functionality for agent skills.(v0.29.0)
- Implemented an admin allowlist for MCP server configurations.(v0.29.0)
- Added support for the /extensions config command.(v0.29.0)
- Added W, B, E Vim motions and operator support for improved text editing.(v0.29.0)
- Implemented support for the `replace` tool in plan mode to allow editing of existing plans.(v0.29.0)
- Added guidance on iterating on approved plans versus creating new plans within plan mode.(v0.29.0)
- Implemented support for MCP servers within Plan mode.(v0.29.0)
- Created a metric for tracking the execution of plans generated in plan mode.(v0.29.0)
- Implemented observation masking for tool outputs to control sensitive information exposure.(v0.29.0)
- Implemented session-linked tool output storage and cleanup mechanisms.(v0.29.0)
- Added behavioral evaluations for plan mode.(v0.29.0)
- Enabled extension configuration by default.(v0.29.0)
- Applied the MCP allowlist to extensions and added a `gemini mcp list` command.(v0.29.0)
- Implemented persistence and resumption for masked tool outputs.(v0.29.0)
- Added WebSocket-based network logging and streaming chunk support.(v0.29.0)
- Added text wrapping capabilities to markdown tables.(v0.30.0-preview.0)
- Added autoconfigure memory usage setting to the dialog.(v0.30.0-preview.0)
- Added setting to hide shortcuts hint UI.(v0.30.0-preview.0)
- Formalized 5-phase sequential planning workflow.(v0.30.0-preview.0)
- Introduced limits for search results.(v0.30.0-preview.0)
- Enabled tool output masking by default.(v0.30.0-preview.0)
- Added multi-line text answers in the ask-user tool.(v0.30.0-preview.0)
- Documented and validated Plan Mode policy overrides.(v0.30.0-preview.0)
- Enabled skills to be enabled in plan mode.(v0.30.0-preview.0)
- Updated internal utility models to Gemini 3.(v0.30.0-preview.0)
- Added value-resolver for auth credential resolution.(v0.30.0-preview.0)
- Deprecated --allowed-tools and excludeTools in favor of the policy engine.(v0.30.0-preview.0)
- Refined Plan Mode system prompt for agentic execution.(v0.30.0-preview.0)
- Created metrics for usage of `AskUser` tool.(v0.30.0-preview.0)
- Added support for Ctrl-Z suspension.(v0.30.0-preview.0)
- Added strict seatbelt profiles and removed unusable closed profiles.(v0.30.0-preview.0)
- Prototyped clean UI toggle and minimal-mode bleed-through.(v0.30.0-preview.0)
- Added generic searchable list to back settings and extensions.(v0.30.0-preview.0)
- Aligned `AskUser` color scheme with UX specification.(v0.30.0-preview.0)
- Added admin control.(v0.30.0-preview.0)
- Added support for Ctrl-Z suspension in the CLI.(v0.30.0-nightly.20260212.207ac6f2d)
- Introduced a prototype clean UI toggle and minimal-mode bleed-through option.(v0.30.0-nightly.20260212.207ac6f2d)
- Added strict seatbelt profiles and removed unusable closed profiles.(v0.30.0-nightly.20260212.207ac6f2d)
- Created metrics for tracking the usage of the `AskUser` tool within Plan Mode.(v0.30.0-nightly.20260212.207ac6f2d)
- Introduced documentation and validation for Plan Mode policy overrides.(v0.30.0-nightly.20260212.099aa9621)
- Enabled skills to be activated within Plan Mode.(v0.30.0-nightly.20260212.099aa9621)
- Updated internal utility models to use Gemini 3.(v0.30.0-nightly.20260212.099aa9621)
- Added a value-resolver for authenticating credential resolution.(v0.30.0-nightly.20260212.099aa9621)
- Deprecated the use of --allowed-tools and excludeTools flags in favor of the new policy engine.(v0.30.0-nightly.20260212.099aa9621)
- Enabled tool output masking by default for enhanced privacy.(v0.30.0-nightly.20260211.f5dd1068f)
- Introduced support for multi-line text answers when using the ask-user tool.(v0.30.0-nightly.20260211.f5dd1068f)
- Implemented observation masking for tool outputs.(v0.30.0-nightly.20260210.8257ec447)
- Enabled session-linked tool output storage and cleanup.(v0.30.0-nightly.20260210.8257ec447)
- Added behavioral evaluations for plan mode.(v0.30.0-nightly.20260210.8257ec447)
- Added extension registry client.(v0.30.0-nightly.20260210.8257ec447)
- Enabled extension configuration by default.(v0.30.0-nightly.20260210.8257ec447)
- Removed previewFeatures and defaulted to Gemini 3.(v0.30.0-nightly.20260210.8257ec447)
- Applied MCP allowlist to extensions and added a gemini mcp list command.(v0.30.0-nightly.20260210.8257ec447)
- Added shortcuts hint and panel for better discoverability.(v0.30.0-nightly.20260210.8257ec447)
- Added positive test cases and updated the eval stability policy for plan mode.(v0.30.0-nightly.20260210.8257ec447)
- Implemented persistence and resumption for masked tool outputs.(v0.30.0-nightly.20260210.8257ec447)
- Added WebSocket-based network logging and streaming chunk support.(v0.30.0-nightly.20260210.8257ec447)
- Updated approval modes UI.(v0.30.0-nightly.20260210.8257ec447)
- Overhauled the system prompt for improved rigor, integrity, and intent alignment.(v0.30.0-nightly.20260210.8257ec447)
- Exempted high-signal tools from output masking.(v0.30.0-nightly.20260210.8257ec447)
- Added remote configuration for tool output masking thresholds.(v0.30.0-nightly.20260210.8257ec447)
- Paused agent timeout budget while waiting for tool confirmation.(v0.30.0-nightly.20260210.8257ec447)
- Truncated shell output in UI history and improved active shell display.(v0.30.0-nightly.20260210.8257ec447)
- Conditionally included the ctrl+f prompt based on the interactive shell setting.(v0.30.0-nightly.20260210.8257ec447)
- Transitioned sub-agents to XML format and improved definitions.(v0.30.0-nightly.20260210.8257ec447)
- Implemented atomic writes and safety checks for trusted folders.(v0.30.0-nightly.20260210.8257ec447)
- Moved shell efficiency guidelines to the tool description.(v0.30.0-nightly.20260210.8257ec447)
- Added the /prompt-suggest slash command for generating prompt suggestions.(v0.28.0)
- Enabled activity logging for non-interactive mode and evaluations.(v0.28.0)
- Added support for custom deny messages in policy rules.(v0.28.0)
- Added support for custom themes in extensions.(v0.28.0)
- Updated undo/redo keybindings to Cmd+Z/Alt+Z and Shift+Cmd+Z/Shift+Alt+Z.(v0.28.0)
- Display user identity (auth, email, tier) on startup.(v0.28.0)
- Show hooksConfig.enabled status in the settings dialog.(v0.28.0)
- Added support for the positron IDE.(v0.28.0)
- Added support for the GOOGLE_GENAI_API_VERSION environment variable.(v0.28.0)
- Implemented interactive and non-interactive consent flows for OAuth.(v0.28.0)
- Implemented dynamic policy registration for subagents.(v0.28.0)
- Preserved End-Of-Line (EOL) characters in files during operations.(v0.28.0)
- Created a generic Checklist component and refactored the Todo component.(v0.28.0)
- Added an npx smoke test to verify tool installability.(v0.28.0)
- Introduced the /plan slash command for planning workflows.(v0.29.0-preview.0)
- Enabled configuration of the user tier name via `getUserTierName` in config.(v0.29.0-preview.0)
- Added default execution limits for subagents to control resource usage.(v0.29.0-preview.0)
- Implemented linking functionality for agent skills.(v0.29.0-preview.0)
- Implemented an admin allowlist for MCP server configurations.(v0.29.0-preview.0)
- Added support for the /extensions config command.(v0.29.0-preview.0)
- Added W, B, E Vim motions and operator support for enhanced editing.(v0.29.0-preview.0)
- Enabled support for the `replace` tool in plan mode to edit existing plans.(v0.29.0-preview.0)
- Added guidance on iterating on approved plans versus creating new plans in the /plan feature.(v0.29.0-preview.0)
- Implemented support for MCP servers within Plan mode.(v0.29.0-preview.0)
- Created a metric for tracking the execution of plans generated in plan mode.(v0.29.0-preview.0)
- Implemented observation masking for tool outputs to control sensitive data exposure.(v0.29.0-preview.0)
- Implemented session-linked tool output storage and cleanup for better session management.(v0.29.0-preview.0)
- Added behavioral evaluations for plan mode.(v0.29.0-preview.0)
- Added an extension registry client.(v0.29.0-preview.0)
- Enabled extension configuration by default.(v0.29.0-preview.0)
- Applied MCP allowlist to extensions and added a `gemini mcp list` command.(v0.29.0-preview.0)
- Added positive test cases and updated the eval stability policy for plan mode.(v0.29.0-preview.0)
- Implemented persistence and resumption for masked tool outputs.(v0.29.0-preview.0)
- Added WebSocket-based network logging and streaming chunk support.(v0.29.0-preview.0)
- Added support for the /extensions config command.(v0.29.0-nightly.20260206.4ffc349c1)
- Added W, B, E Vim motions and operator support.(v0.29.0-nightly.20260206.4ffc349c1)
- Enabled support for using the `replace` tool within plan mode to edit existing plans.(v0.29.0-nightly.20260206.4ffc349c1)
- Added guidance within plan mode on iterating on approved plans versus creating entirely new plans.(v0.29.0-nightly.20260206.4ffc349c1)
- Integrated planning artifacts and tools into primary workflows within plan mode.(v0.29.0-nightly.20260206.4ffc349c1)
- Implemented support for MCP servers specifically within Plan mode.(v0.29.0-nightly.20260206.4ffc349c1)
- Expanded `verify-release` functionality to cover macOS and Windows platforms.(v0.29.0-nightly.20260206.4ffc349c1)
- Created a new metric to track the execution of plans generated in plan mode.(v0.29.0-nightly.20260206.4ffc349c1)
- Introduced tool name aliasing for backward compatibility in the core.(v0.29.0-nightly.20260205.a3af4a8ca)
- Enabled automatic theme switching based on the terminal background.(v0.29.0-nightly.20260205.a3af4a8ca)
- Updated the slash command from `review-frontend-and-fix` to `review-and-fix`.(v0.29.0-nightly.20260205.a3af4a8ca)
- Implemented the `plan` slash command.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added the `enter_plan_mode` tool.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added markdown rendering support to the `ask_user` tool.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added support for MCP configuration via admin controls (part 1).(v0.29.0-nightly.20260205.a3af4a8ca)
- Required user consent before performing MCP server OAuth.(v0.29.0-nightly.20260205.a3af4a8ca)
- Moved user identity display to the header for better visibility.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added pluggable authentication provider infrastructure for A2A.(v0.29.0-nightly.20260205.a3af4a8ca)
- Enabled `getUserTierName` in the configuration.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added default execution limits for subagents.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added support for draft-2020-12 JSON Schema with lenient fallback.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added A2A auth config types.(v0.29.0-nightly.20260205.a3af4a8ca)
- Enabled the `.agents/skills` directory as an alias for skill discovery.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added core logic and the `exit_plan_mode` tool definition for planning.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added the `exit_plan_mode` UI and prompt.(v0.29.0-nightly.20260205.a3af4a8ca)
- Enabled the ability to disable loop detection via a setting.(v0.29.0-nightly.20260205.a3af4a8ca)
- Added logic to handle inconsistency in schedulers within plan mode.(v0.29.0-nightly.20260205.a3af4a8ca)
- Enabled custom deny messages in plan mode policies.(v0.29.0-nightly.20260205.a3af4a8ca)
- Increased the `ask_user` label limit to 16 characters.(v0.29.0-nightly.20260205.a3af4a8ca)
- Implemented an event-driven tool execution scheduler in the CLI.(v0.27.0)
- Added the code-reviewer skill.(v0.27.0)
- Extended Shift+Tab Mode Cycling to include Plan Mode.(v0.27.0)
- Refactored TestRig and eval helper to support configurable approval modes.(v0.27.0)
- Supported recursive workstream labeling and new IDs for workflows.(v0.27.0)
- Enabled dynamic variable substitution in system prompt override.(v0.27.0)
- Enabled recursive directory access for core functionality.(v0.27.0)
- Added approval mode instructions to the system prompt for planning.(v0.27.0)
- Enabled disableLLMCorrection by default in the core.(v0.27.0)
- Consolidated shell inactivity and redirection monitoring in the CLI.(v0.27.0)
- Enabled the EventDrivenScheduler by default in configuration.(v0.27.0)
- Enabled the hooks system by default.(v0.27.0)
- Enabled the AgentRegistry to track all discovered subagents.(v0.27.0)
- Made subagents use a JSON schema type for input.(v0.27.0)
- Replaced large text pastes with a [Pasted Text: X lines] placeholder.(v0.27.0)
- Implemented the /rewind command.(v0.27.0)
- Added detection for JetBrains IDEs.(v0.27.0)
- Added the AskUser tool schema.(v0.27.0)
- Added state management and plumbing for agent configuration dialog.(v0.27.0)
- Added the docs-writer skill.(v0.27.0)
- Implemented the /rewind command.(v0.27.0)
- Added image pasting capabilities for Wayland and X11 on Linux.(v0.27.0)
- Added the /prompt-suggest slash command for generating prompt suggestions.(v0.28.0-preview.0)
- Enabled activity logging for non-interactive mode and evaluations.(v0.28.0-preview.0)
- Added support for Positron IDE.(v0.28.0-preview.0)
- Added support for custom deny messages in policy rules.(v0.28.0-preview.0)
- Added support for custom themes in extensions.(v0.28.0-preview.0)
- Updated undo/redo keybindings to Cmd+Z/Alt+Z and Shift+Cmd+Z/Shift+Alt+Z.(v0.28.0-preview.0)
- Display user identity (auth, email, tier) on startup.(v0.28.0-preview.0)
- Show hooksConfig.enabled status in the settings dialog.(v0.28.0-preview.0)
- Added user identity info to the stats command.(v0.28.0-preview.0)
- Created a generic Checklist component and refactored Todo functionality.(v0.28.0-preview.0)
- Added support for the GOOGLE_GENAI_API_VERSION environment variable.(v0.28.0-preview.0)
- Preserved End-Of-Line (EOL) characters in files during operations.(v0.28.0-preview.0)
- Implemented interactive and non-interactive consent flows for OAuth.(v0.28.0-preview.0)
- Added an npx smoke test to verify installability.(v0.28.0-preview.0)
- Implemented dynamic policy registration for subagents.(v0.28.0-preview.0)
- Added quick clear input shortcuts when using vim mode.(v0.27.0-nightly.20260127.00f60ef53)
- Enabled support for multi-scheduler tool aggregation and nested call IDs.(v0.27.0-nightly.20260127.00f60ef53)
- Implemented first-run experience for project-level sub-agents.(v0.27.0-nightly.20260127.00f60ef53)
- Added persistent plan file storage for better state management.(v0.27.0-nightly.20260127.00f60ef53)
- Migrated subagents to an event-driven scheduler.(v0.27.0-nightly.20260127.00f60ef53)
- Added an option to use a solid background color for the input prompt.(v0.27.0-nightly.20260127.00f60ef53)
- Enabled refreshing the system prompt when the approval mode changes (using Shift+Tab).(v0.27.0-nightly.20260127.00f60ef53)
- Added a global setting to disable UI spinners.(v0.27.0-nightly.20260127.00f60ef53)
- Enabled AgentRegistry to track all discovered subagents.(v0.27.0-nightly.20260126.cb772a5b7)
- Made subagents use a JSON schema type for input.(v0.27.0-nightly.20260126.cb772a5b7)
- Replaced large text pastes with a [Pasted Text: X lines] placeholder.(v0.27.0-nightly.20260126.cb772a5b7)
- Enabled queuing of specific nightly evaluation tests.(v0.27.0-nightly.20260126.cb772a5b7)
- Implemented the /rewind command for conversation history navigation.(v0.27.0-nightly.20260126.cb772a5b7)
- Added IDE detection capabilities for JetBrains IDEs.(v0.27.0-nightly.20260126.cb772a5b7)
- Added the AskUser tool schema.(v0.27.0-nightly.20260126.cb772a5b7)
- Added state management and plumbing for the agent configuration dialog.(v0.27.0-nightly.20260126.cb772a5b7)
- Introduced the docs-writer new skill.(v0.27.0-nightly.20260126.cb772a5b7)
- Added the 'communicate' tool kind for planning.(v0.27.0-nightly.20260126.cb772a5b7)
- Enabled A/B Testing for Numerical Complexity Scoring for Gemini 3.(v0.27.0-nightly.20260126.cb772a5b7)
- Updated the UI Theme for Plan Mode.(v0.27.0-nightly.20260126.cb772a5b7)
- Added the /agents config command and improved agent discovery.(v0.27.0-nightly.20260126.cb772a5b7)
- Added enable/disable commands for MCP servers.(v0.27.0-nightly.20260126.cb772a5b7)
- Enabled agents to refresh after loading an extension.(v0.27.0-nightly.20260126.cb772a5b7)
- Implemented a simple workflow for planning in the main agent.(v0.27.0-nightly.20260126.cb772a5b7)
- Added alternative command names for the Antigravity editor detection.(v0.27.0-nightly.20260126.cb772a5b7)
- Added the AskUserDialog UI component for the AskUser tool.(v0.27.0-nightly.20260126.cb772a5b7)
- Displayed the user tier in the about command output.(v0.27.0-nightly.20260126.cb772a5b7)
- Added `clearContext` to `AfterAgent` hooks.(v0.27.0-nightly.20260126.cb772a5b7)
- Implemented persistent `approvalMode` setting for planning.(v0.27.0-nightly.20260126.cb772a5b7)
- Added an interactive ValidationDialog to handle 403 VALIDATION_REQUIRED errors.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled clearing the prompt by pressing Esc-Esc when the prompt is not empty.(v0.27.0-nightly.20260122.61040d0eb)
- Implemented an event-driven tool execution scheduler.(v0.27.0-nightly.20260122.61040d0eb)
- Added the code-reviewer skill.(v0.27.0-nightly.20260122.61040d0eb)
- Extended Shift+Tab Mode Cycling to include Plan Mode.(v0.27.0-nightly.20260122.61040d0eb)
- Refactored TestRig and eval helper to support configurable approval modes.(v0.27.0-nightly.20260122.61040d0eb)
- Supported recursive workstream labeling and new IDs for workflows.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled dynamic variable substitution in system prompt override.(v0.27.0-nightly.20260122.61040d0eb)
- Added approval mode instructions to the system prompt.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled disableLLMCorrection by default.(v0.27.0-nightly.20260122.61040d0eb)
- Consolidated shell inactivity and redirection monitoring.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled the event-driven scheduler by default.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled the hooks system by default.(v0.27.0-nightly.20260122.61040d0eb)
- Added an experiment gate for the event-driven scheduler in the CLI.(v0.26.0-nightly.20260121.211d2c5fd)
- Enabled improved shell redirection transparency and security in the core.(v0.26.0-nightly.20260121.211d2c5fd)
- Introduced enhanced anchored iterative context compression with self-verification.(v0.26.0-nightly.20260121.211d2c5fd)
- Set admin.skills.enabled based on the advancedFeaturesEnabled setting.(v0.26.0-nightly.20260121.211d2c5fd)
- Applied admin settings to gemini skills, mcp, and extensions commands.(v0.26.0-nightly.20260121.211d2c5fd)
- Added telemetry to track adoption and usage of plan mode.(v0.26.0-nightly.20260121.211d2c5fd)
- Added a flash lite utility fallback chain.(v0.26.0-nightly.20260121.211d2c5fd)
- Added initial evaluation for the generalist agent.(v0.26.0-nightly.20260121.211d2c5fd)
- Unified agent enabled and disabled flags.(v0.26.0-nightly.20260121.211d2c5fd)
- Ensured all properties in the `hooks` object are event names.(v0.26.0-nightly.20260121.211d2c5fd)
- Added support for Hx.(v0.25.0)
- Introduced GEMINI_CLI_HOME environment variable for strict test isolation.(v0.25.0)
- Enabled multi-scope skill enablement and shadowing fixes for Skills.(v0.25.0)
- Added confirmation details support and JSON-RPC vs HTTP REST support.(v0.25.0)
- Enabled filepath autosuggestion after using slash commands in the CLI.(v0.25.0)
- Added an upgrade option for paid users.(v0.25.0)
- Exported chat history in the /bug command and prefilled GitHub issues.(v0.25.0)
- Enabled interactive shell mode for external tooling relying on the a2a server.(v0.25.0)
- Added support for Antigravity terminal in the terminal setup utility.(v0.25.0)
- Wired up model routing to subagents.(v0.25.0)
- Added the /agents slash command to list available agents.(v0.25.0)
- Implemented file system reversion utilities for the Rewind functionality.(v0.25.0)
- Added native Sublime Text support to IDE detection.(v0.25.0)
- Enabled the cli_help agent by default.(v0.25.0)
- Added support for @ suggestions.(v0.25.0)
- Added a Rewind Confirmation dialog and a Rewind Viewer component.(v0.26.0-preview.0)
- Introduced the 'skill-creator' built-in skill and CJS management tools.(v0.26.0-preview.0)
- Undeprecated the --prompt flag for CLI usage.(v0.26.0-preview.0)
- Added an 'auto' alias for default model selection in configuration.(v0.26.0-preview.0)
- Enabled functionality to enable and disable agents.(v0.26.0-preview.0)
- Added an experimental 'plan' flag for planning features.(v0.26.0-preview.0)
- Added security consent prompts for skill installation via the CLI.(v0.26.0-preview.0)
- Added source tracking to policy rules.(v0.26.0-preview.0)
- Enforced '🔒 maintainer only' label permissions and fixed bot loops in automation.(v0.26.0-preview.0)
- Added support for running available commands prior to MCP servers loading.(v0.26.0-preview.0)
- Added experimental 'plan' approval mode.(v0.26.0-preview.0)
- Added functional awaitConfirmation utility for the scheduler.(v0.26.0-preview.0)
- Steered the outer agent to use expert subagents when present.(v0.26.0-preview.0)
- Added source tracking to policy rules for better auditing.(v0.26.0-nightly.20260119.20580d754)
- Introduced an experimental 'plan' approval mode for execution control.(v0.26.0-nightly.20260119.20580d754)
- Added a functional `awaitConfirmation` utility for scheduling operations.(v0.26.0-nightly.20260119.20580d754)
- Added the `generalist` agent.(v0.26.0-nightly.20260119.20580d754)
- Enabled agent skills functionality by default.(v0.26.0-nightly.20260119.20580d754)
- Implemented admin controls polling and added a restart prompt for administrators.(v0.26.0-nightly.20260119.20580d754)
- Renamed negative settings (e.g., `disable*`) to positive naming (e.g., `enable*`) for clarity.(v0.26.0-nightly.20260119.20580d754)
- Replaced relative keyboard shortcuts link with a direct web URL.(v0.26.0-nightly.20260119.20580d754)
- Enforced strict read-only policy within 'plan' mode and halts execution upon violation.(v0.26.0-nightly.20260119.20580d754)
- Removed `read_many_files` capability from 'plan' approval mode policies.(v0.26.0-nightly.20260119.20580d754)
- Added prompt for users to restart the CLI when changing authentication to OAuth mid-session or if no auth type is selected at startup.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Introduced the /memory command for the a2a server.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Added install and uninstall commands for managing skills via the CLI.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Enabled using the Tab key to switch focus between the shell and the input area.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Enabled support for shipping built-in skills directly with the CLI.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Implemented dynamic terminal tab titles to reflect CLI status.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Added the 'disableLLMCorrection' setting to skip automatic correction in edit tools.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Clarified mandatory YAML frontmatter requirements for sub-agents.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Implemented support for subagents as extensions.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Added support for defining agents within settings.json.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Highlighted the persist mode status within the ModelDialog.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Enhanced the retry mechanism and overall user experience for retries.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Added a Rewind Confirmation dialog and a Rewind Viewer component.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Introduced the 'skill-creator' built-in skill and CJS management tools.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Added support for Folder Trust when using Hooks.(v0.24.0)
- Introduced dynamic mode-aware policy evaluation.(v0.24.0)
- Enabled automatic persistence of the last used /model across Gemini CLI sessions.(v0.24.0)
- Implemented unified secrets sanitization and environment variable redaction options.(v0.24.0)
- Added experimental in-CLI subcommands for installing and uninstalling extensions.(v0.24.0)
- Added support for A2A Client Manager.(v0.24.0)
- Enabled terse transformations of image paths within the text buffer.(v0.24.0)
- Added security warnings and improved layout for the Hooks list.(v0.24.0)
- Added project-level hook warnings for security context.(v0.24.0)
- Added modifyOtherKeys protocol support for tmux integration.(v0.24.0)
- Enabled granular shell command allowlisting within the policy engine.(v0.24.0)
- Added support for remote agents and multi-agent TOML files.(v0.24.0)
- Implemented tool input modification support.(v0.24.0)
- Added instructions to the extensions update info notification.(v0.24.0)
- Added extension settings information to the /extensions list command output.(v0.24.0)
- Implemented Core Skill Infrastructure and Tiered Discovery for Agent Skills.(v0.24.0)
- Added remote agents to the agent registry.(v0.24.0)
- Implemented STOP_EXECUTION functionality and enhanced hook decision handling.(v0.24.0)
- Implemented Autonomous Activation Tool and Context Injection for Agent Skills.(v0.24.0)
- Added folder suggestions to the /dir add command.(v0.24.0)
- Implemented Agent Integration and System Prompt Awareness for Agent Skills.(v0.24.0)
- Added Status Bar Integration to display Skill Counts for Agent Skills.(v0.24.0)
- Implemented Unification of Representation and Centralized Loading for Agent Skills.(v0.24.0)
- Restored MessageBus optionality for soft migration (Phase 1).(v0.24.0)
- Standardized Tool and Agent Invocation constructors (Phase 2).(v0.24.0)
- Added support for Hx.(v0.25.0-preview.0)
- Introduced GEMINI_CLI_HOME environment variable for strict test isolation.(v0.25.0-preview.0)
- Enabled multi-scope skill enablement and shadowing.(v0.25.0-preview.0)
- Added confirmation details support and JSON-RPC vs HTTP REST support.(v0.25.0-preview.0)
- Added filepath autosuggestion after using slash commands in the CLI.(v0.25.0-preview.0)
- Added an upgrade option for paid users.(v0.25.0-preview.0)
- Enabled exporting chat history in the /bug command and prefilling GitHub issues.(v0.25.0-preview.0)
- Enabled interactive shell mode for external tooling relying on the a2a server.(v0.25.0-preview.0)
- Added extension linking capabilities in the CLI.(v0.25.0-preview.0)
- Added native Sublime Text support to IDE detection.(v0.25.0-preview.0)
- Added the /agents slash command to list available agents.(v0.25.0-preview.0)
- Implemented file system reversion utilities for the Rewind functionality.(v0.25.0-preview.0)
- Added support for Antigravity terminal in the terminal setup utility.(v0.25.0-preview.0)
- Enabled model routing to subagents.(v0.25.0-preview.0)
- Enabled the cli_help agent by default.(v0.25.0-preview.0)
- Introduced granular stop and block behavior for agent hooks.(v0.25.0-nightly.20260112.15891721a)
- Added a command to manage Gemini skills via the CLI (/skills management command).(v0.25.0-nightly.20260112.15891721a)
- Enabled implementation of /skills reload command.(v0.25.0-nightly.20260112.15891721a)
- Added a setting to support OSC 52 paste functionality.(v0.25.0-nightly.20260112.15891721a)
- Enabled support for context injection via the SessionStart hook.(v0.25.0-nightly.20260112.15891721a)
- Added support for remote agents.(v0.25.0-nightly.20260112.15891721a)
- Added visual indicators for hook execution in the UI.(v0.25.0-nightly.20260112.15891721a)
- Added a hooks.enabled setting to control hook execution.(v0.25.0-nightly.20260112.15891721a)
- Introduced remote admin settings, including secureModeEnabled and mcpEnabled.(v0.25.0-nightly.20260112.15891721a)
- Added support for Hx.(v0.25.0-nightly.20260112.15891721a)
- Implemented multi-scope skill enablement and shadowing fix for Agent Skills.(v0.25.0-nightly.20260112.15891721a)
- Added support for extensions disabled via admin settings (though this was reverted, it was present in the notes).(v0.25.0-nightly.20260112.15891721a)
- Added description for each setting item displayed in the /settings command.(v0.25.0-nightly.20260112.15891721a)
- Enabled opt-in persistence of the model selected via /model.(v0.25.0-nightly.20260112.15891721a)
- Introduced remote agent infrastructure and renamed the local executor.(v0.23.0)
- Added the `/auth logout` command to clear credentials and authentication state.(v0.23.0)
- Launched Gemini 3 Flash model support in the Gemini CLI.(v0.23.0)
- Added support for Windows clipboard image pasting, including an Alt+V workaround.(v0.23.0)
- Added an agent TOML parser.(v0.23.0)
- Added a core tool that incorporates all context from the core package.(v0.23.0)
- Added support for friendly names and descriptions in hooks.(v0.23.0)
- Enabled detection of the background color.(v0.23.0)
- Passed additional environment variables to shell execution.(v0.23.0)
- Implemented JIT context memory loading and UI synchronization.(v0.23.0)
- Put the "Allow for all future sessions" confirmation behind a setting that is off by default.(v0.23.0)
- Added documentation for the System Prompt Override (GEMINI_SYSTEM_MD).(v0.23.0)
- Added an Introspection agent demo.(v0.23.0)
- Added clear logging for hooks telemetry.(v0.23.0)
- Added support for `.geminiignore` in the SearchText tool.(v0.23.0)
- Added support for Folder Trust when using Hooks.(v0.24.0-preview.0)
- Introduced dynamic mode-aware policy evaluation.(v0.24.0-preview.0)
- Enabled automatic persistence of the last used /model across Gemini CLI sessions.(v0.24.0-preview.0)
- Implemented unified secrets sanitization and environment variable redaction options.(v0.24.0-preview.0)
- Added experimental in-CLI subcommands for installing and uninstalling extensions.(v0.24.0-preview.0)
- Added support for A2A Client Manager.(v0.24.0-preview.0)
- Enabled terse transformations of image paths within the text buffer.(v0.24.0-preview.0)
- Added security warnings and improved layout for the Hooks list.(v0.24.0-preview.0)
- Added project-level hook warnings for security context.(v0.24.0-preview.0)
- Added modifyOtherKeys protocol support for tmux integration.(v0.24.0-preview.0)
- Enabled granular shell command allowlisting within the policy engine.(v0.24.0-preview.0)
- Added support for remote agents and multi-agent TOML files.(v0.24.0-preview.0)
- Implemented tool input modification support.(v0.24.0-preview.0)
- Added instructions to the extensions update info notification.(v0.24.0-preview.0)
- Added extension settings information to the /extensions list command output.(v0.24.0-preview.0)
- Implemented Core Skill Infrastructure and Tiered Discovery for Agent Skills.(v0.24.0-preview.0)
- Added remote agents to the agent registry.(v0.24.0-preview.0)
- Implemented STOP_EXECUTION functionality and enhanced hook decision handling.(v0.24.0-preview.0)
- Implemented Autonomous Activation Tool and Context Injection for Agent Skills.(v0.24.0-preview.0)
- Added folder suggestions to the /dir add command.(v0.24.0-preview.0)
- Implemented Agent Integration and System Prompt Awareness for Agent Skills.(v0.24.0-preview.0)
- Added Status Bar Integration to display Skill Counts for Agent Skills.(v0.24.0-preview.0)
- Implemented Unification of Representation and Centralized Loading for Agent Skills.(v0.24.0-preview.0)
- Restored MessageBus optionality for soft migration (Phase 1).(v0.24.0-preview.0)
- Standardized Tool and Agent Invocation constructors (Phase 2).(v0.24.0-preview.0)
- Introduced Autonomous Activation Tool and Context Injection for Agent Skills.(v0.24.0-nightly.20260103.30f5c4af4)
- Enabled Agent Integration and System Prompt Awareness for Agent Skills.(v0.24.0-nightly.20260103.30f5c4af4)
- Added folder suggestions when using the /dir add command.(v0.24.0-nightly.20260103.30f5c4af4)
- Integrated Agent Skill counts into the Status Bar for quick visibility.(v0.24.0-nightly.20260103.30f5c4af4)
- Enabled automatic persistence of the currently selected model (/model) across Gemini CLI sessions.(v0.21.0-nightly.20251226.546baf993)
- Introduced experimental in-CLI subcommands for installing and uninstalling extensions.(v0.21.0-nightly.20251226.546baf993)
- Added support for Folder Trust to Hooks functionality.(v0.21.0-nightly.20251226.546baf993)
- Implemented dynamic mode-aware policy evaluation.(v0.21.0-nightly.20251226.546baf993)
- Added security warning and improved layout for the Hooks list in the UI.(v0.21.0-nightly.20251226.546baf993)
- Implemented unified secrets sanitization and environment variable redaction options.(v0.21.0-nightly.20251226.546baf993)
- Added A2A Client Manager.(v0.21.0-nightly.20251226.546baf993)
- Enabled terse transformations of image paths within the text buffer.(v0.21.0-nightly.20251226.546baf993)
- Added project-level hook warnings for security context.(v0.21.0-nightly.20251226.546baf993)
- Added modifyOtherKeys protocol support for tmux integration.(v0.21.0-nightly.20251226.546baf993)
- Enabled fallback to GEMINI_CLI_IDE_AUTH_TOKEN environment variable for authentication.(v0.22.0)
- Introduced display of quota statistics for unused models in the /stats command.(v0.22.0)
- Ensured the codebase investigator uses the preview model when the main agent does.(v0.22.0)
- Added plumbing for late resolution of model configurations.(v0.22.0)
- Added support for specifying `transcript_path` in hooks for the git-ai/Gemini extension.(v0.22.0)
- Introduced the `/init` command for the A2A server.(v0.22.0)
- Enabled support for multi-file drag and drop of images.(v0.22.0)
- Introduced persistent "Always Allow" policies with granular shell & MCP support.(v0.22.0)
- Polished cached token statistics and simplified stats display when quota information is present.(v0.22.0)
- Added validation for the settings schema.(v0.22.0)
- Enabled the codebase investigator by default in read-only policy.(v0.22.0)
- Introduced remote agent infrastructure and renamed the local executor.(v0.23.0-preview.0)
- Added the `/auth logout` command to clear credentials and authentication state.(v0.23.0-preview.0)
- Launched Gemini 3 Flash support in the Gemini CLI.(v0.23.0-preview.0)
- Added support for Windows clipboard image pasting, including an Alt+V paste workaround.(v0.23.0-preview.0)
- Added an agent TOML parser.(v0.23.0-preview.0)
- Added a core tool that incorporates all context from the core package.(v0.23.0-preview.0)
- Added support for friendly names and descriptions in hooks.(v0.23.0-preview.0)
- Enabled detection of the background color.(v0.23.0-preview.0)
- Added support for passing additional environment variables to shell execution.(v0.23.0-preview.0)
- Implemented JIT context memory loading and UI synchronization.(v0.23.0-preview.0)
- Moved the "Allow for all future sessions" confirmation behind a setting that is off by default.(v0.23.0-preview.0)
- Added documentation for the System Prompt Override (GEMINI_SYSTEM_MD).(v0.23.0-preview.0)
- Added an Introspection agent demo.(v0.23.0-preview.0)
- Added clear logging for hooks via telemetry.(v0.23.0-preview.0)
- Added .geminiignore support to the SearchText tool.(v0.23.0-preview.0)
- Implemented Just-In-Time (JIT) context memory loading and synchronized it with the UI.(v0.21.0-nightly.20251220.41a1a3eed)
- Introduced the ability to override the system prompt using the GEMINI_SYSTEM_MD environment variable.(v0.21.0-nightly.20251220.41a1a3eed)
- Added an Introspection agent demo.(v0.21.0-nightly.20251220.41a1a3eed)
All MCP Bug Fixes
- Fixed streaming failures by retrying additional OpenSSL 3.x SSL errors.(v0.40.0-nightly.20260415.g06e7621b2)
- Fixed terminal color detection by correctly identifying kmscon terminal as supporting true color.(v0.40.0-nightly.20260415.g06e7621b2)
- Fixed crash when the vim editor is not found in PATH on Windows.(v0.38.0)
- Fixed unwanted vertical spacing around various tool output treatments in the UI.(v0.38.0)
- Fixed crash in AnsiOutputText when handling non-array data.(v0.38.0)
- Fixed shell output display issues.(v0.38.0)
- Fixed table styling in the UI.(v0.38.0)
- Fixed restoration of topic headers.(v0.38.0)
- Fixed crash related to ARG_MAX when refactoring the Linux sandbox.(v0.38.0)
- Fixed issue where complete_task tool calls were not being recorded in chat history.(v0.38.0)
- Fixed agent stopping when all declinable tools are cancelled.(v0.38.0)
- Fixed issue where stale retry/loading state was not cleared after cancellation.(v0.38.0)
- Fixed issue where unsafe type assertions occurred in the Core File System.(v0.38.0)
- Fixed issue where model quota was incorrectly shown in /stats; refactored quota display.(v0.38.0)
- Fixed broken PowerShell translation and native __write issues in the Windows sandbox.(v0.38.0)
- Fixed issue where sandbox approvals were not correctly persisted and matched for proactive expansions.(v0.38.0)
- Fixed issue where stale slash command list was shown after /skills reload in the CLI.(v0.38.0)
- Fixed issue where sandbox usability was poor and build errors occurred.(v0.38.0)
- Fixed issue where computer-use model detection failed for analyze_screenshot in the browser component.(v0.38.0)
- Fixed handling of all InvalidStreamError types gracefully in prompts.(v0.38.0)
- Fixed issue where skill system instructions were not injected into subagent prompts if activated.(v0.38.0)
- Improved Windows sandbox reliability and fixed related integration tests.(v0.38.0)
- Fixed issue where the global temp directory was not correctly moved under the tmp directory.(v0.38.0)
- Ensured robust sandbox cleanup across all process execution paths.(v0.39.0-preview.0)
- Fixed Windows symlink bypass issues and stabilized sandbox integration tests.(v0.39.0-preview.0)
- Restored file path display in edit and write tool confirmations.(v0.39.0-preview.0)
- Fixed resume bugs by dynamically injecting session IDs.(v0.39.0-preview.0)
- Fixed an issue where Ctrl+Backspace for word deletion was not enabled in Windows Terminal.(v0.39.0-preview.0)
- Fixed sandbox checkpointing errors by adding explicit git identity environment variables.(v0.39.0-preview.0)
- Fixed an issue where FooterConfigDialog respected hideContextPercentage even when closed without changes.(v0.39.0-preview.0)
- Suppressed unhandled AbortError logs that occurred during request cancellation.(v0.39.0-preview.0)
- Honored retryDelay in RetryInfo for 503 errors.(v0.39.0-preview.0)
- Remediated subagent memory leaks by using AbortSignal in MessageBus.(v0.39.0-preview.0)
- Cleared 5-minute timeouts in the OAuth flow to prevent memory leaks.(v0.39.0-preview.0)
- Fixed PTY exhaustion and orphan MCP subprocess leaks.(v0.39.0-preview.0)
- Silently handled EPERM errors when listing directory structures.(v0.39.0-preview.0)
- Fixed decoding issues for Uint8Array and multi-byte UTF-8 characters in API error messages.(v0.39.0-preview.0)
- Resolved lifecycle memory leaks by cleaning up listeners and root closures.(v0.39.0-preview.0)
- Marked /settings as unsafe to run concurrently.(v0.39.0-preview.0)
- Removed buffer slicing to prevent Out-Of-Memory errors on large output streams.(v0.39.0-preview.0)
- Fixed shell execution configuration fields not being preserved on update.(v0.39.0-preview.0)
- Passed the session ID to interactive shell executions.(v0.39.0-preview.0)
- Resolved text sanitization data loss caused by C1 control characters.(v0.39.0-preview.0)
- Fixed a memory leak related to trace's streamWrapper.(v0.39.0-nightly.20260414.gdaf500623)
- Fixed incorrect quota footer display for models that do not auto-update.(v0.39.0-nightly.20260414.gdaf500623)
- Fixed an infinite render loop by extracting QuotaContext in the CLI.(v0.39.0-nightly.20260414.gdaf500623)
- Fixed incorrect handling of UTF-8 characters (U+FFFD) by replacing custom binary detection with isbinaryfile.(v0.39.0-nightly.20260414.gdaf500623)
- Fixed incorrect incrementing of redirect count during fetchJson operations.(v0.39.0-nightly.20260414.gdaf500623)
- Prevented a secondary crash in the ModelRouterService finally block.(v0.39.0-nightly.20260414.gdaf500623)
- Fixed API error messages failing to decode Uint8Array and multi-byte UTF-8 characters.(v0.39.0-nightly.20260411.0957f7d3e)
- Resolved lifecycle memory leaks by ensuring proper cleanup of listeners and root closures.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed an issue where large output streams caused Out-Of-Memory (OOM) errors by removing unnecessary buffer slicing.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed concurrent execution issues by marking the /settings command as unsafe to run concurrently.(v0.39.0-nightly.20260411.0957f7d3e)
- Preserved shell execution configuration fields correctly during updates.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed data loss during text sanitization caused by C1 control characters.(v0.39.0-nightly.20260411.0957f7d3e)
- Resolved PTY exhaustion and prevented orphan MCP subprocess leaks.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed interactive shell executions failing to pass the session ID.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed audio MIME normalization and validation issues during file reads.(v0.39.0-nightly.20260411.0957f7d3e)
- Fixed an issue where hook system messages were not being displayed in the UI.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue where the BeforeModel hook model override was not being propagated end-to-end.(v0.39.0-nightly.20260408.a39461718)
- Fixed skill linking on Windows to correctly use directory junctions.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue that caused multiple banner increments when remounting.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue where the tmux alternate buffer warning was being shown.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue where shell command descriptions were sometimes not shown, ensuring either the description or the actual command is always visible.(v0.39.0-nightly.20260408.a39461718)
- Fixed a memory leak by ensuring the Scheduler is disposed, preventing McpProgress listener leaks.(v0.39.0-nightly.20260408.a39461718)
- Fixed regressions related to the default terminalBuffer=false setting.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue that allowed concurrent browser agent instances to interfere with each other.(v0.39.0-nightly.20260408.a39461718)
- Ensured robust sandbox cleanup across all process execution paths.(v0.39.0-nightly.20260408.a39461718)
- Fixed an issue where the auto-update spinner was shown even when no update was in progress.(v0.37.0)
- Stabilized copy mode to prevent flickering and cursor resets in the CLI.(v0.37.0)
- Fixed flaky ctrl-c-exit test by moving it to the non-blocking test suite.(v0.37.0)
- Fixed an issue where the shell tool would conditionally expose additional_permissions incorrectly.(v0.37.0)
- Prevented over-eager slash subcommand completion in the UI.(v0.37.0)
- Fixed dynamic model routing for gemini 3.1 pro to correctly use the customtools model.(v0.37.0)
- Fixed console log/info messages appearing in headless mode.(v0.37.0)
- Fixed browser agent bypassing domain restrictions by correctly detecting embedded URLs in query parameters.(v0.37.0)
- Fixed browser agent's domain restriction system prompt to include proxy bypass constraints.(v0.37.0)
- Fixed policy issue where write_file argsPattern in plan mode incorrectly required the session ID.(v0.37.0)
- Fixed an issue where attachments were not allowed by adding a required permission prompt.(v0.37.0)
- Threaded AbortSignal to chat compression requests to prevent hangs.(v0.37.0)
- Resolved TTY hang in headless environments by unconditionally resuming process.stdin before React Ink launch.(v0.37.0)
- Fixed UI composer by cleaning up estimated string length hacks.(v0.37.0)
- Prevented the escape key from cancelling ongoing requests while in shell mode.(v0.37.0)
- Fixed CLI to show helpful guidance when no skills are available.(v0.37.0)
- Fixed chat logs and errors to correctly handle tail tool calls.(v0.37.0)
- Fixed CLI to prioritize primary name matches when searching for slash commands.(v0.37.0)
- Fixed browser agent input blocker to remain active across page navigations.(v0.37.0)
- Fixed plan mode to switch the model back to a flash model after exiting plan mode.(v0.37.0)
- Fixed core logic to strip leading underscores from error types reported in telemetry.(v0.37.0)
- Fixed an issue where non-interactive ASK_USER denials were hardcoded instead of relying on explicit policy rules.(v0.37.0)
- Fixed crash when the vim editor is not found in PATH on Windows.(v0.38.0-preview.0)
- Fixed unwanted vertical spacing around various tool output treatments in the UI.(v0.38.0-preview.0)
- Fixed crash in AnsiOutputText when handling non-array data.(v0.38.0-preview.0)
- Fixed shell output display issues.(v0.38.0-preview.0)
- Fixed table styling in the UI.(v0.38.0-preview.0)
- Fixed the Linux sandbox to resolve ARG_MAX crashes.(v0.38.0-preview.0)
- Fixed broken PowerShell translation and native __write in the Windows sandbox.(v0.38.0-preview.0)
- Fixed computer-use model detection for analyze_screenshot in the browser agent.(v0.38.0-preview.0)
- Fixed agent stopping when all declinable tools are cancelled.(v0.38.0-preview.0)
- Fixed stale retry/loading state clearing after cancellation.(v0.38.0-preview.0)
- Fixed unsafe type assertions in the Core File System.(v0.38.0-preview.0)
- Fixed crash when handling all InvalidStreamError types gracefully in prompts.(v0.38.0-preview.0)
- Fixed issues where complete_task tool calls were not recorded in chat history.(v0.38.0-preview.0)
- Fixed agent not stopping when all declinable tools are cancelled.(v0.38.0-preview.0)
- Fixed issues with sandbox usability and build errors.(v0.38.0-preview.0)
- Fixed crash when handling non-array data in AnsiOutputText.(v0.38.0-preview.0)
- Fixed restoration of topic headers.(v0.38.0-preview.0)
- Fixed issues where sandbox approvals were not correctly persisted and matched for proactive expansions.(v0.38.0-preview.0)
- Fixed issues with stale slash command list refresh after /skills reload.(v0.38.0-preview.0)
- Fixed issues where skill system instructions were not injected into subagent prompts if activated.(v0.38.0-preview.0)
- Fixed issues with project memory directory location, moving it under the tmp directory.(v0.38.0-preview.0)
- Fixed issues with model quota display in /stats (hiding model quota).(v0.38.0-preview.0)
- Fixed issues with Windows sandbox reliability and integration tests.(v0.38.0-preview.0)
- Fixed subagent tool updates not applying configuration overrides immediately.(v0.36.0-nightly.20260402.13ccc1645)
- Resolved flicker observed at boundaries of the selection list in BaseSelectionList.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed commands with paths not correctly allowing the "always" option.(v0.36.0-nightly.20260402.13ccc1645)
- Prevented terminal escape sequences from leaking upon tool exit.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed minimum height reservation for the selection list in AskUserDialog.(v0.36.0-nightly.20260402.13ccc1645)
- Hardened AgentSession replay semantics.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed tool confirmations not taking up the entire terminal height in the UI.(v0.36.0-nightly.20260402.13ccc1645)
- Prevented redundant remote agent loading when switching models.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed policy engine not properly supporting allowRedirect.(v0.36.0-nightly.20260402.13ccc1645)
- Prevented subcommand shadowing and fixed skipping authentication for certain commands.(v0.36.0-nightly.20260402.13ccc1645)
- Patched a memory leak and enforced logPrompts privacy in telemetry.(v0.36.0-nightly.20260402.13ccc1645)
- Prevented Ctrl+D from exiting when the input buffer is not empty.(v0.36.0-nightly.20260402.13ccc1645)
- Separated conversational text from the execute tool command title in ACP.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed accurately reflecting subagent tool failure in the UI.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed API key input dialog user interaction when the Gemini API Key was selected.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed sandbox path resolution in Plan Mode to prevent hallucinations.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed ACP by passing the cwd to AcpFileSystemService to avoid looping failures when asking for permissions to write the plan md file.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed core policy configuration requiring explicit specification of toolName.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed overlapping act() deadlock in app-test-helper during evals.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed updating telemetry attribute keys and adding a timestamp in plan mode.(v0.36.0-nightly.20260402.13ccc1645)
- Prevented premature MCP discovery completion.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed agent loader error formatting when paths are empty.(v0.36.0-nightly.20260402.13ccc1645)
- Fixed use of theme-aware text colors for console warnings and errors in devtools.(v0.36.0)
- Fixed CLI to use the active sessionId in useLogger and improved resume robustness.(v0.36.0)
- Fixed CLI to expand tilde (~) in policy paths read from settings.json.(v0.36.0)
- Added actionable warnings for terminal fallbacks.(v0.36.0)
- Fixed CLI to automatically add all VSCode workspace folders to Gemini context.(v0.36.0)
- Fixed CLI to lazily load settings in onModelChange to prevent stale closure data loss.(v0.36.0)
- Fixed CLI to resolve subagent grouping and UI state persistence issues.(v0.36.0)
- Fixed an issue where subagent thoughts were being appended incorrectly.(v0.36.0)
- Fixed CLI to correctly handle auto-update for standalone binaries.(v0.36.0)
- Fixed core logic to not persist browser consent sentinel in non-interactive mode.(v0.36.0)
- Fixed core logic to narrow browser agent description to prevent stealing URL tasks from web_fetch.(v0.36.0)
- Fixed UI flickering observed on small terminal heights.(v0.36.0)
- Fixed ACP to provide more meta information in tool_call_update.(v0.36.0)
- Fixed core logic to explicitly map execution context in LocalAgentExecutor.(v0.36.0)
- Fixed CLI to resolve duplicate footer appearing when canceling a tool via ESC.(v0.36.0)
- Fixed core logic to circumvent genai sdk requirement for api key when using gateway auth via ACP.(v0.36.0)
- Fixed plan mode policies to clarify they are combined with normal mode policies.(v0.36.0)
- Fixed an issue where the copy mode experienced flickering and cursor resets.(v0.37.0-preview.0)
- Fixed the browser agent bypassing domain restrictions by incorrectly detecting embedded URLs in query parameters.(v0.37.0-preview.0)
- Fixed the browser agent's domain restriction system prompt lacking a proxy bypass constraint.(v0.37.0-preview.0)
- Fixed the shell tool conditionally exposing additional_permissions.(v0.37.0-preview.0)
- Fixed the escape key prematurely cancelling requests when in shell mode.(v0.37.0-preview.0)
- Fixed the CLI showing the updating spinner even when auto-update was not in progress.(v0.37.0-preview.0)
- Fixed the CLI hanging on headless environments by ensuring process.stdin is resumed before React Ink launches.(v0.37.0-preview.0)
- Fixed the ask question dialog not taking the full window height in the CLI.(v0.37.0-preview.0)
- Fixed an issue where chat logs and errors did not handle tail tool calls correctly.(v0.37.0-preview.0)
- Fixed the CLI prioritizing non-primary name matches during slash command search.(v0.37.0-preview.0)
- Fixed the plan mode incorrectly switching the model to a flash model after exiting plan mode.(v0.37.0-preview.0)
- Fixed an issue where the agent loader produced poorly formatted errors when paths were empty.(v0.37.0-preview.0)
- Fixed the UI cleanup of estimated string length calculations in the composer.(v0.37.0-preview.0)
- Fixed an issue where the policy for write_file argsPattern in plan mode was too strict, preventing paths without a session ID.(v0.37.0-preview.0)
- Fixed an issue where non-interactive ASK_USER denials were used instead of explicit policy rules.(v0.37.0-preview.0)
- Fixed an issue where the browser input blocker was not kept active across navigations.(v0.37.0-preview.0)
- Fixed an issue where the model routing for gemini 3.1 pro to customtools model was incorrect.(v0.37.0-preview.0)
- Fixed an issue where console logs/info were shown in headless mode when they should have been skipped.(v0.37.0-preview.0)
- Fixed cursor clamping to the last character after all NORMAL mode deletes in the UI.(v0.35.0)
- Allowed scrolling keys to function correctly in copy mode (Ctrl+S selection mode).(v0.35.0)
- Fixed cleanup issues related to /clear and /resume commands.(v0.35.0)
- Fixed orphaned descendant processes being left behind on PTY abort in the core.(v0.35.0)
- Fixed handling of EISDIR errors in robustRealpath function on Windows.(v0.35.0)
- Fixed numerical routing enablement for API key users.(v0.35.0)
- Preserved dynamic tool descriptions when resuming a session.(v0.35.0)
- Silently retried API errors up to 3 times before halting the session.(v0.35.0)
- Fixed inconsistencies with BeforeAgent/AfterAgent hooks.(v0.35.0)
- Resolved skill uninstall failures when the skill name was updated.(v0.35.0)
- Ensured user policies are loaded correctly when policyPaths is empty.(v0.35.0)
- Secured argsPattern and reverted WEB_FETCH_TOOL_NAME escalation.(v0.35.0)
- Fixed registration of themes to occur on extension load instead of start.(v0.35.0)
- Resolved MCP tool FQN validation, schema export, and wildcard handling in subagents.(v0.35.0)
- Validated the --model argument at startup in the CLI.(v0.35.0)
- Handled policy ALLOW for exit_plan_mode correctly.(v0.35.0)
- Fixed propagation of subagent context to the policy engine.(v0.35.0)
- Fixed console warnings and errors to use theme-aware text colors in devtools.(v0.36.0-preview.0)
- Improved resume robustness by ensuring the active sessionId is used in the logger.(v0.36.0-preview.0)
- Fixed issue where tilde (~) in policy paths from settings.json was not expanded.(v0.36.0-preview.0)
- Added actionable warnings for terminal fallbacks.(v0.36.0-preview.0)
- Fixed automatic inclusion of all VSCode workspace folders into Gemini context.(v0.36.0-preview.0)
- Fixed UI state persistence and subagent grouping issues in the CLI.(v0.36.0-preview.0)
- Fixed lazy loading of settings in onModelChange which could lead to stale closure data loss.(v0.36.0-preview.0)
- Fixed issue where subagent thoughts were incorrectly appended to output.(v0.36.0-preview.0)
- Fixed duplicate footer appearing when canceling a tool via the ESC key.(v0.36.0-preview.0)
- Fixed flickering issues observed on small terminal heights.(v0.36.0-preview.0)
- Fixed providing insufficient meta information in tool_call_update for ACP.(v0.36.0-preview.0)
- Fixed auto-update mechanism for standalone binaries.(v0.36.0-preview.0)
- Fixed circumventing the genai SDK requirement for an API key when using gateway authentication via ACP.(v0.36.0-preview.0)
- Fixed persisting the browser consent sentinel in non-interactive mode.(v0.36.0-preview.0)
- Fixed the browser agent description being too broad, preventing it from stealing URL tasks meant for web_fetch.(v0.36.0-preview.0)
- Fixed logging and virtual list rendering issues.(v0.36.0-preview.0)
- Fixed plan mode policies being incorrectly combined with normal mode policies.(v0.36.0-preview.0)
- Fixed an issue where copied extensions were not writable in the user's local directory.(v0.36.0-preview.0)
- Fixed stale closure data loss by lazily loading settings during onModelChange in the CLI.(v0.36.0-nightly.20260321.fc03891a1)
- Resolved issues related to subagent grouping and UI state persistence.(v0.36.0-nightly.20260321.fc03891a1)
- Fixed an issue where subagent thoughts were incorrectly appended.(v0.36.0-nightly.20260321.fc03891a1)
- Resolved duplicate footer display when canceling a tool via the ESC key.(v0.36.0-nightly.20260321.fc03891a1)
- Fixed flickering issues observed in the UI when running in small terminal heights.(v0.36.0-nightly.20260321.fc03891a1)
- Provided more metadata in tool_call_update for ACP.(v0.36.0-nightly.20260321.fc03891a1)
- Circumvented the requirement for an API key when using gateway authentication via ACP.(v0.36.0-nightly.20260321.fc03891a1)
- Prevented the browser consent sentinel from being persisted in non-interactive mode.(v0.36.0-nightly.20260321.fc03891a1)
- Narrowed the browser agent description to prevent it from incorrectly taking URL tasks meant for web_fetch.(v0.36.0-nightly.20260321.fc03891a1)
- Enabled the "Allow all server tools" session policy for the browser agent.(v0.36.0-nightly.20260321.fc03891a1)
- Fixed sanitization issues related to subagent thoughts and centralized utilities.(v0.36.0-nightly.20260321.fc03891a1)
- Correctly handled auto-update functionality for standalone binaries.(v0.36.0-nightly.20260321.fc03891a1)
- Fixed an issue where plan mode policies were incorrectly overriding normal mode policies.(v0.36.0-nightly.20260321.fc03891a1)
- Fixed crash/issue where configuration was undefined.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed manual deletion of subagent histories.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed issue where j/k navigation keys were intercepting input in the edit buffer.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed descriptive error messages when saving settings failed.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed AskUser evals related to planning.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed false positive evals trigger on merge commits.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed command conflict handling for skills in the CLI.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed issue where tool call filtering incorrectly handled Confirming and Canceled tool calls.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed CJK input and full Unicode scalar values support in terminal protocols.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed race condition during MCP config merge and skipping discovery for incomplete configs.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed three JIT context bugs affecting `read_file`, `read_many_files`, and `memoryDiscovery`.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed handling of equivalent root resource URLs in OAuth validation.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed deduplication of project memory when JIT context is enabled.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed use of session-specific temp directory for the task tracker.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed merging of user settings with extension-provided MCP servers.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed stale PR closer permissions and maintainer detection in automation.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed evaluation of staleness before checking protected labels in automation.(v0.36.0-nightly.20260318.e2658ccda)
- Fixed discoveryMaxDirs not being passed to global config during initialization.(v0.36.0-nightly.20260318.e2658ccda)
- Increased A2A agent timeout to 30 minutes.(v0.34.0)
- Fixed merging of duplicate imports in packages/core.(v0.34.0)
- Fixed model persistence for all scenarios.(v0.34.0)
- Polished shell autocomplete rendering to feel more native to the shell.(v0.34.0)
- Fixed notifications/tools/list_changed support not working in MCP.(v0.34.0)
- Fixed registering extension lifecycle events in DebugProfiler.(v0.34.0)
- Enabled multi-arch docker builds for the sandbox.(v0.34.0)
- Fixed an error message when OAuth succeeds but a project ID is required.(v0.34.0)
- Prevented hang during OAuth in non-interactive sessions for core and a2a-server.(v0.34.0)
- Suppressed redundant failure note when a tool error note is shown in the UI.(v0.34.0)
- Preserved symlinks in extension source path while enforcing folder trust.(v0.34.0)
- Deferred tool exclusions to the policy engine in non-interactive mode.(v0.34.0)
- Removed double padding on rendered content in the UI.(v0.34.0)
- Truncated excessively long lines in grep search output.(v0.34.0)
- Explicitly allowed codebase_investigator and cli_help in read-only mode.(v0.34.0)
- Fixed enabling kitty codes on Windows Terminal.(v0.34.0)
- Added partial output to cancelled shell UI.(v0.34.0)
- Replaced hardcoded keybinding strings with dynamic formatters in the CLI.(v0.34.0)
- Fixed shell calls formatting.(v0.34.0)
- Prevented unhandled AbortError crash during stream loop detection.(v0.34.0)
- Fixed cursor clamping to the last character after all NORMAL mode deletes in the UI.(v0.35.0-preview.1)
- Allowed scrolling keys to function correctly in copy mode (Ctrl+S selection mode).(v0.35.0-preview.1)
- Fixed cleanup issues related to /clear and /resume commands.(v0.35.0-preview.1)
- Fixed orphaned descendant processes being left behind on PTY abort in the core.(v0.35.0-preview.1)
- Fixed language detection to use LSP 3.18 identifiers.(v0.35.0-preview.1)
- Fixed an issue where the OAuth check was incorrectly performed in handleFallback and cleaned up a stray file.(v0.35.0-preview.1)
- Fixed handling of EISDIR errors in robustRealpath function on Windows.(v0.35.0-preview.1)
- Fixed numerical routing to be enabled for API key users.(v0.35.0-preview.1)
- Fixed preservation of dynamic tool descriptions when resuming a session.(v0.35.0-preview.1)
- Silently retried API errors up to 3 times before halting the session.(v0.35.0-preview.1)
- Fixed inconsistencies in BeforeAgent/AfterAgent hooks.(v0.35.0-preview.1)
- Fixed propagation of subagent context to the policy engine.(v0.35.0-preview.1)
- Resolved skill uninstall failure when the skill name was updated.(v0.35.0-preview.1)
- Fixed policy engine to ensure user policies are loaded when policyPaths is empty.(v0.35.0-preview.1)
- Secured argsPattern and reverted WEB_FETCH_TOOL_NAME escalation.(v0.35.0-preview.1)
- Fixed MCP tool FQN validation, schema export, and wildcards in subagents.(v0.35.0-preview.1)
- Validated the --model argument during startup in the CLI.(v0.35.0-preview.1)
- Fixed policy handling for exit_plan_mode to allow policy ALLOW.(v0.35.0-preview.1)
- Fixed theme registration to occur on extension load instead of start.(v0.35.0-preview.1)
- Fixed an issue where the toolRegistry property was incorrectly overridden for sub-agent schedulers.(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed footer items in the CLI not being equally spaced.(v0.35.0-nightly.20260313.bb060d7a9)
- Ensured correct flash model steering during the plan mode implementation phase.(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed handling of headless execution within the credits and upgrade dialogs.(v0.35.0-nightly.20260313.bb060d7a9)
- Treated retryable errors with delays longer than 5 minutes as terminal quota errors.(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed the core logic to handle AbortError when the ESC key cancels tool execution.(v0.35.0-nightly.20260313.bb060d7a9)
- Prevented a race condition during retry attempts related to loop detection.(v0.35.0-nightly.20260313.bb060d7a9)
- Updated authentication terminology to consistently use 'sign in' and 'sign out'.(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed UI rendering to prevent empty tool-group border stubs from appearing after filtering.(v0.35.0-nightly.20260313.bb060d7a9)
- Prevented plan truncation in the approval dialog by supporting unconstrained heights.(v0.35.0-nightly.20260313.bb060d7a9)
- Patched gaxios v7 stream corruption caused by Array.toString().(v0.35.0-nightly.20260313.bb060d7a9)
- Prevented hangs in non-interactive mode and improved agent guidance.(v0.35.0-nightly.20260313.bb060d7a9)
- Stabilized the prompt layout in the CLI to prevent jumping while typing.(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed prompt text preservation when streaming is cancelled.(v0.35.0-nightly.20260313.bb060d7a9)
- Improved UX robustness for handling remote agent errors.(v0.35.0-nightly.20260313.bb060d7a9)
- Removed an unnecessary OAuth check from handleFallback and cleaned up a stray file.(v0.35.0-nightly.20260313.bb060d7a9)
- Clamped the cursor to the last character after all NORMAL mode delete operations.(v0.35.0-nightly.20260313.bb060d7a9)
- Allowed scrolling keys to function correctly in copy mode (Ctrl+S selection mode).(v0.35.0-nightly.20260313.bb060d7a9)
- Fixed parsing of raw ASCII buffer strings in Gaxios errors.(v0.33.0)
- Fixed a race condition between MCP initialization and prompt in ACP/Zed.(v0.33.0)
- Resolved autoThemeSwitching issues when the background hasn't changed but the theme mismatches.(v0.33.0)
- Fixed UI flicker when allowing sub-agent confirmation requests.(v0.33.0)
- Reduced false positives for LLM-based loop detection.(v0.33.0)
- Deflaked plan mode integration tests.(v0.33.0)
- Fixed CI failures by using a robot PAT for automated PRs to pass CLA checks.(v0.33.0)
- Fixed issue where subagents were not using qualified MCP tool names.(v0.33.0)
- Fixed issue where telemetry logging occurred for AbortError exceptions.(v0.33.0)
- Restricted the 'System: Please continue' invalid stream retry to Gemini 2 models only.(v0.33.0)
- Fixed CI failures related to stale PR closers when APP_ID is empty.(v0.33.0)
- Fixed plan mode by cleaning up session directories and plans upon deletion.(v0.33.0)
- Enforced optionality for API response fields in code_assist.(v0.33.0)
- Fixed an issue where background auto-updates blocked relaunching.(v0.33.0)
- Fixed pre-loading of @file references from external editor prompts.(v0.33.0)
- Pinned the clipboardy dependency to ~5.2.x to prevent issues.(v0.33.0)
- Increased the A2A agent timeout to 30 minutes to prevent premature failures.(v0.34.0-preview.0)
- Fixed merging of duplicate imports within packages/core.(v0.34.0-preview.0)
- Fixed model persistence issues across all scenarios.(v0.34.0-preview.0)
- Polished shell autocomplete rendering to feel more native to the shell environment.(v0.34.0-preview.0)
- Fixed Notifications/tools/list_changed support not working in MCP.(v0.34.0-preview.0)
- Fixed extension lifecycle events not registering correctly in DebugProfiler.(v0.34.0-preview.0)
- Fixed enabling of kitty codes on Windows Terminal.(v0.34.0-preview.0)
- Fixed crash related to unhandled AbortError during stream loop detection.(v0.34.0-preview.0)
- Fixed an error message when OAuth succeeds but a project ID is required.(v0.34.0-preview.0)
- Prevented hang during OAuth in non-interactive sessions for core and a2a-server.(v0.34.0-preview.0)
- Suppressed redundant failure notes when a tool error note is already being shown.(v0.34.0-preview.0)
- Preserved symlinks in extension source paths while enforcing folder trust.(v0.34.0-preview.0)
- Deferred tool exclusions to the policy engine in non-interactive mode.(v0.34.0-preview.0)
- Removed double padding on rendered content in the UI.(v0.34.0-preview.0)
- Truncated excessively long lines in grep search output.(v0.34.0-preview.0)
- Fixed enabling kitty codes on Windows Terminal.(v0.34.0-preview.0)
- Fixed partial output display in cancelled shell UI.(v0.34.0-preview.0)
- Fixed shell calls formatting.(v0.34.0-preview.0)
- Fixed overage strategy lifecycle and settings integration issues.(v0.34.0-nightly.20260310.4653b126f)
- Fixed expansion of paste placeholders in TextInput upon submission.(v0.34.0-nightly.20260310.4653b126f)
- Fixed crash potential by adding an in-memory cache to ChatRecordingService to prevent OOM errors.(v0.34.0-nightly.20260310.4653b126f)
- Fixed the experience for Ctrl+O expansion hints across different buffer modes.(v0.34.0-nightly.20260310.4653b126f)
- Corrected shell height reporting in the CLI.(v0.34.0-nightly.20260310.4653b126f)
- Fixed test suite failures when GEMINI_SYSTEM_MD or GEMINI_WRITE_SYSTEM_MD environment variables were set locally.(v0.34.0-nightly.20260310.4653b126f)
- Fixed display of actual graph output in the tracker_visualize tool.(v0.34.0-nightly.20260310.4653b126f)
- Sanitized SSE-corrupted JSON and domain strings during error classification.(v0.34.0-nightly.20260310.4653b126f)
- Resolved symlinks for non-existent paths during validation.(v0.34.0-nightly.20260310.4653b126f)
- Prevented infinite recursion during symlink resolution.(v0.34.0-nightly.20260310.4653b126f)
- Corrected Windows line ending and path separation bugs across the CLI.(v0.34.0-nightly.20260310.4653b126f)
- Handled GUI editor non-zero exit codes gracefully.(v0.34.0-nightly.20260310.4653b126f)
- Destroyed PTY on kill() and exception to prevent file descriptor leaks.(v0.34.0-nightly.20260310.4653b126f)
- Fixed logic for task tracker strategy and removed tracker tools.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed the display of media type and size for inline data parts.(v0.34.0-nightly.20260307.6c3a90645)
- Added screen reader support to the RewindViewer for accessibility.(v0.34.0-nightly.20260307.6c3a90645)
- Propagated stopHookActive in the AfterAgent retry path.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed deduplication of GEMINI.md files by device/inode on case-insensitive filesystems.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed graceful handling of the --resume flag when no sessions exist.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed keeping the approved plan during chat compression.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed appending correct OTLP paths for HTTP exporters.(v0.34.0-nightly.20260307.6c3a90645)
- Implemented the --all flag for extensions uninstall command.(v0.34.0-nightly.20260307.6c3a90645)
- Fixed unsafe type assertions in the activityLogger.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed an issue related to punycode2 processing.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed an issue where allowedTools settings were not being passed correctly to the core Config in the a2a-server.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed the creation of McpClient upon restart, ensuring updated configuration is applied.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed an issue where the PR rate limiter incorrectly excluded maintainers in CI actions.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed extension environment directory loading issues.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed skipping of 404 errors during setup-github file downloads.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed a trailing comma issue in issue triage workflow settings (v0.30.0-nightly.20260226.f9f916e1d)
- Fixed the authentication handshake in the ACP to adhere to the specification.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed robust A2A streaming reassembly and ensured task continuity.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed additional high vulnerabilities related to minimatch and cross-spawn dependencies.(v0.30.0-nightly.20260226.f9f916e1d)
- Fixed a race condition in policy persistence.(v0.30.0)
- Prevented false positives in the hierarchical memory test.(v0.30.0)
- Allowed closing the debug console after auto-open via flicker.(v0.30.0)
- Fixed MCP discovery completion when configured servers are skipped.(v0.30.0)
- Cached the CLI version to ensure consistency during sessions.(v0.30.0)
- Resolved double rendering in shpool and addressed vscode lint warnings.(v0.30.0)
- Fixed behavior when pressing any key to exit select mode.(v0.30.0)
- Updated F12 behavior to only open the drawer if the browser fails.(v0.30.0)
- Ensured that stdio MCP server execution has the GEMINI_CLI=1 env variable populated.(v0.30.0)
- Improved headless mode detection for flags and query arguments.(v0.30.0)
- Fixed maintainer detection for automated PR actions.(v0.30.0)
- Ensured sub-agents are registered regardless of tools.allowed.(v0.30.0)
- Dismissed '?' shortcuts help on hotkeys and active states.(v0.30.0)
- Isolated plan files per session.(v0.30.0)
- Fixed character truncation in raw markdown mode.(v0.30.0)
- Fixed flicker due to AppContainer continuous initialization.(v0.30.0)
- Fixed the bottom border color in the UI.(v0.31.0-preview.0)
- Fixed an issue where the `onConfirm` callback was not supported in `ToolActionsContext`.(v0.31.0-preview.0)
- Fixed a safe fallback issue when the experiment setting for plan mode is disabled but startup approval mode is set to plan.(v0.31.0-preview.0)
- Fixed an issue where empty history items were being added to the UI.(v0.31.0-preview.0)
- Fixed an issue where the directory was not ensured to exist before writing a conversation file.(v0.31.0-preview.0)
- Fixed an issue where unknown slash commands were incorrectly showing an error instead of being treated as regular input.(v0.31.0-preview.0)
- Fixed an issue where ripgrep failed when the search pattern looked like a ripgrep flag.(v0.31.0-preview.0)
- Fixed auto-completion disabling Shift+Tab, preserving mode cycling.(v0.31.0-preview.0)
- Fixed an issue preventing empty history items from being added.(v0.31.0-preview.0)
- Fixed an issue where user rejections were not updating tool outcomes for telemetry.(v0.31.0-preview.0)
- Fixed an issue where the `ClearcutLogger` crashed when `os.cpus()` returned an empty result.(v0.31.0-preview.0)
- Fixed extra padding in the Composer component.(v0.31.0-preview.0)
- Fixed preventing duplicate tool approval entries from being written to `auto-saved.toml`.(v0.31.0-preview.0)
- Fixed an issue where tool confirmation timed out when no UI listeners were present.(v0.31.0-preview.0)
- Fixed an issue where messages that were too large caused problems.(v0.31.0-preview.0)
- Fixed an issue where manual model selection was not persisted across restarts.(v0.30.0-nightly.20260224.544df749a)
- Resolved silent failures occurring when running in non-interactive mode.(v0.30.0-nightly.20260224.544df749a)
- Fixed plan mode writes failing on Windows and corrected prompt paths.(v0.30.0-nightly.20260224.544df749a)
- Prevented an OAuth server crash caused by unexpected requests.(v0.30.0-nightly.20260224.544df749a)
- Fixed state corruption issues in McpClientManager during concurrent operations.(v0.30.0-nightly.20260224.544df749a)
- Fixed an issue where utility calls could unintentionally change the session's active model.(v0.30.0-nightly.20260224.544df749a)
- Skipped workspace policy loading when the CLI is executed from the home directory.(v0.30.0-nightly.20260224.544df749a)
- Added Windows (win32/x64) support to lint.js scripts.(v0.30.0-nightly.20260224.544df749a)
- Removed unsafe type assertions in the agent within the a2a-server component.(v0.30.0-nightly.20260224.544df749a)
- Fixed handling of corrupted token files when switching authentication types.(v0.30.0-nightly.20260224.544df749a)
- Fixed an issue where ask headers longer than 16 characters were disallowed.(v0.30.0-nightly.20260224.544df749a)
- Fixed an issue where environment variable expansion and explicit overrides for MCP servers were not working correctly.(v0.30.0-nightly.20260224.544df749a)
- Hardened image packaging integrity checks within the sandbox environment.(v0.30.0-nightly.20260224.544df749a)
- Fixed a critical dependency vulnerability.(v0.30.0-nightly.20260224.544df749a)
- Fixed the bottom border color display issue.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed an issue where the release note generator failed.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed optimization issues related to height calculations in the `ask_user` dialog.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed support for the legacy `onConfirm` callback in `ToolActionsContext`.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed a fallback issue when the experiment setting for plan mode is disabled but startup approval mode is set to plan.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed core logic to clarify plan mode constraints and the exit mechanism.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed prevention of empty history items from being added to the UI.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed ensuring the conversation directory exists before writing the conversation file.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed UI margin placement in `ToolGroupMessage`, moving it from top to bottom.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed treating unknown slash commands as regular input instead of showing an error.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed an issue where ripgrep failed when the search pattern resembled a ripgrep flag.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed disabling auto-completion on Shift+Tab to correctly preserve mode cycling.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed preventing duplicate tool approval entries from being written to `auto-saved.toml`.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed a crash in `ClearcutLogger` when `os.cpus()` returns an empty result.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed improving error type extraction for telemetry purposes.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed removing extra padding in the Composer component.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed ensuring user rejections correctly update the tool outcome for telemetry.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed initialization of the ACP configuration.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed adding error logging for IDE fetch failures.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed robust workspace-based IDE connection discovery.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed avoiding tool confirmation timeouts when no UI listeners are present.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed an issue causing messages that were too large to be processed.(v0.30.0-nightly.20260223.c537fd5ae)
- Fixed flicker in the CLI caused by continuous AppContainer initialization.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed issues with Vim support, making it feel more complete.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed drag and drop escaping behavior.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed an issue where the question type was not required in the AskUser tool.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed an issue where the --yolo flag incorrectly forced headless mode.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed loop detection false positives occurring on lists with long shared prefixes.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed inconsistent commandRegex example in policy engine documentation.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed an issue where the approval mode was not persisted in the UI while the agent was thinking.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed escaping issues on Windows and corrected broken tests related to it.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed unsafe type assertion lint errors in VSCode integration.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed flicker when refreshing the background color in the UI.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed GitHub App token permissions for maintainer detection in workflows.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed hook integration test flakiness observed on Windows CI.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed propagating the User-Agent header to setup-phase CodeAssist API calls.(v0.30.0-nightly.20260218.ce84b3cb5)
- Fixed telemetry file exporters by replacing stringify with safeJsonStringify.(v0.30.0-nightly.20260218.ce84b3cb5)
- Removed the `ask_user` tool from non-interactive modes to prevent unexpected prompts.(v0.29.0)
- Fixed an issue where the agent would get stuck when encountering interactive commands.(v0.29.0)
- Fixed an issue where restricted .env loading was disallowed in untrusted sandboxed folders.(v0.29.0)
- Removed unnecessary logging related to hook registration.(v0.29.0)
- Ensured the MCP transport is properly closed to prevent potential memory leaks.(v0.29.0)
- Added retry logic for transient SSL/TLS connection errors.(v0.29.0)
- Fixed handling of non-compliant mcpbridge responses originating from Xcode 26.3.(v0.29.0)
- Fixed Windows-specific agent quality and system prompt issues.(v0.29.0)
- Fixed invalid tool calls occurring within plan mode.(v0.29.0)
- Fixed a permission check issue.(v0.29.0)
- Resolved an infinite loop that occurred when using the 'Modify with external editor' feature.(v0.29.0)
- Fixed focus navigation issues for interactive and background shells in the CLI.(v0.29.0)
- Fixed a configuration issue where system settings were treated as writable during migration; now warns the user instead.(v0.29.0)
- Fixed a bug preventing subagent bypass when operating in plan mode.(v0.29.0)
- Fixed a bug related to saving plans.(v0.29.0)
- Ensured storage is initialized early, even if configuration loading fails.(v0.29.0)
- Fixed an issue on Windows where `shell: true` was missing for `spawnSync`, causing EINVAL errors with .cmd editors.(v0.29.0)
- Fixed race condition in policy persistence.(v0.30.0-preview.0)
- Prevented false positive in hierarchical memory test.(v0.30.0-preview.0)
- Allowed closing debug console after auto-open via flicker.(v0.30.0-preview.0)
- Completed MCP discovery when configured servers are skipped.(v0.30.0-preview.0)
- Cached CLI version to ensure consistency during sessions.(v0.30.0-preview.0)
- Resolved double rendering in shpool and addressed vscode lint warnings.(v0.30.0-preview.0)
- Fixed behavior when pressing any key to exit select mode.(v0.30.0-preview.0)
- Updated F12 behavior to only open drawer if browser fails.(v0.30.0-preview.0)
- Ensured that stdio MCP server execution has the `GEMINI_CLI=1` env variable populated.(v0.30.0-preview.0)
- Improved headless mode detection for flags and query arguments.(v0.30.0-preview.0)
- Improved maintainer detection for automated PR actions.(v0.30.0-preview.0)
- Ensured sub-agents are registered regardless of tools.allowed.(v0.30.0-preview.0)
- Dismissed '?' shortcuts help on hotkeys and active states.(v0.30.0-preview.0)
- Prioritized conditional policy rules and hardened Plan Mode.(v0.30.0-preview.0)
- Isolated plan files per session.(v0.30.0-preview.0)
- Fixed character truncation in raw markdown mode.(v0.30.0-preview.0)
- Fixed flicker due to AppContainer continuous initialization.(v0.30.0-preview.0)
- Fixed character truncation issues when operating in raw markdown mode.(v0.30.0-nightly.20260212.207ac6f2d)
- Fixed an issue where the '?' shortcuts help was not being dismissed correctly when hotkeys were active or in certain states.(v0.30.0-nightly.20260212.207ac6f2d)
- Fixed Plan Mode to isolate plan files per session, preventing cross-session interference.(v0.30.0-nightly.20260212.207ac6f2d)
- Fixed an issue where conditional policy rules were not being prioritized correctly; hardened Plan Mode execution.(v0.30.0-nightly.20260212.207ac6f2d)
- Fixed an issue where pressing any key would exit select mode.(v0.30.0-nightly.20260212.099aa9621)
- Resolved double rendering issues in shpool and addressed VS Code lint warnings.(v0.30.0-nightly.20260212.099aa9621)
- Fixed F12 behavior to only open the drawer if the browser fails to open.(v0.30.0-nightly.20260212.099aa9621)
- Ensured that the stdio MCP server execution correctly populates the GEMINI_CLI=1 environment variable.(v0.30.0-nightly.20260212.099aa9621)
- Improved headless mode detection for flags and query arguments.(v0.30.0-nightly.20260212.099aa9621)
- Fixed an issue where sub-agents were not being registered when tools.allowed was set.(v0.30.0-nightly.20260212.099aa9621)
- Fixed an issue where the debug console could not be closed after being automatically opened due to a flicker.(v0.30.0-nightly.20260211.f5dd1068f)
- Resolved an issue preventing complete MCP discovery when configured servers were skipped.(v0.30.0-nightly.20260211.f5dd1068f)
- Fixed focus navigation for interactive and background shells.(v0.30.0-nightly.20260210.8257ec447)
- Fixed crash when migrating system settings by treating them as read-only and warning the user.(v0.30.0-nightly.20260210.8257ec447)
- Fixed crash on Windows when using editors with .cmd extensions by adding shell: true for spawnSync.(v0.30.0-nightly.20260210.8257ec447)
- Fixed bug when saving plans.(v0.30.0-nightly.20260210.8257ec447)
- Fixed issue where storage was not initialized early if configuration was missing.(v0.30.0-nightly.20260210.8257ec447)
- Prevented subagent bypass in plan mode.(v0.30.0-nightly.20260210.8257ec447)
- Fixed reload of skills and agents upon extension restart.(v0.30.0-nightly.20260210.8257ec447)
- Expanded excludeTools to include legacy aliases for renamed tools.(v0.30.0-nightly.20260210.8257ec447)
- Fixed focus navigation for interactive and background shells.(v0.30.0-nightly.20260210.8257ec447)
- Fixed SS3 Shift+Tab support for Windows terminals.(v0.30.0-nightly.20260210.8257ec447)
- Corrected escaped interpolation in the system prompt.(v0.30.0-nightly.20260210.8257ec447)
- Ensured the `enter_plan_mode` tool registration respects the `experimental.plan` setting.(v0.30.0-nightly.20260210.8257ec447)
- Fixed newline insertion bug in the replace tool.(v0.30.0-nightly.20260210.8257ec447)
- Fixed shortened tool call IDs and resolved duplicate tool names in truncated output filenames.(v0.30.0-nightly.20260210.8257ec447)
- Fixed linting support for cjs files.(v0.30.0-nightly.20260210.8257ec447)
- Fixed crash when getText() was called with undefined instead of a string by adding "" as a default value.(v0.30.0-nightly.20260210.8257ec447)
- Fixed verification of audio/video MIME types using content checks.(v0.28.0)
- Suppressed duplicate hook failure warnings that occurred during streaming.(v0.28.0)
- Fixed unintended credential exposure to MCP Servers.(v0.28.0)
- Fixed persistence and restoration of workspace directories upon session resume.(v0.28.0)
- Fixed wrapping issues caused by fixed-width characters in the UX.(v0.28.0)
- Fixed activity log directory path in evals to use an absolute path.(v0.28.0)
- Fixed listing of installed extensions when an update target is missing.(v0.28.0)
- Fixed handling of PAT tokens and credentials during git remote URL parsing.(v0.28.0)
- Fixed error display by using returnDisplay for error results.(v0.28.0)
- Fixed detection of 'bun' as a package manager.(v0.28.0)
- Prevented a ghost border from appearing for the AskUserDialog.(v0.28.0)
- Resolved an error thrown when processing sensitive values.(v0.28.0)
- Fixed an issue where GIT_CONFIG_GLOBAL was not used to isolate the shadow git repo configuration.(v0.28.0)
- Fixed the Cloud Shell authentication setting key.(v0.28.0)
- Fixed the HalfLinePaddedBox rendering issue in screenreader mode.(v0.28.0)
- Fixed Vim mode issues, ensuring it starts in insert mode and unblocking F12 and ctrl-X.(v0.28.0)
- Removed the `ask_user` tool from non-interactive modes to prevent unexpected prompts.(v0.29.0-preview.0)
- Fixed an issue where the agent would get stuck when encountering interactive commands.(v0.29.0-preview.0)
- Fixed an issue where restricted .env loading was disallowed in untrusted sandboxed folders.(v0.29.0-preview.0)
- Removed unnecessary logging related to hook registration.(v0.29.0-preview.0)
- Ensured the MCP transport is properly closed to prevent memory leaks.(v0.29.0-preview.0)
- Added retry logic for transient SSL/TLS errors during communication.(v0.29.0-preview.0)
- Fixed handling of non-compliant mcpbridge responses originating from Xcode 26.3.(v0.29.0-preview.0)
- Fixed Windows-specific agent quality and system prompt issues.(v0.29.0-preview.0)
- Fixed invalid tool calls occurring within plan mode.(v0.29.0-preview.0)
- Fixed a permission check issue.(v0.29.0-preview.0)
- Resolved an infinite loop that occurred when using the 'Modify with external editor' feature.(v0.29.0-preview.0)
- Fixed focus navigation issues for interactive and background shells in the CLI.(v0.29.0-preview.0)
- Fixed an issue where system settings were treated as writable during migration, now warning the user instead.(v0.29.0-preview.0)
- Fixed a bug preventing the saving of plans.(v0.29.0-preview.0)
- Ensured storage is initialized early, even if configuration is missing.(v0.29.0-preview.0)
- Prevented subagent bypass when operating in plan mode.(v0.29.0-preview.0)
- Added retry logic to handle transient SSL/TLS errors during operations.(v0.29.0-nightly.20260206.4ffc349c1)
- Fixed handling of non-compliant mcpbridge responses originating from Xcode 26.3.(v0.29.0-nightly.20260206.4ffc349c1)
- Fixed issues related to Windows Specific Agent Quality and System Prompt configuration.(v0.29.0-nightly.20260206.4ffc349c1)
- Fixed invalid tool calls that occurred when operating in plan mode.(v0.29.0-nightly.20260206.4ffc349c1)
- Fixed a permission check issue.(v0.29.0-nightly.20260206.4ffc349c1)
- Resolved an infinite loop that occurred when using the 'Modify with external editor' feature.(v0.29.0-nightly.20260206.4ffc349c1)
- Fixed a regression affecting up/down arrow navigation and added a corresponding test.(v0.29.0-nightly.20260205.a3af4a8ca)
- Prevented content leak in `MaxSizedBox` bottom overflow issues.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where the agent would get stuck at interactive commands.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where the agent was not respecting the user's `.gitignore` preference.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed environment loading and authentication validation issues when running in ACP mode.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where the `ask_user` tool was incorrectly used in non-interactive modes.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where detailed error messages were not prioritized during code assist setup.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where Google environment variables (`GOOGLE_GEMINI_BASE_URL` & `GOOGLE_VERTEX_BASE_URL`) were not propagated correctly in the sandbox.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed handling of empty settings.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where folder trust was not enforced for workspace settings, skills, and context.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue allowing restricted `.env` loading in untrusted sandboxed folders.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed an issue where an error message was shown in the `/plan` command.(v0.29.0-nightly.20260205.a3af4a8ca)
- Fixed efficiency and bugs related to the label-workstream-rollup GitHub feature.(v0.27.0)
- Ensured CLI commands run correctly in non-interactive mode for admin tasks.(v0.27.0)
- Fixed an issue where recursive directory access was not enabled.(v0.27.0)
- Fixed crash when initial authentication fails.(v0.27.0)
- Resolved layout jitter in the settings bar.(v0.27.0)
- Fixed the edit tool diffs to correctly show whitespace changes.(v0.27.0)
- Fixed the ide-client connection to ide-companion when operating inside Docker via SSH/devcontainer.(v0.27.0)
- Resolved AbortSignal MaxListenersExceededWarning in the core.(v0.27.0)
- Fixed the /policy command to display policies according to the current mode.(v0.27.0)
- Fixed the scheduler to prevent stale tool re-publication and stuck UI states.(v0.27.0)
- Fixed token count and telemetry updates when resuming chat history load.(v0.27.0)
- Fixed the tool replace error message to be simpler.(v0.27.0)
- Fixed rewind to start at the bottom and ensured loadHistory refreshes static content.(v0.27.0)
- Fixed verification of audio/video MIME types using content checks.(v0.28.0-preview.0)
- Fixed crash or incorrect behavior related to streaming by suppressing duplicate hook failure warnings.(v0.28.0-preview.0)
- Fixed persistence and restoration of workspace directories upon session resume.(v0.28.0-preview.0)
- Fixed wrapping issues caused by fixed-width characters in the UX.(v0.28.0-preview.0)
- Fixed incorrect path usage for the activity log directory in evaluations.(v0.28.0-preview.0)
- Fixed listing of installed extensions when an update target is missing.(v0.28.0-preview.0)
- Fixed handling of PAT tokens and credentials during git remote URL parsing.(v0.28.0-preview.0)
- Fixed error display by using returnDisplay for error results.(v0.28.0-preview.0)
- Fixed detection of 'bun' as a package manager.(v0.28.0-preview.0)
- Prevented a ghost border from appearing for the AskUserDialog.(v0.28.0-preview.0)
- Fixed unintended credential exposure to MCP Servers.(v0.28.0-preview.0)
- Fixed an error thrown for sensitive values.(v0.28.0-preview.0)
- Fixed the use of GIT_CONFIG_GLOBAL to isolate the shadow git repo configuration.(v0.28.0-preview.0)
- Fixed the Cloud Shell auth setting key.(v0.28.0-preview.0)
- Fixed HalfLinePaddedBox rendering in screenreader mode.(v0.28.0-preview.0)
- Fixed Vim mode issues, ensuring it starts in insert mode and unblocking F12 and ctrl-X.(v0.28.0-preview.0)
- Fixed an issue where URI-encoded workspace paths were not handled correctly in IdeClient.(v0.27.0-nightly.20260127.00f60ef53)
- Fixed a bug related to detecting paths that were already added.(v0.27.0-nightly.20260127.00f60ef53)
- Fixed streaming issue for grep/ripgrep output, preventing Out-of-Memory (OOM) errors.(v0.27.0-nightly.20260127.00f60ef53)
- Fixed an issue where subagent invocations were incorrectly included in plan mode.(v0.27.0-nightly.20260127.00f60ef53)
- Fixed an error related to extensions configuration.(v0.27.0-nightly.20260127.00f60ef53)
- Fixed crash when initial authentication fails.(v0.27.0-nightly.20260126.cb772a5b7)
- Resolved layout jitter in the CLI settings bar.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed showing whitespace changes correctly in edit tool diffs.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed IDE-client connection issues to ide-companion when running inside Docker via SSH/devcontainer.(v0.27.0-nightly.20260126.cb772a5b7)
- Resolved AbortSignal MaxListenersExceededWarning.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed rewind command to start at the bottom and ensure loadHistory refreshes static content.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed UI stabilization during terminal resize in alternate buffer mode.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed default behavior to use interactive mode for positional arguments.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed including source information in policy rule display.(v0.27.0-nightly.20260126.cb772a5b7)
- Removed obsolete CloudCode PerDay quota and the 120s terminal threshold.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed including the MCP server name in the OAuth message.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed the pr-triage.sh script to correctly update pull requests with tags "help wanted" and "maintainer only".(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed the exit code to be non-zero when esbuild is missing.(v0.27.0-nightly.20260126.cb772a5b7)
- Ensured correct ordering and testing for @-command UI messages.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed awaiting MCP initialization in non-interactive mode.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed gracefully handling disk full errors during chat recording.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed OAuth to use 127.0.0.1 instead of localhost.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed using RFC 9728 compliant path-based OAuth protected resource discovery.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed preserving input text when declining tool approval.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed the image paste location to use the global temp directory.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed line endings issue with the Notice file.(v0.27.0-nightly.20260126.cb772a5b7)
- Fixed newline support broken in the previous release in the CLI.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed home/end keybinding conflicts in the CLI.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed the display of 'http' type on the mcp list.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed bad fallback logic for external editor operations.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed a bug where System scopes were not migrated correctly.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed the lookup of mcp tools in the tool registry.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed spurious warnings about unexpected renders triggered by appEvents and coreEvents.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed efficiency and bugs related to the label-workstream-rollup GitHub integration.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed an issue preventing CLI commands from running correctly in non-interactive mode.(v0.27.0-nightly.20260122.61040d0eb)
- Enabled recursive directory access for core and CLI components.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed token count and telemetry updates when resuming chat history load.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed the /policy command to display policies according to the current mode.(v0.27.0-nightly.20260122.61040d0eb)
- Simplified the replace tool error message in the core.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed the scheduler to prevent stale tool re-publication and stuck UI states.(v0.27.0-nightly.20260122.61040d0eb)
- Fixed UI issue where rationale was not rendering before tool calls.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed maintainer check in workflows to use author_association.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed return type of fireSessionStartEvent to defaultHookOutput.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed deduplication of ModelInfo emission in GeminiClient.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed update of currentSequenceModel when the model changes.(v0.26.0-nightly.20260121.211d2c5fd)
- Provided compact, actionable errors for agent delegation failures.(v0.26.0-nightly.20260121.211d2c5fd)
- Migrated BeforeModel and AfterModel hooks to the HookSystem.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed telemetry token count update after session resume.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed Windows crash related to resizing a pty that has already exited.(v0.26.0-nightly.20260121.211d2c5fd)
- Resolved auto model selection in the default strategy.(v0.26.0-nightly.20260121.211d2c5fd)
- Resolved Ctrl+Enter and Ctrl+J newline issues in the CLI.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed sending the gemini-cli version as the mcp client version.(v0.26.0-nightly.20260121.211d2c5fd)
- Fixed an issue with overrides to bases in the core.(v0.25.0)
- Fixed the spewie getter in `hookTranslator.ts`.(v0.25.0)
- Fixed a typo in the MCP servers settings description.(v0.25.0)
- Fixed Yolo mode so it auto allows redirection.(v0.25.0)
- Fixed an issue where `disableYoloMode` incorrectly enforced default approval mode against arguments.(v0.25.0)
- Fixed CI for forks.(v0.25.0)
- Fixed and renamed the introspection agent to cli help agent.(v0.25.0)
- Fixed an issue where the pr-triage.sh script used incorrect and limited labels for PRs.(v0.25.0)
- Fixed PDF token estimation calculation.(v0.26.0-preview.0)
- Prevented errors from being thrown for GeminiEventType Retry and InvalidStream in A2A communication.(v0.26.0-preview.0)
- Ensured exit cleanup runs when stdin closes in ACP.(v0.26.0-preview.0)
- Replaced three consecutive periods with the proper ellipsis character.(v0.26.0-preview.0)
- Fixed handling of colons within skill description frontmatter.(v0.26.0-preview.0)
- Resolved a circular dependency issue via tsconfig paths in the core.(v0.26.0-preview.0)
- Corrected status/need-triage label application to ensure it is never cleared automatically.(v0.26.0-preview.0)
- Corrected status/need-issue label matching wildcard in automation.(v0.26.0-preview.0)
- Prevented label-enforcer loops by ignoring all bots during label enforcement.(v0.26.0-preview.0)
- Fixed a race condition by correctly awaiting scheduleToolCalls.(v0.26.0-preview.0)
- Prevented ModelInfo event emission on aborted signals in the core.(v0.26.0-preview.0)
- Fixed OOM crash risk by limiting file search traversal and adding a timeout in the CLI.(v0.26.0-preview.0)
- Safely handled /dev/tty access on macOS.(v0.26.0-preview.0)
- Fixed a dependency issue by explicitly adding 'color-convert' in the CLI.(v0.26.0-preview.0)
- Fixed infrastructure issue by updating the maintainer rollup label to 'workstream-rollup'.(v0.26.0-preview.0)
- Fixed infrastructure issue by using GraphQL to detect direct parents in the rollup workflow.(v0.26.0-preview.0)
- Fixed label enforcement robustness by including permission checks in automation.(v0.26.0-preview.0)
- Fixed a crash that could occur when aborting a ModelInfo event emission.(v0.26.0-nightly.20260119.20580d754)
- Prevented an infinite loop in the label-enforcer automation by ignoring all bots.(v0.26.0-nightly.20260119.20580d754)
- Fixed robust label enforcement logic to correctly incorporate permission checks.(v0.26.0-nightly.20260119.20580d754)
- Prevented an Out-Of-Memory (OOM) crash by limiting file search traversal depth and adding a timeout.(v0.26.0-nightly.20260119.20580d754)
- Safely handled access to `/dev/tty` on macOS environments.(v0.26.0-nightly.20260119.20580d754)
- Fixed a shell descriptor leak in the PTY handling core component.(v0.26.0-nightly.20260119.20580d754)
- Truncated large telemetry log entries to prevent excessive logging.(v0.26.0-nightly.20260119.20580d754)
- Surfaced configuration warnings for invalid hook event names.(v0.26.0-nightly.20260119.20580d754)
- Enabled the `/hooks disable` command to reliably stop single hooks.(v0.26.0-nightly.20260119.20580d754)
- Fixed a race condition by ensuring `scheduleToolCalls` is properly awaited.(v0.26.0-nightly.20260119.20580d754)
- Fixed OAuth redirect port stabilization and resolved a PKCE length issue.(v0.26.0-nightly.20260119.20580d754)
- Fixed an issue where experiment values, specifically `compressionThreshold`, were not showing in the settings UI.(v0.26.0-nightly.20260119.20580d754)
- Fixed an issue where the 'activate_skill' command would incorrectly trigger a re-registration warning.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Restored fastreturn support.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Fixed an issue where the CLI would set only the window title instead of both the tab and window title.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Ensured MCP policies correctly match unqualified names when running in non-interactive mode.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Refined the 'Action Required' indicator and focus hints in the CLI.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Fixed an unknown argument error when using 'gemini skills install'.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Allowed the @ file selector to work correctly on slash command lines.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Resolved a sticky header regression that occurred in tool messages.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Ensured OSC52 copy functionality is only used when operating in SSH or WSL environments.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Fixed an issue where ctrl+x did not use the preferred editor.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Resolved a race condition that occurred during tool response reporting.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Prevented errors from being thrown for GeminiEventType Retry and InvalidStream in the a2a context.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Fixed PDF token estimation calculations.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Corrected an issue where the 'gemini skills install' command threw an unknown argument error.(v0.26.0-nightly.20260115.6cb3ae4e0)
- Fixed a crash that occurred when cleaning up unhandled EIO errors during stdin read operations.(v0.24.0)
- Fixed missing `type` field in MCPServerConfig.(v0.24.0)
- Resolved an unhandled promise rejection in ide-client.ts.(v0.24.0)
- Fixed failure to check if a directory is a repository in GitService.initialize.(v0.24.0)
- Fixed an issue where the `enableShellOutputEfficiency` setting was missing from the settings schema.(v0.24.0)
- Fixed the policy engine by enabling granular shell command allowlisting.(v0.24.0)
- Deduplicated agent hooks and added cross-platform integration tests.(v0.24.0)
- Fixed a build issue caused by a year-specific linter rule.(v0.24.0)
- Fixed an unhandled promise rejection in mcp-client-manager.(v0.24.0)
- Improved shell command detection when dealing with redirection.(v0.24.0)
- Mocked powershell output in shell-utils tests to ensure stability.(v0.24.0)
- Fixed an issue with overrides to bases in the core functionality.(v0.25.0-preview.0)
- Fixed the spewie getter in `hookTranslator.ts`.(v0.25.0-preview.0)
- Fixed a typo in the MCP servers settings description.(v0.25.0-preview.0)
- Fixed an issue where yolo should auto allow redirection.(v0.25.0-preview.0)
- Fixed an issue where `disableYoloMode` incorrectly enforced default approval mode against arguments.(v0.25.0-preview.0)
- Fixed an issue with the `pr-triage.sh` script limiting labels.(v0.25.0-preview.0)
- Fixed CI failures when running on forks.(v0.25.0-preview.0)
- Fixed an issue where the telemetry.js script failed to handle paths containing spaces.(v0.25.0-preview.0)
- Fixed a terminal hang that occurred when a user exited the browser without logging in.(v0.25.0-nightly.20260112.15891721a)
- Fixed an issue where the SDK warning was triggered by accessing the .text getter during logging.(v0.25.0-nightly.20260112.15891721a)
- Prevented a crash when using the /copy command on Windows by skipping /dev/tty.(v0.25.0-nightly.20260112.15891721a)
- Resolved a paste issue specifically on Windows terminals.(v0.25.0-nightly.20260112.15891721a)
- Fixed the order of preflight checks.(v0.25.0-nightly.20260112.15891721a)
- Fixed an issue where the writeTodo construction was incorrect.(v0.25.0-nightly.20260112.15891721a)
- Fixed image token estimation calculations.(v0.25.0-nightly.20260112.15891721a)
- Fixed an issue where broken skills objects were incorrectly displayed in the settings dialog.(v0.25.0-nightly.20260112.15891721a)
- Fixed scheduled issue triage queries that were causing an "argument list too long" error by limiting the queries.(v0.25.0-nightly.20260112.15891721a)
- Fixed the /settings command fallback mechanism.(v0.25.0-nightly.20260112.15891721a)
- Fixed an issue where keyboard modes were not properly disabled upon exit.(v0.25.0-nightly.20260112.15891721a)
- Fixed an issue where eager slash command completion hid sibling commands.(v0.23.0)
- Corrected initial history length handling for chat commands.(v0.23.0)
- Prevented an infinite loop during prompt completion when an error occurred.(v0.23.0)
- Fixed an issue where showing command suggestions was suppressed even on a perfect match; suggestions are now shown and sorted correctly.(v0.23.0)
- Fixed hook command expansion to sanitize input and prevent injection vulnerabilities.(v0.23.0)
- Added validation for the trusted folder level in folder trust settings.(v0.23.0)
- Fixed a right border overflow issue in trust dialogs.(v0.23.0)
- Fixed a bug where accepting edits continued even after the feature was turned off.(v0.23.0)
- Prevented an infinite relaunch loop when the `--resume` flag failed.(v0.23.0)
- Fixed the placeholder text displayed during shell mode input.(v0.23.0)
- Fixed a crash that occurred when cleaning up unhandled EIO errors during stdin reading.(v0.24.0-preview.0)
- Fixed missing `type` field in MCPServerConfig.(v0.24.0-preview.0)
- Resolved an unhandled promise rejection in ide-client.ts.(v0.24.0-preview.0)
- Fixed failure in GitService.initialize when checking if the current directory is a repository.(v0.24.0-preview.0)
- Fixed an issue where the `enableShellOutputEfficiency` setting was missing from the settings schema.(v0.24.0-preview.0)
- Fixed the policy engine by enabling granular shell command allowlisting.(v0.24.0-preview.0)
- Fixed deduplication issues for agent hooks and added cross-platform integration tests.(v0.24.0-preview.0)
- Fixed a build issue caused by a year-specific linter rule.(v0.24.0-preview.0)
- Fixed an unhandled promise rejection in mcp-client-manager.(v0.24.0-preview.0)
- Improved shell command detection when handling redirection.(v0.24.0-preview.0)
- Fixed mocking of powershell output in shell-utils tests.(v0.24.0-preview.0)
- Fixed build issues that were caused by a year-specific linter rule.(v0.24.0-nightly.20260103.30f5c4af4)
- Fixed an unhandled promise rejection error occurring within the mcp-client-manager.(v0.24.0-nightly.20260103.30f5c4af4)
- Improved shell command handling to correctly detect redirection.(v0.24.0-nightly.20260103.30f5c4af4)
- Fixed an issue where PowerShell output was incorrectly mocked in shell-utils tests.(v0.24.0-nightly.20260103.30f5c4af4)
- Fixed an issue where the SearchText tool did not respect the contents of the .geminiignore file.(v0.21.0-nightly.20251226.546baf993)
- Prevented a crash that occurred during readStdin cleanup when encountering an unhandled EIO error.(v0.21.0-nightly.20251226.546baf993)
- Fixed missing 'type' field in MCPServerConfig.(v0.21.0-nightly.20251226.546baf993)
- Made schema validation errors non-fatal, preventing them from halting operations.(v0.21.0-nightly.20251226.546baf993)
- Fixed a typo in a code comment.(v0.22.0)
- Correctly detected MCP tool errors.(v0.22.0)
- Updated the IDE extension to write the auth token in the environment variable.(v0.22.0)
- Corrected license generation for workspace dependencies in the vscode-ide-companion.(v0.22.0)
- Applied a temporary fix for subagent invocation until subagent delegation is merged to stable.(v0.22.0)
- Fixed commandPrefix word boundary and compound command safety issues.(v0.22.0)
- Fixed authentication logic to prioritize GEMINI_API_KEY environment variable and skip unnecessary key checks.(v0.22.0)
- Used Zod for safety check result validation.(v0.22.0)
- Fixed tool output fragmentation by encapsulating content in functionResponse.(v0.22.0)
- Marked A2A requests as interactive.(v0.22.0)
- Fixed settings merging so that settings using the new JSON format take priority over older formats.(v0.22.0)
- Fixed an issue where eager slash command completion was hiding sibling commands.(v0.23.0-preview.0)
- Corrected initial history length handling for chat commands.(v0.23.0-preview.0)
- Prevented an infinite loop during prompt completion when an error occurred.(v0.23.0-preview.0)
- Fixed an issue where command suggestions were not shown on a perfect match and improved sorting.(v0.23.0-preview.0)
- Corrected the placeholder text displayed during shell mode input.(v0.23.0-preview.0)
- Fixed an issue where accepting edits continued even after the feature was turned off.(v0.23.0-preview.0)
- Prevented an infinite relaunch loop when the `--resume` flag failed.(v0.23.0-preview.0)
- Fixed an issue where hook command expansion was not sanitized, preventing potential injection.(v0.23.0-preview.0)
- Added validation for the trusted folder level in folder trust settings.(v0.23.0-preview.0)
- Fixed a right border overflow issue observed in trust dialogs.(v0.23.0-preview.0)
- Fixed a security vulnerability by sanitizing hook command expansion to prevent injection.(v0.21.0-nightly.20251220.41a1a3eed)
- Corrected the input placeholder text displayed when using shell mode.(v0.21.0-nightly.20251220.41a1a3eed)
Releases with MCP Changes
v0.40.0-nightly.20260415.g06e7621b22 features2 fixesThis release introduces new capabilities for plan mode prompts and improves CLI interaction by enabling mouse clicking for cursor positioning in multi-line answers. Key fixes include improved handling of OpenSSL 3.x streaming errors and better terminal color detection.
v0.38.017 features21 fixesThis release introduces several new features, including enabling 'TerminalBuffer' mode to reduce flicker, adding support for default environment variable values, and implementing a background memory service for skill extraction. Numerous bug fixes address stability issues across the CLI and core, particularly around sandbox reliability on Linux and Windows, and improved chat history recording. Users will also benefit from UI refinements like compact tool output being enabled by default and better handling of input prompts.
v0.39.0-preview.013 features20 fixesThis release introduces significant stability and capability enhancements, including a new /memory inbox command for skill review and required user confirmation for Plan Mode skill activation. Numerous bug fixes address critical issues related to sandbox cleanup, session resumption, and memory leaks across core components. Users will also benefit from improved CLI usability, such as word deletion support in Windows Terminal and refined tool output formatting.
v0.39.0-nightly.20260414.gdaf5006235 features6 fixesThis release introduces significant architectural updates, including a decoupled ContextManager and Sidecar, and new capabilities like skill patching integration. Several critical bugs were resolved, such as fixing an infinite render loop and improving UTF-8 file handling. Users will also notice that model responses now fully display thoughts and text without suppression.
v0.39.0-nightly.20260411.0957f7d3e2 features9 fixesThis release introduces new guidance for Strategic Re-evaluation in the system prompt and ensures subagent IDs are persisted in tool call records. Numerous bug fixes address critical issues like memory leaks, OOM errors on large outputs, and data corruption during text sanitization. Performance is also improved, particularly on Windows sandbox initialization.
v0.39.0-nightly.20260408.a394617187 features10 fixesThis release introduces several new features, including the "/help" command, experimental Agent Protocol UI types, and memory management commands like "/memory inbox". Several critical bugs were fixed, such as issues with hook propagation, Windows skill linking, and memory leaks in the scheduler. Additionally, user experience improvements include better sandbox error handling and a switch in the default command shortcut from Ctrl+X to Ctrl+G.
v0.37.011 features22 fixesThis release introduces significant new capabilities, including experiment-gated support for gemini flash 3.1 lite and new skills for code duplication checking and automated failure replication. Numerous bug fixes stabilize the CLI experience, particularly around browser agent security, TTY handling in headless environments, and improved error reporting for agent loading.
v0.38.0-preview.018 features23 fixesThis release introduces several new features focused on enhancing agent capabilities, including context-aware policy approvals and a background memory service for skill extraction. Numerous bug fixes address stability issues across the CLI and core, particularly around sandbox reliability on Linux and Windows. Users will also benefit from UI improvements like the new 'TerminalBuffer' mode to reduce flicker and compact tool output being enabled by default.
v0.36.0-nightly.20260402.13ccc164512 features22 fixesThis release introduces significant new features, including a behavioral-evals skill, support for multi-turn interactive evals, and a refreshed UX for the Composer layout. Numerous bug fixes address stability, including resolving terminal escape sequence leaks and improving sandbox path resolution in Plan Mode. Performance has also been boosted by parallelizing startup cleanup tasks and backgrounding the IDE client.
v0.36.018 features17 fixesThis release introduces significant architectural updates, including multi-registry support and tool filtering for subagents, alongside an experimental memory manager agent. Key fixes address issues with context handling, session logging, and UI rendering, while new features enhance plan mode usability and macOS sandboxing security.
v0.37.0-preview.010 features18 fixesThis 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.
v0.35.014 features17 fixesThis release introduces significant new customization options, including user-configurable keyboard shortcuts and support for custom base URLs. Several Vim editing features have been added to the UI, alongside core stability fixes like improved API error retries and better process cleanup. Users will also notice better visibility into tool execution status with pending tool calls now being displayed.
v0.36.0-preview.018 features18 fixesThis release introduces significant architectural updates, including multi-registry support and enhanced subagent capabilities like local execution and memory injection. Key fixes address issues with path expansion, session handling, and UI rendering stability across various terminal sizes. Users will benefit from improved context awareness, better sandboxing on macOS, and new features like plan mode support in non-interactive sessions.
v0.36.0-nightly.20260321.fc03891a119 features13 fixesThis release introduces significant enhancements to agent execution, including local execution and tool isolation for subagents, alongside native sandboxing support for both macOS and Windows. Key fixes address issues with subagent state persistence and UI rendering, while new features enable plan mode in non-interactive sessions and add Git worktree support for parallel development.
v0.36.0-nightly.20260318.e2658ccda12 features18 fixesThis release introduces major new capabilities, including the full enablement of subagents and the default activation of JIT context loading. Security and reliability have been enhanced with cryptographic integrity verification for extensions and sandboxing for process-spawning tools. Numerous bug fixes address issues related to configuration saving, terminal input handling, and JIT context operations.
v0.34.017 features20 fixesThis release introduces significant new capabilities, including experimental LXC and native gVisor sandboxing support, and new CLI features like chat resume and custom footers. Numerous bug fixes address stability, particularly around OAuth handling, model persistence, and UI rendering issues.
v0.35.0-preview.114 features19 fixesThis release introduces significant enhancements to user customization, including customizable keyboard shortcuts and support for new Vim motions and keybindings in the CLI. Several core stability fixes address issues related to process handling, API error retries, and Windows compatibility. Users will also benefit from improved telemetry and better handling of tool calls and session resumes.
v0.35.0-nightly.20260313.bb060d7a917 features18 fixesThis release introduces significant new capabilities, including OAuth2 Authorization Code support for A2A agents and customizable keyboard shortcuts. Key fixes address stability issues like preventing prompt jumping and race conditions during retries. Users will also benefit from new Vim mode motions and improved reporting of per-model token usage in stream-json output.
v0.33.016 features16 fixesThis release introduces several new capabilities, including the addition of the github-issue-creator skill and support for HTTP authentication for A2A remote agents. Several critical bugs were fixed, such as resolving race conditions in ACP/Zed and reducing false positives in LLM loop detection. Users will also benefit from UX improvements like a redesigned compact header and better context display for the context window usage.
v0.34.0-preview.015 features18 fixesThis release introduces significant new capabilities, including experimental LXC and native gVisor sandboxing support, along with new CLI features like chat session resumption and custom footer configuration. Numerous bug fixes address stability, including increased agent timeouts and preventing hangs during OAuth in non-interactive sessions. Users will also benefit from improved UI styling and better filename prioritization in autocompletion.
v0.34.0-nightly.20260310.4653b126f14 features13 fixesThis release introduces several new CLI features, including enabling Plan Mode by default, adding the /upgrade command, and unifying the /chat and /resume UX. Numerous bug fixes address stability issues related to memory usage, environment variables, and path resolution across platforms. Performance has also been improved through increased caching of API key and settings loads.
v0.34.0-nightly.20260307.6c3a906456 features9 fixesThis release introduces several core enhancements, including a unified KeychainService for token storage and a generic CacheService to optimize user setup. Several bugs related to task tracking, chat compression, and file handling on case-insensitive filesystems have been resolved. Users will also benefit from dynamically generated keybinding hints and improved accessibility support in the RewindViewer.
v0.30.0-nightly.20260226.f9f916e1d18 features11 fixesThis release introduces significant new capabilities, including the foundation for a task tracker, support for RuntimeHook functions, and the ability to steer models within the workspace. Several fixes address critical issues like unsafe type assertions and dependency vulnerabilities. Users will also benefit from parallel extension loading and improved Plan Mode features like automatic model switching.
v0.30.020 features16 fixesThis release introduces significant enhancements to Plan Mode, including a formalized 5-phase workflow and the ability to enable skills within plans. Users will benefit from new UI controls like text wrapping in markdown tables and the option to hide the shortcuts hint. Numerous bug fixes address stability issues related to policy persistence, MCP discovery, and UI rendering flicker.
v0.31.0-preview.014 features15 fixesThis release introduces several new features, including support for macOS run-event notifications, experimental steering hints, and project-level policy implementation. Key fixes address issues with tool confirmation timeouts, incorrect handling of unknown slash commands, and improved stability in plan mode fallback scenarios. Users will also benefit from enhanced editing capabilities via improved ranged reads and fuzzy editing.
v0.30.0-nightly.20260224.544df749a8 features14 fixesThis release introduces significant enhancements to policy handling, including support for Gemini 3.1 policy chains and MCP Server Wildcards. Users will benefit from an experimental direct web fetch capability and improved stability, as several critical bugs related to model persistence, non-interactive mode failures, and Windows compatibility have been resolved. Additionally, retry configuration is now user-adjustable via a new setting.
v0.30.0-nightly.20260223.c537fd5ae14 features22 fixesThis release introduces several new features, including experimental in-progress steering hints and support for macOS run-event notifications. Numerous bug fixes address issues with UI rendering, tool handling, and configuration stability. Key improvements include centralizing compatibility checks and migrating core rendering to use xterm.js.
v0.30.0-nightly.20260218.ce84b3cb521 features15 fixesThis release introduces significant new capabilities, including initial SDK support with custom skills and dynamic system instructions, alongside new command structure management via the model subcommand. Numerous bug fixes address UI flickering, platform-specific issues like Windows escaping, and improved tool validation, resulting in a more stable and feature-rich experience.
v0.29.018 features17 fixesThis release introduces significant enhancements to planning capabilities, including the new /plan slash command, support for editing plans with the `replace` tool, and better integration of planning artifacts into workflows. Several critical bugs were fixed, such as resolving agents getting stuck in interactive commands and improving connection stability with retry logic for SSL/TLS errors. Additionally, new features include Vim motion support in the CLI and better management of tool outputs via observation masking and persistence.
v0.30.0-preview.020 features17 fixesThis release introduces significant enhancements to planning capabilities, including a formalized 5-phase workflow and new settings for managing memory and skills within Plan Mode. Users will benefit from new UI controls like text wrapping in markdown tables and the ability to hide shortcut hints. Several stability fixes address issues related to policy persistence, debug console closing, and rendering consistency.
v0.30.0-nightly.20260212.207ac6f2d4 features4 fixesThis release introduces new CLI capabilities, including Ctrl-Z suspension and a prototype clean UI toggle for a more minimal experience. Key fixes address character truncation in markdown mode and improve Plan Mode isolation and policy rule prioritization. Users will also benefit from refined system prompts for better agentic execution.
v0.30.0-nightly.20260212.099aa96215 features6 fixesThis release focuses heavily on enhancing Plan Mode by allowing skill enablement and validating policy overrides. Key updates include migrating internal models to Gemini 3 and deprecating older tool configuration flags in favor of the new policy engine. Several bugs related to selection mode exiting and environment variable population have also been resolved.
v0.30.0-nightly.20260211.f5dd1068f2 features2 fixesThis release introduces multi-line text answers for the ask-user tool and enables tool output masking by default for improved privacy. Performance has been boosted through optimized table rendering and better management of large debug logs. Several critical bugs related to debug console closing and MCP discovery have also been resolved.
v0.30.0-nightly.20260210.8257ec44721 features16 fixesThis release introduces significant enhancements to tool handling, including observation masking, session-linked storage, and remote configuration for masking thresholds. The core model has been updated with an overhauled system prompt for better rigor, and sub-agents now use XML format. Several stability fixes address issues in plan mode, Windows shell execution, and configuration initialization.
v0.28.014 features16 fixesThis release introduces several new capabilities, including the /prompt-suggest command and support for custom themes in extensions. Key fixes address issues with streaming hook warnings, credential exposure, and Vim mode usability. Users will also benefit from improved startup information display and updated undo/redo keybindings.
v0.29.0-preview.020 features16 fixesThis release introduces significant enhancements to planning capabilities, including the new /plan slash command, support for editing plans with the `replace` tool, and improved plan execution metrics. Key fixes address stability issues such as agents getting stuck, memory leaks in MCP transport, and improved handling of SSL/TLS errors. Additionally, the CLI now supports Vim motions and operators for better text interaction.
v0.29.0-nightly.20260206.4ffc349c18 features6 fixesThis release introduces significant enhancements to planning capabilities, including support for the `replace` tool and MCP server integration in plan mode. Key fixes address transient SSL/TLS errors and resolve an infinite loop when using external editors. Users will also benefit from new Vim motion support and improved input prompt autocomplete.
v0.29.0-nightly.20260205.a3af4a8ca21 features12 fixesThis 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.
v0.27.022 features13 fixesThis release introduces significant new capabilities, including the event-driven tool execution scheduler and the new code-reviewer skill. Several core improvements focus on stability, such as fixing issues with authentication failures and improving GitHub workstream rollup efficiency. Users will also benefit from the new /rewind command and enhanced planning features like configurable approval modes.
v0.28.0-preview.015 features16 fixesThis release introduces several new features, including the /prompt-suggest command, support for Positron IDE, and improved OAuth consent flows. Key fixes address issues with MIME type verification, session restoration, and Vim mode usability. Users will also benefit from UX refinements like updated keybindings and clearer settings visibility.
v0.27.0-nightly.20260127.00f60ef538 features5 fixesThis release introduces several new capabilities, including persistent plan file storage and support for multi-scheduler tool aggregation. Key fixes address issues with URI-encoded paths and prevent OOM errors when streaming grep output. Users will also benefit from new CLI options like quick clear shortcuts in vim mode and the ability to disable UI spinners globally.
v0.27.0-nightly.20260126.cb772a5b721 features21 fixesThis release introduces significant new capabilities, including the /rewind command, persistent planning settings, and improved agent discovery via the new `/agents config` command. Several critical bugs were addressed, such as fixing crashes on authentication failure and resolving connection issues within Docker/SSH environments. Users will also benefit from better handling of large text pastes and new image pasting support on Linux.
v0.27.0-nightly.20260122.61040d0eb13 features14 fixesThis release introduces significant new capabilities, including an event-driven tool execution scheduler and the new code-reviewer skill. Users will benefit from improved stability with fixes for keybinding conflicts and better handling of validation errors via a new interactive dialog. Several core systems, like hooks and the scheduler, are now enabled by default for a more streamlined experience.
v0.26.0-nightly.20260121.211d2c5fd10 features12 fixesThis release introduces several new capabilities, including an experiment gate for the event-driven scheduler and enhanced context compression for agents. Several critical bugs were resolved, such as fixing UI rendering order, resolving Windows crashes, and improving error reporting for agent delegation failures.
v0.25.015 features8 fixesThis release introduces significant enhancements to agent interaction and tooling, including filepath autosuggestion, the new /agents slash command, and native Sublime Text support. Key improvements focus on stabilizing Skills management, refining the Rewind functionality with file system reversion utilities, and enabling better isolation for testing via GEMINI_CLI_HOME. Several bugs related to CI stability and Yolo mode behavior have also been resolved.
v0.26.0-preview.013 features17 fixesThis release introduces significant new capabilities, including the 'skill-creator' skill, experimental 'plan' features, and enhanced agent steering to use expert subagents. Numerous bug fixes address issues related to token estimation, event handling, and automation stability. Users will also benefit from improved keybindings and the reintroduction of the --prompt flag.
v0.26.0-nightly.20260119.20580d75410 features12 fixesThis release introduces significant new capabilities, including an experimental 'plan' approval mode and the addition of the 'generalist' agent, while enabling Agent Skills by default. Numerous stability fixes address OOM crashes, PTY descriptor leaks, and automation loop issues. Performance has also been improved with optimizations for handling large text inputs.
v0.26.0-nightly.20260115.6cb3ae4e014 features14 fixesThis release introduces significant new capabilities, including the ability to install and uninstall skills directly via CLI commands and support for defining agents in settings.json. Key fixes address issues with command registration warnings, title setting inconsistencies, and improved shell command safety. Users will also benefit from new UI features like dynamic terminal tab titles and enhanced retry mechanisms.
v0.24.025 features11 fixesThis release introduces significant enhancements to Agent Skills, including infrastructure setup, autonomous activation tools, and system prompt awareness. Key new features also include experimental in-CLI extension management and support for remote agents. Several stability improvements were made, such as fixing crashes related to I/O errors and enhancing policy engine granularity for shell command execution.
v0.25.0-preview.015 features8 fixesThis release introduces significant new capabilities, including native Sublime Text support, filepath autosuggestion in the CLI, and the ability to export chat history directly into GitHub issues via the /bug command. Key improvements focus on skill management centralization, better model routing to subagents, and enhanced stability for CI workflows and terminal compatibility.
v0.25.0-nightly.20260112.15891721a14 features11 fixesThis release focuses heavily on enhancing Agent Skills management, introducing new CLI commands for skills, and centralizing their representation. Key functional additions include support for remote agents, granular control over agent hooks, and improved security policies. Numerous bug fixes address stability issues across platforms, particularly on Windows, related to pasting and file system operations.
v0.23.015 features10 fixesThis release introduces significant new capabilities, including support for the Gemini 3 Flash model and the introduction of remote agent infrastructure. Key fixes address issues with command suggestion display and prevent infinite loops during prompt completion and relaunch attempts. Users will also benefit from new features like Windows clipboard image support and improved hook management.
v0.24.0-preview.025 features11 fixesThis release introduces significant enhancements to Agent Skills, including infrastructure setup, autonomous activation tools, and system prompt awareness. New features also include experimental in-CLI extension management and support for remote agents. Several stability improvements were made, such as fixing crashes related to I/O errors and enhancing policy engine granularity for shell command allowlisting.
v0.24.0-nightly.20260103.30f5c4af44 features4 fixesThis release significantly enhances Agent Skills by introducing autonomous activation, context injection, and system prompt awareness. Users will also benefit from new folder suggestions when adding directories via the /dir add command. Several core stability issues, including promise rejections and shell command handling, have also been resolved.
v0.21.0-nightly.20251226.546baf99310 features4 fixesThis release introduces significant new capabilities, including automatic model persistence across sessions and experimental in-CLI extension management. Key fixes address issues with .geminiignore support in search tools and prevent crashes related to I/O errors. Users will also benefit from improved security visibility around hooks and better error handling for schema validation.
v0.22.011 features11 fixesThis release introduces several new capabilities, including persistent "Always Allow" policies, support for multi-file image drag and drop, and the new `/init` command for the A2A server. Several critical bugs were fixed, such as correcting MCP tool error detection and improving authentication logic by prioritizing the GEMINI_API_KEY environment variable. Additionally, the codebase investigator now aligns its model usage with the main agent.
v0.23.0-preview.015 features10 fixesThis release introduces significant new capabilities, including the launch of Gemini 3 Flash and the introduction of remote agent infrastructure. Key usability updates include native Windows clipboard image support and a new `/auth logout` command for managing credentials. Several fixes address UI glitches, command completion behavior, and improve core stability, especially around hooks and tool execution.
v0.21.0-nightly.20251220.41a1a3eed3 features2 fixesThis release focuses on security and core functionality by implementing JIT context memory loading and sanitizing hook command expansion to prevent injection vulnerabilities. Users will also benefit from improved UX through simplified tool confirmations and reduced log verbosity.