Grok Build Updates & Release Notes

Subagents can always use the parent model, with that choice locked in when the session starts

Earlier image attachments survive multiple compactions

Subagent spawning no longer requires choosing a type

The agent keeps helper scripts, logs, and PR drafts in the system temp directory instead of the repository

Model effort menus pick up live effort levels, and config aliases inherit them

MCP tools that receive the wrong JSON shape are automatically fixed when the mismatch is unambiguous

Local worktree capture no longer refuses repositories that contain uninitialized submodules

Memory captures no longer include opaque model reasoning blobs

Idle timeouts show a clear message instead of raw internal text

Images attached to the last user prompt survive compaction

Custom agent profiles persist correctly across session resume and reload

Subagent labels appear as "Subagent" instead of "General" in transcripts and the tasks pane

Dashboard preview setting lets you hide the selected-session preview panel

Policy setting allows organizations to disable hooks that are not from managed policy

Background shell commands appear as live task rows with streaming output

grok --sandbox no longer exits when accepting folder trust

Pinned agents in the dashboard no longer jump when their activity changes

Swift code with triple-quoted strings highlights correctly in the pager

Mixed-case reasoning_effort values (for example xHigh) are accepted in config.toml

Pressing Enter right after sending from the dashboard no longer opens an empty agent session

Sessions respect an [agent] set in config.toml instead of always using a builtin plan

MCP plugin authentication succeeds for servers that declare a client ID in their manifest

Shell command approvals from background subagents are shown instead of being silently rejected

Headless MCP status reporting and connecting reminders match actual server handshake state

Swift string interpolation with nested parentheses highlights correctly

Copy-paste in --minimal mode no longer inserts extra blank lines or breaks long paths at wrap points

/memory modal shows copy confirmations, searches note contents, and works on narrow terminals

Memory delete is much faster, and toggling memory on from a config-disabled session takes effect immediately

Syntax highlighting uses far less memory and runs faster on large TypeScript and other files

MCP tool results include structured JSON data when the server provides it

Composer footer shows the Alt+Enter chord for inserting a newline when Cmd+Enter is unavailable over SSH

Pasting an image on macOS no longer attaches the wrong image

MCP tool calls listed directly show their name, arguments, and any error message

Background subagent tasks show cancelled status when the parent session closes

The Subagents dock no longer shows an unclickable "Done N completed" row after all subagents finish

Cancelled or timed-out MCP tool calls stop instead of continuing in the background

/rewind no longer fails when older compaction checkpoints are missing after a session sweep

Large skill files are capped at the same 25k-token limit as other reads; oversized skill bodies show a truncation note

Long-running sessions no longer lose older compacted history during the 30-day cleanup

Auto "where was I" recaps no longer appear while a scheduled task, monitor, or workflow is still running

In minimal mode, confirming /delete or /exit immediately opens a new empty session

In minimal mode, resuming a session after /new reloads its history instead of showing the wrong session

MCP tools that take no arguments no longer show a stray {} line in the permission sheet

grok clone on Windows no longer fails when the repository git config contains backslashes or quotes

Folded subagent groups in scrollback correctly label how many are still running versus completed

Dashboard search and the resume picker indicate the active text field and restore list selection after search

Dock sections no longer keep a selection highlight after you collapse them with a click

Worktree paths no longer show an extra suffix in the header or dashboard

MCP tools with long server prefixes are kept instead of being silently dropped

Turn cancelled banners name the cause instead of always saying "by user"

Long paths in headers are shortened to the last two components

Successful hook runs are silent; only blocking or failing hooks show status

Vim-mode Shift+J/K jump to the next or previous turn at the top of the viewport

Double-click selects a radio button in settings choosers

Dashboard actions row (+ New Agent, Open Previous, Worktree) is navigable with arrow keys and Enter

/workflow can pause and stop background runs the agent launched

/theme picker matches aliases such as transparent, dark, or system and still inserts the canonical name

Bash command output in the pager is the complete result instead of a truncated tail

Voice dictation inserts text at the cursor instead of always appending

Headless prompts time out cleanly instead of hanging when the agent does not respond

Scheduled tasks remind the agent to check results and clean up or update recurring schedules

grok -c no longer attaches the wrong empty session after starting without --continue then resuming

Dock and queue hover colors match the rest of the terminal

Image uploads no longer fail on Adobe SVGs that contain a PNG thumbnail prefix

Dock subagent rows show the same [stop] label as other killable rows

Startup settings and model catalog are no longer shared or clobbered across concurrent boots or accounts

Dock Tasks panel spinner aligns with the header

/feedback sends immediately again instead of starting a model turn; failures save to Drafts

Built-in agent tools take precedence over user MCP servers when tool names collide

edit_file diffs show real file line numbers instead of starting at line 1

Permission prompts for file edits auto-expand the diff so you can review the change

Sending a message to a finished subagent continues that same subagent instead of failing

Background subagent completion messages show the actual output instead of a pointer to run a tool

Dashboard has a unified header with a location picker and an actions row for new or previous sessions

MCP connectors show when they need re-authentication even if no tools are listed

Extensions modal shortcuts work when the tab bar is focused, with helpful messages for invalid header actions

Send now works while an automatic background turn is running

Resumed sessions show the exact text you typed for mid-turn follow-ups

/skills picks up newly installed skills even when they were written from another machine

Auto mode no longer instantly runs destructive git checkout -- commands

Restoring pasted content after skill injections or rewinds no longer crashes

Subagents and workflows respect the parent session bash timeout and auto-background settings

Slash command advertisements no longer spam when the current directory is your home folder

Foreground shell commands no longer produce spurious background-task reminders or incorrect kill results

Queued follow-up buttons appear in the order [Send now][edit][cancel]

/goal (and resume) after Esc starts the goal planner instead of failing with "Planning failed"

Foreground commands report they were backgrounded after about 15 seconds instead of the 120s timeout value

Skill announcements no longer repeat the same list of available skills in a session

Monitor events wake idle sessions in multi-session processes instead of waiting for the next prompt

Opening or resuming sessions no longer pauses the interface while reading MCP config files