Grok · 4 min read

Search session contents from the /resume picker

The /resume picker now searches what was said inside your sessions as you type and opens a match directly. The week also brings a minimal sandbox profile, a workspace_write option and new keyboard shortcuts for long messages and copying replies.

Key takeaways 5
  • The /resume picker searches session contents as you type, and you can open a match directly (1.0.85).
  • In minimal mode, /resume asks for confirmation with 'y confirm delete · n cancel' before it deletes a session (1.0.87).
  • A new `minimal` sandbox profile and a `workspace_write` option let custom profiles control writes to the launch folder (1.0.82).
  • Enter or Ctrl+F expands and collapses long user messages in the transcript, and Ctrl+O copies the latest assistant reply (1.0.82).
  • Smaller changes cover `/connectors`, argument hints in slash command autocomplete, terminal notifications for question and plan-approval cards, and live updates in the model picker.

Sessions pile up faster than anyone names them, and a title rarely says which one contained the fix you need now. This week Grok Build lets you search what was actually said. It also adds a leaner sandbox profile and a few keyboard shortcuts for long transcripts.

Week 46 (November 9-14, 2026) covers versions 1.0.80 to 1.0.87: 8 releases and 180 changelog entries. Below are the changes that affect daily work, plus a batch of smaller wins in the TUI, skills and permissions.


Find a session by what was said in it

Content search in /resume. The /resume picker now searches the contents of your sessions as you type, and you can open a match directly (1.0.85). Before, you had to recognize a session from its list entry. Now a phrase you remember from the conversation is enough. In practice this matters most when you run many sessions and the list stops being scannable.

Safer deletion in minimal mode. In minimal mode, /resume now shows ‘y confirm delete · n cancel’ before it deletes a session (1.0.87). It is a fix, and it removes the chance of deleting a session with one wrong keypress while you browse.

What to check. The input gives no plan restrictions or extra prerequisites for the search, so there is nothing to enable. The sessions documentation is the place for the full behavior.


Changes for everyday work

A minimal sandbox profile. Version 1.0.82 adds a new minimal sandbox profile and a workspace_write option. The option lets custom profiles control whether the launch folder can be written to. If you maintain custom profiles, you now have a direct switch for that one question. The sandbox documentation describes the profiles.

Long messages in the transcript. Enter (or Ctrl+F) expands or collapses long user messages directly in the transcript (1.0.82). Long pasted prompts no longer take over the screen, and you can still read them in place when needed.

Copy the last reply. Ctrl+O now copies the latest assistant reply (1.0.82). The keyboard shortcuts page lists the rest.


Technical notes

Slash commands and skills

  • /mcps has a new alias, /connectors, which opens the extensions modal at the MCP tab.
  • /skills with a name opens the Skills tab filtered to that name (1.0.85).
  • Slash command autocomplete shows each command’s argument hint after its name (1.0.85).
  • Skills expand ${ARGUMENTS} and ${ARGUMENTS[N]} in the skill body (1.0.85).
  • Unknown or disabled slash commands get a direct reply instead of being sent to the model (1.0.85). A mistyped command no longer spends a model turn.

Permissions and notifications

  • Question, MCP input and plan-approval cards send the same terminal notification as permission requests (1.0.85). If you work in another window while a session waits, these prompts now reach you the same way.
  • Typing on the permission card sends your text to the model as feedback instead of changing the selected row (1.0.87).

Transcript and links

  • Drag-selecting in the transcript skips collapsed tool calls and the turn summary line, so you copy only the conversation (1.0.87).
  • Links in replies highlight when you hover over them, and Cmd/Ctrl+click opens them even with the confirmation card visible (1.0.87, 1.0.86).

Models, updates and voice

  • The model picker updates live when your account models change or another client switches the model. Open /model and /context-window dropdowns refresh their rows and markers immediately (1.0.84).
  • grok update delegates to the self-updating build in ~/.grok/bin when needed (1.0.82).
  • /voice on macOS now names the app (your terminal) that is missing Microphone permission and says how to fix it (1.0.87).

What to take away

If you live in the terminal. Try searching for a phrase from an old conversation in /resume. Then try Enter or Ctrl+F on a long message and Ctrl+O for copying the last reply. If you use drag-selection to copy, you will see that collapsed tool calls and the summary line no longer come along.

If you run parallel sessions. Content search in /resume is the change for you, since names alone get ambiguous across many sessions. The extra terminal notifications for question, MCP input and plan-approval cards help when sessions wait in the background. The model picker also reflects a model switch made by another client, so check /model if you mix clients.

If you administer integrations and company deployments. Look at the minimal sandbox profile and decide whether your custom profiles should set workspace_write explicitly. /connectors now opens the MCP tab, so adjust any internal instructions that mention /mcps. Also note that grok update can delegate to the self-updating build in ~/.grok/bin, which is worth knowing if you manage how installs get updated.


Sources: the Grok Build changelog, versions 1.0.80 to 1.0.87 (November 9-14, 2026), and the documentation at docs.x.ai/build, specifically the pages on sessions, sandbox and keyboard shortcuts. Details of individual features can change, so check the documentation for the current state.

Patrick Zandl trains teams in Claude Code and agentic software development in Prague and writes about what actually works in real-world AI-assisted projects. Follow on LinkedIn

Read next

Get the next one by email

Every Tuesday: what changed in Claude Code, Codex and other coding agents, plus new articles. No spam.