astradevlabsastradevlabs
← All posts
Dev Tips4 min

Cheat Sheet: 5 VS Code 1.130 switches for a saner agent workflow

Dev Tips

VS Code 1.130 landed on July 22, 2026, and it is one of the clearer signs that the editor is turning agent workflows into a first-class development surface instead of a side panel experiment. The release is not just about one shiny feature. It changes where agent sessions run, how you review multi-file changes, and how much approval friction you deal with while an agent is working.

This cheat sheet is the practical version: five switches and habits worth trying first if you want the release to save time instead of just adding another settings rabbit hole.

1. Turn on the Agent Host for sessions that should survive the window

The biggest architectural change in 1.130 is the Agent Host. VS Code now runs supported coding agents in a dedicated process, instead of tying the session to one editor window.

Why that matters in practice:

  • The same session can be viewed from multiple VS Code windows.
  • Long-running work is less coupled to a busy extension host.
  • Remote and multi-client workflows get simpler because the host is the source of truth for session state.

If you want to try it, start with:

json
{
  "chat.agentHost.enabled": true,
  "chat.agents.claude.preferAgentHost": true
}

That second setting matters if you use Claude in VS Code and want it to prefer the new host path when available. One caveat from the release notes: rollout is gradual, and some org-managed installs may control this setting centrally.

2. Use assisted approvals when prompts are killing your flow

The second switch is less flashy but more immediately useful: assisted tool approvals.

In 1.130, VS Code adds chat.assistedPermissions.enabled, which lets the model evaluate the risk of a tool call and decide whether it can proceed or should still stop for approval. If you are running longer agent tasks, this can remove a lot of repetitive confirmations.

json
{
  "chat.assistedPermissions.enabled": true
}

The right mental model is not “full auto mode.” It is “fewer low-value interruptions.” Keep using normal trust boundaries for destructive or high-risk tasks. If your team already relies on sandboxing and explicit review, this feature works best as a friction reducer, not as a substitute for judgment.

3. Move project-wide work into the Agents window

1.130 also makes the Agents window more useful as a real work surface instead of a novelty.

You can open it from the title bar, from Chat: Open Agents Window, or from the command line with:

bash
code --agents

The payoff is simple: the Agents window is organized around sessions, not files. That makes it better for higher-level tasks like “update dependencies in three repos,” “review what changed across two chats,” or “keep one autonomous task running while you keep coding in the main editor.”

The preview improvements in 1.130 are mostly about review speed:

  • file-level diff statistics
  • a more compact multi-file diff layout
  • cleaner quick-chat rows in the sessions list

If you mostly use AI inline, stay in the editor. If the task is bigger than the current file, the Agents window now has a much better argument.

4. Default to worktrees when you want parallel experiments

One small but meaningful 1.130 improvement: worktree support now applies across all agent harnesses on the Agent Host, including Claude and Codex sessions in the Agents window.

That means you can keep the lazy, safe workflow:

  1. Start a new agent session.
  2. Check New Worktree.
  3. Let the agent make broad changes without polluting your main workspace.
  4. Review and merge only what survives inspection.

This is the kind of feature that prevents cleanup work later. If your agent task could touch more than a couple of files, isolating it in a worktree is usually cheaper than untangling a noisy working tree after the fact.

5. Turn on the tiny visibility upgrades too

The last bucket is easy to skip because it sounds minor. It is not.

First, enable chat timestamps if you care about how long agent turns actually take:

json
{
  "chat.verbose": true
}

Second, if you spend time in terminal diffs, set Git's mnemonic prefixes so VS Code can open those file links correctly from diff output:

bash
git config --global diff.mnemonicPrefix true

This helps when you are reviewing patches from the terminal and want the jump back into the editor to be immediate instead of manual.

One more practical preview note: the Agents window still has limitations. The docs call out that the agent cannot directly open the integrated browser for you yet, so if your workflow depends on browser validation, start that browser session yourself from the command palette or a localhost link.

Starter setup

If you want one quick trial run after upgrading to 1.130, this is a sensible baseline:

json
{
  "chat.agentHost.enabled": true,
  "chat.agents.claude.preferAgentHost": true,
  "chat.assistedPermissions.enabled": true,
  "chat.verbose": true
}

Then open the Agents window, launch one task in a new worktree, and see whether the approval and review loop actually feels lighter. That is the real test of this release. The headline feature is the Agent Host, but the practical win is that 1.130 makes long-running agent work easier to isolate, inspect, and resume.

References