Zed v1.6 Ships Shareable AI Skills and Smarter Git Workflows

Zed v1.6 is out, and this release is less about headline features and more about tightening the screws on everything that matters day-to-day: Git workflows, AI agent ergonomics, and the small editor interactions that add up to a lot of friction over time. The biggest additions land squarely in the agent panel, where the team has been steadily building out a skills system to give your AI coding assistant reusable, shareable instructions.

Skills you can actually share

Skills in Zed are reusable instruction sets for the agent -- think of them as named prompts that the agent can discover and invoke automatically when a task matches their description. A skill is a folder containing a SKILL.md file with metadata and instructions, and the agent sees a catalog of all installed skills and can load one on demand, or you can invoke any skill directly from the message editor with a slash command.

The new feature in v1.6 is the ability to share those skills with teammates via a link. You can hand a skill to a teammate without hosting it anywhere: in the Skills settings page, click the link icon on a skill row to copy a zed://skill?data=… link to your clipboard. The link is self-contained -- it embeds the full SKILL.md contents (base64url-encoded), so the recipient doesn't need access to your project or any registry.

The recipient can review the name, description, and full body, choose a scope (global or project-local), and click Save to install it. Nothing is written to disk until they explicitly save, so a shared link can never silently install instructions into someone's agent. That last part is a meaningful security property -- no surprise prompt injection from a link someone pastes into Slack.

Fast Mode: pay more, wait less

Zed added "Fast mode" support in the agent panel for Anthropic and OpenAI models that support a toggle to get faster responses -- fast mode for Anthropic and priority service tier for OpenAI -- at increased per-token cost. It's a simple tradeoff: if you're in a flow state and want the model to respond immediately rather than wait in a queue, you can flip the toggle and pay the premium. The option surfaces directly in the agent panel UI, no config file edits needed.

On the model side, v1.6 also adds support for specifying reasoning effort for Grok 4.3 from xAI, and introduces an agent.commit_message_instructions setting to customize AI-generated git commit messages. That last one is particularly useful for teams enforcing Conventional Commits or other commit message standards -- you can now bake the format directly into your agent config:

// settings.json
{
  "agent": {
    "commit_message_instructions": "Use Conventional Commits format: <type>(<scope>): <description>."
  }
}

Git panel gets a real upgrade

The Git improvements in this release are the kind that sound small but change how you actually work. This week's release includes the ability to open a Git diff for a single file in its own dedicated tab from the Git panel, split diff mode in the commit history view, total added/removed line counts in the Git panel, and the ability to open files from the file finder without dismissing it.

Here's what that looks like in practice:

• Split diffs in commit history -- the commit history diff view now lets you toggle split diff mode , so you can see old and new side by side instead of a unified inline view.
• Single-file diff tab -- clicking a changed file in the Git panel opens the multi-buffer project diff as before, but now cmd/ctrl-click opens a dedicated single-file diff in its own tab.
• Line count totals -- the Git panel now shows total added and removed line counts across all changed files, so you can gauge the scope of a change at a glance without opening each file.
• Compare with branch -- a new git: compare with branch action lets you directly compare the current branch with an arbitrary branch.

File finder stays open

A small but genuinely useful quality-of-life change: you can now open files from the file finder without dismissing it, using the right arrow key at the end of the input or the "Keep Open" footer button. This means you can open multiple files in succession without re-invoking the finder each time -- something power users have wanted for a while.

Agent sandbox gets smarter

The agent's terminal sandboxing -- which controls what the AI can write to disk when running commands -- got a meaningful upgrade. The agent's terminal sandbox now lets commands request write access to specific paths instead of all-or-nothing filesystem access, with the option to grant access for a single command or the rest of the conversation. Previously, you had to choose between giving the agent full write access or blocking it entirely. Now you can approve writes to /tmp/build-output without opening up your entire home directory.

What's available and what it costs

Zed v1.6.3 is available now on macOS, Windows, and Linux. The core editor is free forever. The Personal plan is free forever with 2,000 edit predictions per month; Pro is $10/month with unlimited predictions and hosted models. Business is $30 per seat per month and adds org-wide model policies, data governance, role-based access controls, and unified spend visibility. A two-week free trial of the Pro plan is available, which comes with $20 of token credits and unlimited accepted edit predictions.

The skills sharing feature is available to all users regardless of plan -- you don't need a Pro subscription to share or receive skills. If you're already running Zed, the update is available through the built-in updater or from the releases page.