Zed Editor v1.14.2 リリースZed Editor Releases v1.14.2

Zed, the Rust-based code editor from the team behind Atom, has published version 1.14.2, a weekly update that focuses heavily on making its built-in AI Agent safer to run and giving developers finer control over both file management and the editing environment. The headline change is the addition of sandboxing for the Agent's terminal commands and web fetches, a move that reflects growing industry attention to the security risks of letting AI assistants act autonomously inside a developer's workspace.

Sandboxing, in this context, means the Agent's outbound actions are confined to a restricted execution environment rather than running with the same unrestricted access as the user. When an AI Agent can open a terminal, execute shell commands, and pull content from the web, it gains the ability to touch files, network resources, and system state well beyond the code a developer intended it to edit. By isolating those terminal commands and web fetches, Zed appears to be limiting the blast radius of an unexpected or malicious instruction, whether that comes from a prompt-injection attack embedded in fetched content or simply from a model generating a command with unintended consequences. Zed points to a companion blog post for the details, and users evaluating the feature are likely to want to review exactly which operations are constrained and how the boundaries are enforced.

Alongside the security work, the release adds support for undoing and redoing file operations in the Project Panel, backed by documentation. The Project Panel is Zed's file-tree sidebar, and until now actions such as creating, renaming, moving, or deleting files there did not participate in a familiar undo history the way in-editor text edits do. Bringing undo and redo to these operations reduces the risk of an accidental rename or deletion becoming a permanent mistake, and it aligns the panel's behavior more closely with the expectations set by mainstream editors and file managers. This is a quality-of-life improvement, but it also complements the Agent changes: as AI tools increasingly perform file operations on a developer's behalf, having a reliable way to reverse those actions becomes more important.

The update also introduces a Skip Hooks option for commits. Git hooks are scripts that run automatically at certain points in the version-control workflow, and pre-commit hooks in particular are commonly used to enforce linting, formatting, or test checks before a commit is recorded. Those checks are valuable, but they can be slow or occasionally get in the way of a work-in-progress commit. A Skip Hooks toggle gives developers a deliberate way to bypass them for a given commit, mirroring the long-standing command-line behavior of git commit with the no-verify flag while surfacing it directly in Zed's interface.

Rounding out the release, Zed now offers configurable Agent Panel fonts, letting users set the typeface used in the AI Agent's conversation view independently of the main editor. Because the Agent Panel mixes prose, code snippets, and command output, giving it its own font settings is a small but practical refinement for readability and personal preference.

Taken together, these changes fit a broader pattern across the developer-tools market, where editors are racing to integrate agentic AI features while grappling with the trust and safety questions those features raise. Zed competes in a space that now includes AI-first editors such as Cursor and Windsurf, as well as assistants like GitHub Copilot that plug into existing environments. Sandboxing autonomous agents is emerging as a differentiator in that landscape, since the more capable an assistant becomes at executing real actions, the more scrutiny its permissions attract. Zed's emphasis on GPU-accelerated performance and native collaboration has been part of its pitch, and layering in guardrails for its Agent suggests the project is trying to balance capability with control.

For existing users, v1.14.2 should arrive through Zed's standard update mechanism, and the marking of these items as "Shipped" indicates they are live in this build rather than experimental previews. As with any release that touches security boundaries, teams relying on the Agent for real work would be well advised to read the linked blog post and documentation before assuming the new sandbox fully covers their particular workflows, since the precise scope of what is and is not restricted will shape how much the feature can be depended upon.