Zed v1.4 Kills Its Rules System and Ships Skills for AI Agents
Zed v1.4 replaces its old Rules system with Agent Skills — reusable, on-demand instruction packages that give the AI agent specialized knowledge per project or globally.
- Zed v1.4 ships Skills, replacing the old Rules library entirely —
@ruleis gone,@skillis in. - Skills are reusable instruction packages stored as
SKILL.mdfiles that give the agent specialized knowledge on demand. - Progressive disclosure keeps context lean — the agent only loads a skill's full body when it activates it, not on every message.
- Global or project-local scope: skills in
~/.agents/skills/work everywhere; project skills require explicit trust before loading. - Git improvements include a base branch selector in diff views and a toggle-all-diff-hunks action.
- Available now on the preview channel; free to download.
Zed v1.4 ships with a new agent feature that replaces the editor's previous Rules library entirely: Skills. The old @rule system is gone. In its place is a more structured, on-demand approach to giving the AI agent specialized knowledge , one that scales from a single personal workflow to a shared team standard.
Rules are dead, long live Skills
The rules library has been removed and replaced with skills support, and the @rule autocomplete entry in the agent chat input has been replaced with @skill.
This is a breaking change , if you have existing rules configured, you'll need to migrate them.
Skills are reusable instruction packages that give the agent specialized knowledge for specific tasks: test-driven development workflows, document processing, database integrations, or your team's internal coding standards. Think of them as prompt templates with structure , not just a wall of text dumped into every conversation, but something the agent can selectively load when relevant.
How Skills actually work
A skill is a folder containing a SKILL.md file with metadata and instructions. 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.
This lazy-loading design is intentional.
Skills use progressive disclosure: the agent sees only the name and description until it activates a skill, then loads the full body. This keeps the model's context window lean , it doesn't pay the token cost for a skill it doesn't need on a given task.
Here's what a minimal skill directory looks like:
~/.agents/skills/
my-tdd-workflow/
SKILL.md # metadata + instructions
references/ # optional detailed docs
scripts/ # optional runnable scripts
Structure your skill to take advantage of progressive disclosure: put the most important instructions near the top of the body, keep SKILL.md under 500 lines, and move detailed references to a references/ subdirectory.
Global vs. project-local
Skills can live in two places, and the distinction matters for how you use them:
- Global skills , stored at
~/.agents/skills/, accessible to the agent even though they're outside your project. Good for personal workflows you want everywhere. - Project-local skills , stored at
<worktree>/.agents/skills/. Project-local skills only load from trusted worktrees. Skills from a freshly cloned or untrusted project are excluded from the catalog and slash commands until you grant trust. This is a meaningful security guardrail for open-source contributors.
Creating a skill
You create one with the new agent: open skill creator action, then add it globally or to a specific project. You can also choose whether the agent discovers each skill automatically, or requires explicit invocation via / in the agent panel.
Zed also includes a built-in create-skill skill , invoke it with /create-skill and the agent walks you through the process.
The creator window lets you fill in the skill's name, description, scope, and body without touching the file system directly.
Skills support live reload: adding, removing, or editing a SKILL.md takes effect immediately without restarting your session.
That makes iteration fast during authoring.
What else shipped in v1.4
Skills are the headline, but the release includes meaningful Git improvements too:
- Added the ability to choose the base branch in the branch diff view.
-
Added
editor: toggle all diff hunksto expand or collapse all diff hunks at once. -
Added support for a global
AGENTS.mdfile alongsidesettings.jsonfor user-wide instructions that are included in every project's system prompt. - Added support for image output from MCP tools.
There is a 50KB catalog budget: the total size of all skill names and descriptions is capped at 50KB. Skills that don't fit are dropped from the catalog with a warning in the UI. Keep descriptions tight.
What skills are good for , and where they fall short
Skills shine for anything repetitive and context-heavy that you'd otherwise re-explain in every thread:
- Enforcing team code style or architecture patterns on a specific repo
- Encoding a multi-step workflow (e.g., "always write a failing test first, then implement")
- Giving the agent knowledge of internal tooling or APIs it won't find in public docs
- Storing database schema conventions or deployment runbook steps
The main limitation to be aware of:
there is no remote registry. Zed does not fetch skills from URLs or support custom search paths. Skills come from ~/.agents/skills/ and <worktree>/.agents/skills/ only.
Sharing skills across a team means committing them to the repo or distributing them manually , though a subsequent release added the ability to import skills from GitHub Markdown URLs in the Skill Creator.
Zed v1.4 is available now on the preview channel, with the stable release following shortly after. Zed is free to download.