Guardrails

Guardrails are workspace rules that intercept Claude Code tool calls before they run. Each enabled guardrail is enforced in every member's session — when a tool call matches, the guardrail can block it, prompt the member for approval, or note a warning and let it proceed.

How Guardrails Work

When Claude Code is about to run a tool (such as editing a file or running a shell command), each enabled guardrail checks whether the call matches its match configuration. If it does, the guardrail's action decides what happens next, and the guardrail's message is surfaced to the member.

• Guardrails are evaluated before the tool call executes.
• They run in every member's session, with no per-member setup.
• A guardrail only takes effect while it is enabled.

The Three Actions

Every guardrail has exactly one action that determines what happens when a tool call matches:

Match Configuration

A guardrail's match defines which tool calls it applies to. Every guardrail must target at least one tool; the optional pattern matchers narrow it further.

Tools

Choose which Claude Code tools the guardrail applies to. Select one or more specific tools, or All tools (*) to match every tool call.

• Edit, Write, MultiEdit — file changes
• Bash — shell commands
• WebFetch — fetching remote content
• All tools (*) — match every tool call

Path globs

Optional glob patterns matched against the file path of an edit (for Edit, Write, and MultiEdit). Use them to scope a guardrail to sensitive areas of the codebase.

**/auth/**
**/payments/**
**/migrations/**

Command patterns

Optional regular expressions matched against the command of a Bash tool call. Use them to catch destructive or risky commands.

\brm\s+-rf\b
git\s+push\s+.*--force

Content patterns

Optional regular expressions matched against the content being written (for Write and Edit). Use them to catch secrets, keys, or other content that shouldn't land in the codebase.

AKIA[0-9A-Z]{16}
(sk|pk)_(live|test)_[A-Za-z0-9]{8,}

Presets

When you add a guardrail, you can start from a curated preset that prefills the form with a known-good rule, or configure a custom guardrail from scratch. A preset is just a starting point — you can adjust any field before saving.

Creating a Guardrail

1. Open Guardrails in your workspace
2. Click Add guardrail
3. Choose a preset, or pick Custom guardrail
4. Fill in the details:
   • Name — a short, kebab-case identifier (lowercase letters, numbers, and hyphens, e.g. protect-secrets)
   • Action — Deny, Ask, or Warn
   • Description — what the guardrail does (shown in the list)
   • Tools — which tool calls it applies to (select at least one)
   • Path globs / Command patterns / Content patterns — optional matchers for the selected tools
   • Message — the text surfaced to the member when the rule fires
5. Toggle Enabled and Experimental as needed
6. Click Create guardrail

Enabled & Experimental

Each guardrail has two toggles that control when it is enforced:

• Enabled — when on, the guardrail is eligible for enforcement in every member's session. Disabled guardrails stay in the list but have no effect.
• Experimental — when on, the guardrail's live draft is enforced only for members running in Experimental mode. Use this to trial a draft rule before promoting it for everyone.

Enforcement also follows the guardrail's released version: a normal session enforces the stable release, while an Experimental session enforces the live draft (when the rule is flagged experimental). A guardrail with no stable release yet is not enforced. See Experimental Mode for how the per-member toggle works.

Favorites

Mark frequently referenced guardrails as favorites with the star icon. Favorites are personal to your account — other members won't see yours, and you won't see theirs.

Managing Guardrails

Guardrails are managed by workspace owners, admins, and maintainers. Members can view the list but cannot create, edit, enable/disable, or delete guardrails.

• Search and filter — find guardrails by name, or filter by enabled/disabled status.
• Enable / Disable — toggle enforcement without deleting the rule.
• Mark / Unmark experimental — switch a rule between everyone and Experimental-mode-only.
• Edit — change any field. Edits live in the draft until you publish and promote a release (see below).
• Versions — open the version history to publish, promote, or restore releases.
• Export as JSON / Share — download a guardrail, or create an unlisted link to share it.
• Delete — removes the rule. It stops being enforced in your team's sessions. This cannot be undone.

Versions & Releases

Guardrails are versioned just like skills, agents, commands, and workflows. Editing a guardrail changes its draft; publishing freezes that draft as an immutable release.

• Publish — freeze the current draft as a new version. Your first publish automatically becomes the stable release, so the guardrail starts enforcing.
• Promote — make any version the stable release. This is also how you roll back to an earlier version.
• Restore — copy an older version's content back into the draft so you can keep editing from it. History and the stable release are untouched until you publish again.

Because a normal session enforces the stable release, your edits don't affect the team until you publish and promote them. Flag a guardrail Experimental and test the draft in an Experimental session first, then promote it with confidence.

Sharing, Export & Import

Share a guardrail with another workspace, or move one between your own workspaces:

• Share links — create an unlisted, revocable link (from the guardrail's menu) once it has a stable release. Anyone with the link can preview the rule and import their own copy. Links can be pinned to a version or follow your latest stable release, with optional expiry and import limits.
• Export as JSON — download a guardrail as a portable JSON file (its name, description, match, action, and message).
• Import — upload a guardrail JSON, or open a share link. The imported guardrail is added as a draft (auto-renamed if the name is taken); publish and promote it to start enforcing.

Limits

The number of guardrails per workspace depends on your billing tier:

Each guardrail also allows up to 20 patterns per matcher list (path globs, command patterns, content patterns). See Workspace Item Limits for all workspace limits.