Supported Agents

GuardClaw works with 7 major coding agents out of the box.

Agent	Integration	Pre-tool blocking	Approval flow
Claude Code	HTTP hooks	✅	✅
Codex CLI	Command hooks	✅	✅
Gemini CLI	HTTP hooks	✅	✅
OpenCode	HTTP hooks	✅	✅
OpenClaw	WebSocket plugin	✅	✅
Cursor	Shell hooks	⚠️ shell only	✅
GitHub Copilot CLI	HTTP hooks	✅	✅

Cursor note: File operations (read/write/edit) are not intercepted — only shell commands.

How Hooks Work

HTTP hooks (Claude Code, Gemini CLI, OpenCode, GitHub Copilot CLI)

GuardClaw registers as an HTTP hook handler. Before each tool call, the agent POSTs to localhost:3002/api/cc/pre-tool. GuardClaw scores the call and returns allow or block.

Install:

node scripts/install-claude-code.js

Uninstall:

node scripts/install-claude-code.js --uninstall

OpenClaw WebSocket plugin

The guardclaw-interceptor plugin hooks before_tool_call directly in the OpenClaw gateway. This enables pre-execution blocking of all tool types.

Install:

guardclaw plugin install
openclaw gateway restart

Command hooks (Codex CLI, Cursor)

GuardClaw wraps shell command execution. Shell commands are intercepted; file read/write operations are not.

Installing Hooks

The setup wizard (guardclaw setup) auto-detects installed agents and installs hooks with one confirm. You can also install manually per agent using the commands above.

After installing hooks, restart the target agent to activate them.