OpenClaw Mainctrl Adds Runtime Tool Guards

A new ClawHub skill called mainctrl landed this morning with a practical safety goal: let OpenClaw operators toggle destructive tool access for selected agents at runtime.

The skill was published by iFeel and is currently at version 1.0.2. Its description says it manages OpenClaw agent tool-call permissions through a runtime plugin hook that intercepts write, edit, exec, process, and apply_patch operations from controlled agents.

What Mainctrl Does

Mainctrl is built around a companion OpenClaw plugin. The plugin listens on before_tool_call, checks whether the caller is in the configured controlledAgents list, and blocks selected tools when safety is on.

The default blocked tools are the ones operators usually worry about:

• write for file creation and overwrite.
• edit for in-place file changes.
• exec for shell commands.
• process for background process management.
• apply_patch for patching files.

Agents outside the controlled list are not affected. That design lets a main agent be constrained while a delegated sub-agent can still do the implementation work, assuming the operator wants that separation.

Why It Is Different From Static Deny Lists

The interesting part is that mainctrl is runtime-controlled. The skill argues that static tools.deny rules require config edits and gateway restarts, while a plugin can read a local state.json file on each tool call.

That gives operators a fast switch:

• Turn blocking on or off without changing agent config.
• Change which agents are controlled.
• Change which tools are blocked.
• Keep sub-agents or specialist agents outside the restriction set.

The included mainctrl.sh script exposes commands for status, on, off, controlled-agent selection, blocked-tool selection, plugin install, and plugin removal.

The Delegation Model

Mainctrl also includes an opinionated agent behavior rule. When a controlled agent is blocked, the skill tells it to inform the user briefly and immediately delegate the work to a sub-agent such as coder, tester, auditor, or publicist.

That makes the skill more than a raw deny mechanism. It is trying to enforce a workflow where the main agent becomes a coordinator and destructive work moves to a role that the operator has allowed to use those tools.

This is a pattern worth watching. The OpenClaw ecosystem has seen growing interest in runtime policy, audit trails, and human approval boundaries. Mainctrl takes a direct route: keep the policy local, make the switch immediate, and force a delegation habit when an agent hits the guardrail.

Who Should Look At It

Mainctrl is best suited for operators who already run multi-agent OpenClaw setups and want a simple local control plane for tool access. It is especially relevant for teams experimenting with "main agent coordinates, sub-agents execute" workflows.

The skill is not a replacement for full sandboxing, code review, or least-privilege system design. It is a local runtime guard for a specific class of tool calls. But that narrow scope is also its appeal: it gives OpenClaw builders a concrete, inspectable way to change agent permissions while the system is running.