I’ve been working on cerone-openclaw-plugin, a native OpenClaw plugin that checks tool calls with Cerone before execution. (Cerone is an AI agent runtime.)

OpenClaw agents can read files, write files, call APIs, and trigger other tools.

This plugin adds a pre-execution check layer using OpenClaw’s existing before_tool_call hook, without modifying OpenClaw core.

What it does:

• loads as a normal third-party OpenClaw plugin
• intercepts tool calls through before_tool_call
• sends the tool name, parameters, and runtime context to Cerone
• gets back a decision: approved flagged rejected
• maps that into OpenClaw behavior: approved -> tool runs flagged -> require approval by default rejected -> block the tool call

Example shape:

{
  "plugins": {
    "entries": {
      "cerone-openclaw-plugin": {
        "enabled": true,
        "config": {
          "trialMode": "auto",
          "flaggedBehavior": "requireApproval",
          "networkFailureBehavior": "allow",
          "agentPurpose": "Read repository files and inspect code inside OpenClaw for software engineering tasks.",
          "agentCapabilities": ["file_read", "file_write", "network_access", "api_call"]
        }
      }
    }
  }
}

In practice, this means OpenClaw keeps its normal workflow, but the sdk gets a chance to allow, flag, or block a tool call before it executes.

Repo:

https://github.com/AnantDhavale/cerone-openclaw-plugin

npm:

https://www.npmjs.com/package/cerone-openclaw-plugin

Would appreciate feedback from people using OpenClaw for:

1. coding agents
2. tool-heavy assistants
3. agents that can touch files or external APIs
4. workflows where some actions should pause for approval before execution

Thanks!