Auto Reviewer

OpenCode plugin that acts as a supervisor to detect conceptual loops

Installation

npm install opencode-auto-reviewer

Features

• Acts as an LLM-powered supervisor observing the agent trajectory.
• Detects conceptual rabbit holes where the agent is stuck but technically not in an exact loop.
• Provides hints, warnings, and critical interventions to nudge the agent back on track.
• Forces strategy changes when approaches fail to converge.

Usage

This is a plugin for OpenCode.

You can easily enable this plugin globally by adding it to your opencode.json (or opencode.jsonc) configuration file under the plugins array.

{
  "plugins": [
    "opencode-auto-reviewer"
  ]
}

Once added to your configuration, OpenCode will automatically load and activate the plugin whenever it starts.

Alternatively, if you are configuring OpenCode programmatically via the SDK, you can register it like this:

import { configureAgent } from '@opencode-ai/sdk';
import plugin from 'opencode-auto-reviewer';

const agent = configureAgent({
  plugins: [
    plugin({
      // Provide configuration options here
    })
  ]
});

License

MIT