OpenCode Supervisor Plugin for OpenClaw

Autonomous task execution via OpenCode HTTP API with reliable verification through actual file changes.

What's New in v1.1.3

- Reliable error detection: Uses exit codes instead of text pattern matching
- Change verification: Confirms task completion via /session/:id/diff API
- No false positives: Doesn't mistake code containing "Error" for actual errors
- Progress tracking: Detects when model isn't making changes and prompts accordingly

Installation

``bash
openclaw plugins install openclaw-opencode-supervisor
`

Or install from GitHub:

`bash
openclaw plugins install github:zvenoai/openclaw-opencode-supervisor
`

Configuration

After installation, configure the plugin in your OpenClaw config:

`bash
openclaw config set plugins.openclaw-opencode-supervisor.enabled true
openclaw config set plugins.openclaw-opencode-supervisor.config.password "your-opencode-password"
`

Or edit ~/.openclaw/config.json:

`json
{
"plugins": {
"openclaw-opencode-supervisor": {
"enabled": true,
"config": {
"apiUrl": "http://127.0.0.1:4096",
"username": "opencode",
"password": "openclaw2026",
"sandboxDir": "/root/clawd/sandbox",
"credentialsDir": "/root/clawd/credentials",
"maxIterations": 50,
"timeoutMs": 180000
}
}
}
}
`

Restart the gateway after configuration:

`bash
openclaw gateway restart
`

Configuration Options

| Option | Type | Default | Description |
|--------|------|---------|-------------|
| apiUrl | string | http://127.0.0.1:4096 | OpenCode HTTP API URL |
| username | string | opencode | Basic auth username |
| password | string | - | Basic auth password |
| sandboxDir | string | /root/clawd/sandbox | Sandbox directory for code |
| credentialsDir | string | /root/clawd/credentials | Credentials directory (read-only) |
| maxIterations | integer | 50 | Max iterations before stopping |
| timeoutMs | integer | 180000 | Timeout per API call (3 min) |

Usage

The plugin registers an opencode_task tool that the agent uses automatically when appropriate.

$3

`
"Create a Node.js project that connects to BCMS API"

"Refactor the Python scripts to use python-dotenv"

"Build a web scraper for product prices"
`

$3

The agent can call the tool directly:

`json
{
"tool": "opencode_task",
"params": {
"task": "Refactor knowledge_factory.py to use .env and add type hints",
"projectName": "knowledge-factory",
"continueOnError": true
}
}
`

$3

| Parameter | Type | Required | Description |
|-----------|------|----------|-------------|
| task | string | yes | Detailed task description |
| projectName | string | no | Project folder name in sandbox |
| continueOnError | boolean | no | Auto-retry on errors (default: true) |

How It Works

`
1. Create Session POST /session
↓
2. Send Task POST /session/:id/message
↓
3. Check Response
├── Tool exit code ≠ 0? → Send fix prompt
└── finish = "stop"?
↓
4. Verify Changes GET /session/:id (summary.files > 0?)
├── files > 0 → ✅ Success
└── files = 0 → Prompt to make changes
↓
5. Get Diff GET /session/:id/diff
↓
6. Return Result with file changes summary
`

Reliable Indicators

The plugin uses only reliable indicators:

| Indicator | Source | Use |
|-----------|--------|-----|
| Exit code | tool.state.metadata.exit | Error detection (≠ 0 = error) |
| File count | session.summary.files | Task completion verification |
| Diff | /session/:id/diff | Actual changes made |

Not used (unreliable):
- Text pattern matching (e.g., /Error:/i)
- "TASK_COMPLETE" marker in model output

Task Status

| Status | Meaning |
|--------|---------|
| ✅ completed | Files changed, task done |
| ⚠️ completed_no_changes | Model finished but no files changed |
| ❌ failed | Tool errors (non-zero exit codes) |
| ⏱️ max_iterations | Stopped after max attempts |

Requirements

- OpenClaw gateway running
- OpenCode server running (opencode serve --port 4096)
- Network access between OpenClaw and OpenCode

Verify Installation

`bash


List installed plugins

openclaw plugins list

Check plugin status

openclaw plugins info openclaw-opencode-supervisor
``

License

MIT