@meego-harness/opencode-worker

Standalone worker bridge that runs tasks through the OpenCode CLI and connects to a meego-harness worker server.

Commands

meego-opencode-worker setup
meego-opencode-worker list
meego-opencode-worker doctor
meego-opencode-worker start --worker <workerId> [--tmux] [--opencode-shell]
meego-opencode-worker stop --worker <workerId>
meego-opencode-worker stop-all
meego-opencode-worker enable --worker <workerId>
meego-opencode-worker disable --worker <workerId>
meego-opencode-worker uninstall --worker <workerId>

Use --opencode-shell when opencode is only available through the user's interactive shell configuration.

Setup Inputs

Non-interactive setup accepts:

meego-opencode-worker setup \
  --server-url ws://127.0.0.1:3000/workers \
  --email [email protected] \
  --worker opencode-worker-1 \
  --capability-summary "Handles TypeScript work" \
  --workspace /path/to/workspace \
  --permission-preset default \
  --model anthropic/claude-sonnet-4-5 \
  --variant high \
  --agent build

The worker runs tasks with opencode run --format json. --permission-preset full-access adds --dangerously-skip-permissions; safe and default leave OpenCode permission handling unchanged.

| Preset | OpenCode behavior | | --- | --- | | safe | Do not skip permissions | | default | Do not skip permissions | | full-access | Pass --dangerously-skip-permissions |

Task Metadata

| Key | Type | Description | | --- | --- | --- | | repo | string | Repository directory under the configured workspace or an explicit repo mapping | | openCodeVariant | string | Provider-specific model variant, for example high or max | | openCodeAgent | string | OpenCode agent name, for example build or plan | | openCodePlanMode | boolean | Prepends a planning-only preamble on the first message in that context |

Operational Notes

• The worker server treats this bridge as a normal worker.
• One OpenCode session is reused per harness contextId.
• Task cancellation aborts the active OpenCode process when possible.
• Repo routing follows the same rules as codex-worker: absent repo uses the default workspace, empty mappings infer <workspace>/<repo>, and non-empty mappings require an explicit match.