How to use the Add OpenCode Provider agent skill

What does the add-opencode skill do?

The skill adds OpenCode as an optional provider for NanoClaw. It copies a set of TypeScript provider files into your NanoClaw project, registers them, and adds helper scripts for host-side setup, sign-in and updates. It is not a standalone app: it only works inside a NanoClaw checkout.

Provider choice is made per group. The source describes the change as a per-group setting, applied with the ncl command, and the host passes OPENCODE_* settings and an XDG mount into containers it starts.

• Adds OpenCode as a selectable provider in the standard setup provider picker
• Lets each chat group use OpenCode instead of the default runtime
• Includes test files for provider behaviour, useful to maintainers

What do you need before installing?

Check these requirements first. The skill refuses to apply partial changes if your NanoClaw core is too old, so an unsupported setup receives no files or dependency changes.

You also need a model provider account or sign-in for the backend you choose, for example OpenRouter, OpenAI or ChatGPT. The skill itself does not supply model access.

• NanoClaw core with the host contract and credential-connection seam versions the skill targets
• OpenCode CLI, stable release, new enough to support the --prompt option (the skill's check states the minimum)
• pnpm and tsx for running setup scripts
• Node.js for the readiness check
• A provider account or sign-in for your chosen backend

How do you install the skill?

Download the ZIP from the skill page and unzip it into ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder for one project. In the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

Inside NanoClaw, the skill's own steps copy only the listed payload files to matching paths at the project root. Do not copy ignored dependency directories or generated native-test files. Keep core-owned files such as the cwd-shim, registries and contract files where they are.

After installation, run the host setup helper for OpenCode configuration. Its --update and --debug options launch the matching operational helper.

• Back up any local edits to payload files before refreshing, because refresh overwrites them
• Use --refresh only when you want to replace the installed payload and update its pins

How do you turn OpenCode on for a chat group?

OpenCode is offered by the standard setup provider picker. For an existing install, you can add or authenticate it with the provider-auth step. Backend defaults apply to the whole installation, while model and reasoning effort can be overridden per group through the existing container configuration.

Per-group backend and auth selection is described in the source as separate work, so check the current behaviour in your version before relying on it.

• Existing installs: run the provider-auth step with opencode
• Re-authenticating leaves installed files and the container image unchanged
• Recreate affected containers after a refresh so old config symlinks are discarded

How do you trigger the skill with natural requests?

The source does not list fixed trigger phrases, so ask in plain language what you want. Name OpenCode and the provider you want, and say which chat group should change. Examples of clear requests include asking the assistant to add OpenCode as a provider, to switch one group to OpenCode with OpenRouter, or to check why OpenCode sign-in is failing.

If the assistant's answer is unclear, name the skill directly, for example by saying to use the add-opencode skill, and state the outcome you want.

• Add OpenCode as a provider in this NanoClaw checkout and walk me through sign-in
• Switch the support group to OpenCode and use a model through OpenRouter
• Check whether the OpenCode installation files match the skill's declarations

What examples show it working?

These examples describe what the skill is designed to support. Confirm each one in your own setup, because a working backend and account are only checked when a real request is sent.

• Run one group on a model from OpenRouter while other groups keep the default runtime
• Compare how a provider behaves by using the included provider test files
• Use OpenCode's read tool for channel attachments, so models that accept images or PDFs can read them

What safety notes should you read?

The skill runs scripts and uses the network. It copies many files and runs setup scripts that edit the project, so review the changes before you accept them. Host sign-in uses OpenCode's own settings and is separate from the container's gateway credentials.

The debug, update and failure-help modes launch OpenCode with a permission override, so it asks before every edit and command, even if your own config allows them. The configure mode keeps OpenCode's native permissions. Diagnostic context sent to the model may remain in OpenCode's native history, and deleting the private temporary file does not erase it.

• Do not use this where provider credentials must stay strictly isolated
• Do not use it if you cannot review code changes
• Keep backups of local payload edits before any refresh

What tips help you avoid problems?

If you only need the default Anthropic setup, skip this skill, since it adds complexity without benefit. If an installation step fails, the existing provider failure-assist hook can offer help, including for wizard authentication and installation-check failures.

To remove the skill, first switch each OpenCode group back to an installed provider with the ncl command, restart the group, and do not edit container.json files or clear database rows by hand. The removal guide lists the barrel files and skill-owned destinations to clean up.

• Use the readiness check output to confirm core is compatible before installing
• Expect ordinary sign-in not to change installed files
• Keep other tools' settings and persisted session data when refreshing