--- title: How to Create an Environment Using a Coding Adapter (public) description: Choose a Harness Runtime adapter for Codex CLI, Claude Code, or OpenCode and start a session from a YAML spec or the Control Panel. product: Managed Agents url: https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/create-managed-agent/ last_updated: "2026-10-13" --- > **For AI agents:** The documentation index is at [https://docs.digitalocean.com/llms.txt](https://docs.digitalocean.com/llms.txt). Markdown versions of pages use the same URL with `index.html.md` in place of the HTML page (for example, append `index.html.md` to the directory path instead of opening the HTML document). # How to Create an Environment Using a Coding Adapter (public) DigitalOcean Harness Runtime provides managed, hardware-isolated microVM sandboxes with built-in tools such as Chromium to run harnesses and execute arbitrary code. Rich lifecycle APIs preserve conversational history and working state across sessions, letting you pause, resume, and fork work to control costs and adapt to the nonlinear nature of agentic workflows. Scale complete agents such as Claude Code or use sandboxes independently for code execution, all through the same service. See [What You Can Build](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/details/what-you-can-build/index.html.md) for example use cases. Harness Runtime supports multiple coding agent adapters. Each adapter runs in its own sandbox. Pick an adapter, set credentials under `secrets`, and start a session from the spec. To run a framework agent instead of a coding agent, see [Run a LangGraph Agent](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/run-langgraph-agent/index.html.md) or [Run a CrewAI Agent](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/run-crewai-agent/index.html.md). ## Create a Spec File for a Coding Adapter A spec file is a YAML file that describes the environment you want. You can choose from one of the following coding adapters in your spec: ### Codex CLI Set `codex` as your `agent` and declare your OpenAI API key under `secrets`. ```yaml name: codex-agent agent: codex size: mars-2vcpu-4gb secrets: OPENAI_API_KEY: ${OPENAI_API_KEY} permissions: default: ask ``` Export the key, then save the spec as `codex-spec.yaml`: ```shell export OPENAI_API_KEY= ``` ### Claude Code Set `claude-code` as your `agent` and declare your Anthropic API key under `secrets`. ```yaml name: claude-code-agent agent: claude-code size: mars-2vcpu-4gb secrets: ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY} permissions: default: ask ``` Export the key, then save the spec as `claude-code-spec.yaml`: ```shell export ANTHROPIC_API_KEY= ``` ### OpenCode Set `opencode` as your `agent` and declare your provider API key under `secrets`. ```yaml name: opencode-agent agent: opencode size: mars-2vcpu-4gb secrets: OPENAI_API_KEY: ${OPENAI_API_KEY} permissions: default: ask ``` Export the key, then save the spec as `opencode-spec.yaml`: ```shell export OPENAI_API_KEY= ``` To pin a model, or to use DigitalOcean Inference instead of an external provider key, put the model access key under `secrets` and the non-secret endpoint settings in `env`: ```yaml env: HARNESS_INFERENCE_MODEL: deepseek-v4-pro secrets: HARNESS_INFERENCE_API_KEY: ${HARNESS_INFERENCE_API_KEY} ``` **Note**: Declare every credential under `secrets`. DigitalOcean delivers each secret to the sandbox as an environment variable, strips the `secrets` section from the stored YAML, and keeps the values in the secure secrets store. For more information about the environment spec, see the [environment spec reference](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/reference/environment-spec/index.html.md). ## Start and Attach to the Session Pass the spec file to `launch`, which creates the sandbox, starts the agent inside it, and attaches your terminal, all in one step: ```shell doctl harness-runtime launch .yaml ``` Press `Ctrl+D` to detach. The session keeps running in the cloud, and the same `launch` command reconnects you to it. Passing the name of a session that already exists resumes it rather than creating a new one: ```shell doctl harness-runtime launch ``` If you start the same setup repeatedly, save the spec as an Environment Config and start sessions from it instead of uploading the YAML each time. See [Use Environment Configs](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/use-agent-configs/index.html.md). ## Create an Environment in the Control Panel Instead of writing a spec file, you can also create an environment in the [DigitalOcean Control Panel](https://cloud.digitalocean.com/managed-agents/harness-runtime/). On the **Create Environment** page, you configure the harness, model, sandbox, tools, skills, and tool permissions in one form. 1. In the Control Panel, under **Managed Agents**, click **Harness Runtime**, and then click **Create Environment**. 2. On **Create Environment**, under **Select a Harness**, choose a coding harness (OpenCode, Codex CLI, Claude Code, or Hermes) or a framework harness (LangGraph). If **Build your own** and **Pre-built templates** tabs appear, select **Build your own** to configure a harness from scratch. If you choose a framework harness, under **Framework repository**, provide the repository the platform clones at boot. 3. Under **Model**, choose **Serverless Inference** or **Custom**. For Serverless Inference, select the model and provide a model access key when prompted. For Custom, enter the API base URL, API key, and model name. 4. Under **Configure**, choose a size under **Select a Sandbox Size**. See [Sandbox Sizes](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/details/features/index.html.md#sandbox-sizes) for available slugs and resources. Set **Outbound Network Access** to **Unrestricted** or **Restricted**, and optionally add environment variables or secrets. 5. Under **Tools**, click **Add tools** and choose the Action Gateway tools the agent can use. To let the agent run git commands in the sandbox, optionally connect **GitHub** under **CLI Authentication**: under **Use an existing connection**, select one of your GitHub connections, or click **Connect a different GitHub account**, enter an **Actor ID**, and click **Connect account** to authorize one without leaving the page. The list shows only connections you created. You can also click **Revoke** to revoke a connection here, which succeeds only if no Environment Config still uses it. Under **Tool permissions**, set the **Default permission**, and click **Add rule** to allow, ask for approval, or deny specific tools. To authorize Action Gateway providers for those tools, see [Authorize Providers with Connections](https://docs.digitalocean.com/products/managed-agents/action-gateway/how-to/manage-connections/index.html.md). 6. Under **Skills**, click **Add Skill**, then choose **Upload skill** or **Create skill**: - **Upload skill**: Select a `SKILL.md` Markdown file. The file must include YAML front matter with `name` and `description`. Review or edit the parsed skill in **Review Skill**, then save it. - **Create skill**: Enter the skill name, description, and instructions in **Create New Skill**. An environment accepts up to 32 skills. For every field a skill accepts, see [`skills` in the Environment Spec Reference](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/reference/environment-spec/index.html.md#skills). 7. Enter a **Name** for the environment, then click **Create Environment**. The environment appears in the **Environments** tab. Start a session from it by clicking **Create Session**, or from the command line as described in [Use Environment Configs](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/use-agent-configs/index.html.md). You can then start sending requests to your agent. **Note**: You cannot edit the environment configuration after it is created. If you want to make changes, you need to create a new environment. Because the environment keeps a reference to the GitHub connection it was created with, that connection cannot be revoked while the environment exists. To revoke it, delete the environment first. ## Troubleshooting Use this table to diagnose common errors when creating or attaching to a session: | Symptom | Likely cause | What to do | |---|---|---| | `unknown command "harness-runtime"` | `doctl` build doesn’t include Harness Runtime commands | Install the correct `doctl` build | | `401 Unable to authenticate` | Invalid or missing token | Recreate the token and run `doctl auth init` | | `402 Payment Required` | All of the team’s balances are empty, which blocks starting, resuming, forking, and restoring sessions | Add funds, then start the session again. See [How to Pay for Harness Runtime](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/how-to/pay-for-harness-runtime/index.html.md#empty-balances). To have a paused session resume on its own once the balance is topped off, start it with `--resume-on-topoff` | | `404` or maintenance page | Feature not enabled for the team | Confirm the account has access and is in a [supported region](https://docs.digitalocean.com/products/managed-agents/agent-harness-runtime/details/availability/index.html.md) | | `429 quota exceeded` | Concurrent session limit reached | Remove an unused session, then start again | | Run is terminal | Session ended or was destroyed | Start a new session | | Connection drops | Transient network issue | Reattach to the same session |