For the complete documentation index, see llms.txt. Markdown versions of all docs pages are available by appending .md to any docs URL.

Understand what a harness agent is, how it differs from a BYO agent, and how plugins, harnesses, and runtimes fit together.

A harness agent is an agent that you compose from existing catalog artifacts, including plugins, skills, and prompts. You do not provide the agent source code or container image for the agent. Instead, you describe the artifacts that you want the agent to use and the application (harness) that you want the agent to be compatible with, such as Claude Code or OpenClaw, in the agent resource.

When you deploy a harness agent, Solo Enterprise for agentregistry deploys a pre-built harness runner container to your deployment runtime, such as AWS Bedrock AgentCore. The runner container pulls the plugin bundles, assembles the files the harness expects, and starts the harness binary, such as Claude Code in headless mode so that the agent can run inside the container.

Harness agents stand in contrast to a BYO (bring-your-own) agent, where you write and build the agent code yourself, and publish a container image or repository URL to the catalog. During the agent deployment, Solo Enterprise for agentregistry pulls the image or packages the source code files so that the agent can run in your connected deployment runtime.

Limitations

Harness agents can only be deployed to the AWS Bedrock AgentCore deployment runtime. Deploying harness agents to Solo Enterprise for kagent is currently not supported.

How it works

Review the following diagram to see how harness agents are deployed to AWS Bedrock AgentCore.

    sequenceDiagram
    actor User
    participant Registry
    participant GitHub
    participant S3
    participant AgentCore
    participant Harness as Claude Code / OpenClaw

    User->>Registry: Add a harness agent to the catalog
    User->>Registry: Deploy the agent to AWS Bedrock AgentCore
    Registry->>GitHub: Resolve plugin git references
    GitHub-->>Registry: Plugin bundle
    Registry->>S3: Pack and store bundle
    Registry->>AgentCore: Deploy harness runner container
    AgentCore->>S3: Pull bundle at startup
    AgentCore->>Harness: Start harness binary
    User->>Harness: Chat
    Harness-->>User: Response
  

When you deploy a harness agent, the registry works through the following steps:

1. Resolve plugins and skills: The registry fetches the pinned commits for each referenced plugin and skills from the source repository.
2. Assemble the bundle: Plugin files are packaged as an archive and stored in S3.
3. Deploy the harness runner: The registry deploys a pre-built harness runner container to AgentCore. The runner downloads the bundle from S3 at startup.
4. Load the composition: The runner installs plugins, links skills into the Claude or OpenClaw config directory, and writes the MCP server configuration and system prompt.
5. Run Claude Code: For each conversation turn, the runner invokes Claude Code headlessly with the assembled context.

The result is a running agent on AgentCore that you can chat with through the Solo Enterprise for agentregistry UI or the AWS Bedrock AgentCore console.

Next steps