Harness Agents reference

Harness Agents Reference

Worker Agent definitions use agent.uses to select a versioned template, agent.with to configure it, and agent.inputs to declare reusable parameters. The Harness-managed template supplies the container and runtime configuration.

Worker Agent form field reference

The following fields define a Worker Agent. Required fields are marked in the Required column.

Field	Required	Description	Example
Name	Yes	Human-readable identifier displayed in the catalog and pipeline step picker.	PR Reviewer Agent
Description	No	Free-text summary of what the agent does. Helps teams discover and reuse agents from the catalog.	Reviews PRs for security, schema, and architectural issues.
Instructions	Yes	The system prompt sent to the model at runtime. Supports Harness variable expressions for dynamic context injection. Go to Configure instructions and Harness expressions to review dynamic context injection.	Example below
Model Connector	Yes	The LLM provider connector. When you configure the connector, you select a default model. Go to Configure Model Connectors for supported providers and models.	anthropic_bedrock_99cf4be5
Model Name	No	Optional override for the model used at runtime. If not specified, the agent uses the model on the Model Connector. If that model is also unset, Harness uses a provider default. Go to Configure Model Connectors. Accepts an AWS Bedrock inference profile ARN for Anthropic connectors.	arn:aws:bedrock:us-east-1:123456789012:application-inference-profile/a1b2c3d4e5f6
MCP Connectors	No	One or more MCP server connectors granting the agent access to Harness platform data and external services (such as GitHub), or a local MCP server that runs inside the agent's container. Go to Configure MCP connectors for remote and local setup.	harness_hosted_mcp
Inputs	No	Named parameters declared under agent.inputs and referenced in the prompt with <+inputs.inputName>.	projectName, orgId, repoName
Environment variables	No	String values supplied to the runtime through agent.with.env.	demo: "true"
Allowed domains	No	Hostnames or regex patterns for domains the agent can reach. Harness domains are allowed by default.	harness.io

---

Agent definition YAML reference

The following YAML shows a Worker Agent definition in the YAML tab of the Worker Agent Catalog. Replace harness_hosted_mcp with your MCP connector identifier, and set the project and organization identifiers for your environment. Input names are case-sensitive: projectName must be referenced as <+inputs.projectName>.

agent:
  uses: harnessAI@1.0.0
  with:
    prompt: |
      Analyze the pipelines in project <+inputs.projectName> in organization <+inputs.orgId>.
      Use Harness MCP to list the pipelines and retrieve each complete pipeline YAML.
      Identify repeated patterns and recommend improvements.
      Write the complete analysis to pipeline_analysis_report.md in the working directory.
      Confirm that the file exists and report its path.
      When you finish, output a variable TASK_SUCCESS with value true if the
      analysis succeeded, or false if it failed.
    connector: account.harnessAnthropic
    mcp:
      - harness_hosted_mcp
    allowed_domains: harness.io
    env:
      demo: "true"
      swarm: "true"
  inputs:
    projectName:
      type: string
      required: true
      default: Unscripted_Tour
    orgId:
      type: string
      required: true
      default: default

YAML field reference

Field	Description
agent.uses	Versioned base template. Use harnessAI@1.0.0 for the Harness AI agent runtime.
agent.with.prompt	Agent instructions. Use a YAML block scalar (`
agent.with.connector	Model connector identifier, such as account.harnessAnthropic. The connector supplies the default model and credentials.
agent.with.mcp	Optional list of MCP connector identifiers. Harness resolves their server configuration.
agent.with.allowed_domains	Optional comma-separated string of hostnames or regex patterns for domains the agent can reach over the network. Harness domains are allowed by default.
agent.with.env	Optional map of environment variable names to string values. Quote boolean-like strings such as "true". Values can reference agent inputs or Harness secrets.
agent.with.max_turns	Optional limit on agent iterations. Supply a string, such as "40".
agent.with.output	Not required for named pipeline variables. Tell the agent in Instructions which variables to output. Do not use this field to declare output variables.
agent.with.workdir	Optional working directory. Defaults to /harness.
agent.with.backend	Optional runtime backend. For the OpenAI-compatible LiteLLM setup, set openai.
agent.with.image	Optional custom agent container image.
agent.with.docker_connector	Optional Docker connector used to pull the agent image.
agent.inputs	Typed parameters referenced by the agent configuration. Each input has a type and can specify required and default.

Go to OpenAI Model Connector to configure the LiteLLM backend.