Task Configuration

Configuration models for tasks, agents, environments, and verifiers

TaskConfig (Task-Level)

Defines task-level configuration from task.toml files.

Import: from harbor.models.task.config import TaskConfig

Fields

version · str · default: "'1.0'"

Task configuration version.

metadata · dict[str, Any] · default: "{}"

Arbitrary task metadata (author, description, tags, etc.).

verifier · VerifierConfig · default: "VerifierConfig()"

Verifier configuration for this task.

agent · AgentConfig · default: "AgentConfig()"

Agent execution configuration.

environment · EnvironmentConfig · default: "EnvironmentConfig()"

Environment resource configuration.

solution · SolutionConfig · default: "SolutionConfig()"

Solution script configuration.

source · str | None · default: "None"

Source dataset name if task is from a dataset.

Methods

model_validate_toml

@classmethod
def model_validate_toml(cls, toml_data: str) -> "TaskConfig"

Create a TaskConfig from TOML string.

toml_data · str · required

TOML-formatted configuration string.

TaskConfig · TaskConfig

Parsed configuration object.

model_dump_toml

def model_dump_toml(self) -> str

Serialize TaskConfig to TOML string.

toml_string · str

TOML-formatted configuration.

Example

from harbor.models.task.config import TaskConfig

# Load from TOML file
toml_content = Path("task.toml").read_text()
task_config = TaskConfig.model_validate_toml(toml_content)

# Create programmatically
task_config = TaskConfig(
    version="1.0",
    metadata={
        "author": "Harbor Team",
        "difficulty": "hard",
        "tags": ["python", "testing"]
    },
    agent=AgentConfig(timeout_sec=900.0),
    environment=EnvironmentConfig(
        cpus=2,
        memory_mb=4096,
        gpus=1
    ),
    verifier=VerifierConfig(timeout_sec=300.0)
)

# Export to TOML
toml_str = task_config.model_dump_toml()
Path("task.toml").write_text(toml_str)

AgentConfig (Task-Level)

Agent execution settings in task configuration.

Import: from harbor.models.task.config import AgentConfig

Fields

timeout_sec · float · default: "600.0"

Agent execution timeout in seconds (10 minutes default).

Example

from harbor.models.task.config import AgentConfig

agent_config = AgentConfig(
    timeout_sec=1800.0  # 30 minutes
)

EnvironmentConfig (Task-Level)

Environment resource configuration in task definition.

Import: from harbor.models.task.config import EnvironmentConfig

Fields

build_timeout_sec · float · default: "600.0"

Environment build timeout in seconds (10 minutes default).

docker_image · str | None · default: "None"

Pre-built Docker image to use instead of building from Dockerfile.

cpus · int · default: "1"

Number of CPU cores allocated to the environment.

memory_mb · int · default: "2048"

Memory allocation in megabytes (2GB default).

storage_mb · int · default: "10240"

Storage allocation in megabytes (10GB default).

gpus · int · default: "0"

Number of GPUs to allocate.

gpu_types · list[str] | None · default: "None"

Acceptable GPU types (e.g., ['H100', 'A100', 'T4']). None means any GPU is acceptable.

allow_internet · bool · default: "True"

Whether to allow internet access in the environment.

mcp_servers · list[MCPServerConfig] · default: "[]"

MCP (Model Context Protocol) servers available to agents.

skills_dir · str | None · default: "None"

Path to skills directory in the environment. Contents are copied to the agent’s skills directory.

Deprecated Fields

memory · str | None · deprecated

Deprecated. Use memory_mb instead. Format: '2G', '512M', etc.

storage · str | None · deprecated

Deprecated. Use storage_mb instead. Format: '10G', '5120M', etc.

Example

from harbor.models.task.config import EnvironmentConfig, MCPServerConfig

env_config = EnvironmentConfig(
    build_timeout_sec=900.0,
    cpus=4,
    memory_mb=8192,  # 8GB
    storage_mb=20480,  # 20GB
    gpus=1,
    gpu_types=["H100", "A100"],
    allow_internet=True,
    mcp_servers=[
        MCPServerConfig(
            name="filesystem",
            transport="stdio",
            command="mcp-server-filesystem",
            args=["--root", "/workspace"]
        )
    ],
    skills_dir="/workspace/.skills"
)

VerifierConfig (Task-Level)

Verifier execution configuration.

Import: from harbor.models.task.config import VerifierConfig

Fields

timeout_sec · float · default: "600.0"

Verifier execution timeout in seconds (10 minutes default).

env · dict[str, str] · default: "{}"

Environment variables for the verifier script.

Example

from harbor.models.task.config import VerifierConfig

verifier_config = VerifierConfig(
    timeout_sec=300.0,  # 5 minutes
    env={
        "EXPECTED_OUTPUT": "success",
        "TOLERANCE": "0.01"
    }
)

SolutionConfig

Solution script configuration.

Import: from harbor.models.task.config import SolutionConfig

Fields

env · dict[str, str] · default: "{}"

Environment variables for the solution script.

Example

from harbor.models.task.config import SolutionConfig

solution_config = SolutionConfig(
    env={
        "API_KEY": "test-key",
        "MODE": "production"
    }
)

MCPServerConfig

Configuration for an MCP (Model Context Protocol) server.

Import: from harbor.models.task.config import MCPServerConfig

Fields

name · str · required

Name of the MCP server.

transport · str · default: "'sse'"

Transport protocol: 'sse', 'streamable-http', or 'stdio'.

url · str | None · default: "None"

Server URL (required for sse and streamable-http transports).

command · str | None · default: "None"

Command to start the server (required for stdio transport).

args · list[str] · default: "[]"

Command-line arguments (for stdio transport).

Validation

The config validates that:

• url is provided for sse and streamable-http transports
• command is provided for stdio transport

Examples

SSE Transport

from harbor.models.task.config import MCPServerConfig

mcp_sse = MCPServerConfig(
    name="remote-filesystem",
    transport="sse",
    url="https://mcp.example.com/filesystem"
)

STDIO Transport

mcp_stdio = MCPServerConfig(
    name="local-git",
    transport="stdio",
    command="npx",
    args=["-y", "@modelcontextprotocol/server-git"]
)

Multiple Servers

from harbor.models.task.config import EnvironmentConfig, MCPServerConfig

env_config = EnvironmentConfig(
    mcp_servers=[
        MCPServerConfig(
            name="filesystem",
            transport="stdio",
            command="mcp-server-filesystem",
            args=["--root", "/workspace"]
        ),
        MCPServerConfig(
            name="git",
            transport="stdio",
            command="npx",
            args=["-y", "@modelcontextprotocol/server-git"]
        ),
        MCPServerConfig(
            name="remote-db",
            transport="sse",
            url="https://mcp.example.com/database"
        )
    ]
)

Complete task.toml Example

version = "1.0"

[metadata]
author = "Harbor Team"
difficulty = "medium"
tags = ["python", "api", "testing"]
description = "Build and test a REST API"

[agent]
timeout_sec = 1200.0

[environment]
build_timeout_sec = 600.0
cpus = 2
memory_mb = 4096
storage_mb = 15360
gpus = 0
allow_internet = true
skills_dir = "/workspace/.skills"

[[environment.mcp_servers]]
name = "filesystem"
transport = "stdio"
command = "mcp-server-filesystem"
args = ["--root", "/workspace"]

[[environment.mcp_servers]]
name = "git"
transport = "stdio"
command = "npx"
args = ["-y", "@modelcontextprotocol/server-git"]

[verifier]
timeout_sec = 300.0

[verifier.env]
EXPECTED_STATUS = "200"
TOLERANCE = "0.01"

[solution]

[solution.env]
API_KEY = "test-key"

Related Types

• TrialConfig - Trial-level configuration that uses these types
• JobConfig - Job-level configuration
• BaseAgent - Accesses mcp_servers and skills_dir
• BaseEnvironment - Manages environment resources