Tool bundle configuration

Tool bundle configuration

This is the page for configuring a tool bundle, not for setting up the tools that are being used for the agent. For the latter, see tools configuration.

sweagent.tools.bundle.BundleConfig pydantic-model

Bases: BaseModel

Fields:

• tools (dict[str, dict])
• state_command (str | None)

tools pydantic-field

tools: dict[str, dict]

state_command pydantic-field

state_command: str | None = None

sweagent.tools.commands.Command pydantic-model

Bases: BaseModel

Represents an executable command with arguments and documentation.

A command can be either a simple bash command or a multi-line command terminated by an end marker.

Attributes:

Name	Type	Description
name	str	The command name
docstring	str | None	Human readable description of what the command does
signature	str | None	Optional custom signature override
end_name	str | None	For multi-line commands, the terminating marker
arguments	list[Argument]	List of arguments accepted by the command

Properties

invoke_format: Format string for constructing the full command invocation

Fields:

• name (str)
• docstring (str | None)
• signature (str | None)
• end_name (str | None)
• arguments (list[Argument])

Validators:

• validate_arguments

name pydantic-field

name: str

docstring pydantic-field

docstring: str | None

signature pydantic-field

signature: str | None = None

end_name pydantic-field

end_name: str | None = None

arguments pydantic-field

arguments: list[Argument] = []

invoke_format cached property

invoke_format: str

Gets the format string for invoking this command with arguments.

Returns either the custom signature with argument placeholders replaced, or a default format of "command arg1 arg2 ...".

get_function_calling_tool

get_function_calling_tool() -> dict

Converts this command into an OpenAI function calling tool definition.

Returns:

Type	Description
dict	Dict containing the OpenAI function schema for this command

Source code in sweagent/tools/commands.py

def get_function_calling_tool(self) -> dict:
    """Converts this command into an OpenAI function calling tool definition.

    Returns:
        Dict containing the OpenAI function schema for this command
    """
    tool = {
        "type": "function",
        "function": {
            "name": self.name,
            "description": self.docstring or "",
        },
    }
    properties = {}
    required = []
    if self.arguments:
        for arg in self.arguments:
            properties[arg.name] = {"type": arg.type, "description": arg.description}

            if arg.items:
                properties[arg.name]["items"] = arg.items

            if arg.required:
                required.append(arg.name)

            # Handle enum if present
            if arg.enum:
                properties[arg.name]["enum"] = arg.enum
    tool["function"]["parameters"] = {"type": "object", "properties": properties, "required": required}
    return tool

validate_arguments pydantic-validator

validate_arguments() -> Command

Validates command argument configuration.

Checks: - Required arguments come before optional ones - Argument names are unique - Argument names match the pattern - Arguments match the signature

Returns:

Type	Description
Command	The validated Command instance

Raises:

Type	Description
ValueError	If validation fails

Source code in sweagent/tools/commands.py

@model_validator(mode="after")
def validate_arguments(self) -> Command:
    """Validates command argument configuration.

    Checks:
    - Required arguments come before optional ones
    - Argument names are unique
    - Argument names match the pattern
    - Arguments match the signature

    Returns:
        The validated Command instance

    Raises:
        ValueError: If validation fails
    """
    if not self.arguments:
        return self
    found_optional = False
    for arg in self.arguments:
        if found_optional and arg.required:
            msg = f"Command '{self.name}': Required argument '{arg.name}' cannot come after optional arguments"
            raise ValueError(msg)
        if not arg.required:
            found_optional = True

    name_counts = Counter(arg.name for arg in self.arguments)
    duplicates = {name for name, count in name_counts.items() if count > 1}
    if duplicates:
        msg = f"Command '{self.name}': Duplicate argument names: {duplicates}"
        raise ValueError(msg)
    for arg in self.arguments:
        if not re.match(ARGUMENT_NAME_PATTERN, arg.name):
            msg = f"Command '{self.name}': Invalid argument name: '{arg.name}'"
            raise ValueError(msg)
    if (invoke_keys := _extract_keys(self.invoke_format)) != {arg.name for arg in self.arguments}:
        msg = f"Command '{self.name}': Argument names ({invoke_keys}) in signature / invoke_format {self.invoke_format!r} do not match argument names"
        raise ValueError(msg)
    return self

sweagent.tools.commands.Argument pydantic-model

Bases: BaseModel

Fields:

• name (str)
• type (str)
• items (dict[str, str] | None)
• description (str)
• required (bool)
• enum (list[str] | None)
• argument_format (str)

Validators:

• validate_argument_format → argument_format

name pydantic-field

name: str

type pydantic-field

type: str

items pydantic-field

items: dict[str, str] | None = None

description pydantic-field

description: str

required pydantic-field

required: bool

enum pydantic-field

enum: list[str] | None = None

argument_format pydantic-field

argument_format: str = '{{value}}'

How to invoke the argument in the command. Make sure to use jinja syntax ({{value}}) instead of {value}).

validate_argument_format pydantic-validator

validate_argument_format(value: str) -> str

Source code in sweagent/tools/commands.py

@field_validator("argument_format")
def validate_argument_format(cls, value: str) -> str:
    _warn_probably_wrong_jinja_syntax(value)
    return value

Our projects

Mini-SWE-Agent SWE-ReX SWE-smith SWE-bench sb-cli