manifest

Class info

Classes

Name	Children	Inherits
AgentConfig llmling_agent.models.agents Configuration for a single agent in the system.		NodeConfig
AgentWorkerConfig llmling_agent_config.workers Configuration for agent workers.		BaseWorkerConfig
AgentsManifest llmling_agent.models.manifest Complete agent configuration manifest defining all available agents.		Schema
BaseMCPServerConfig llmling_agent_config.mcp_server Base model for MCP server configuration.	StdioMCPServerConfig SSEMCPServerConfig StreamableHTTPMCPServerConfig	Schema
BaseProviderConfig llmling_agent_config.providers Base configuration for agent providers.	PydanticAIProviderConfig HumanProviderConfig CallbackProviderConfig	Schema
BaseWorkerConfig llmling_agent_config.workers Base configuration for workers.	TeamWorkerConfig AgentWorkerConfig	Schema
ConversionConfig llmling_agent_config.converters Global conversion configuration.		Schema
Job llmling_agent_config.task A task is a piece of work that can be executed by an agent.		Schema Generic
MCPPoolServerConfig llmling_agent_config.pool_server Configuration for pool-based MCP server.		Schema
ObservabilityConfig llmling_agent_config.observability Global observability configuration - supports single backend only.		Schema
PromptLibraryConfig llmling_agent_config.system_prompts Complete prompt configuration.		Schema
ResourceRegistry llmling_agent.resource_registry Registry for filesystem resources.		BaseRegistry
SourceResourceConfig llmling_agent_config.resources Configuration for a single filesystem source.		BaseResourceConfig
StorageConfig llmling_agent_config.storage Global storage configuration.		Schema
StructuredResponseConfig llmling_agent_config.output_types Base class for response definitions.		Schema
TeamConfig llmling_agent_config.teams Configuration for a team or chain of message nodes.		NodeConfig
TeamWorkerConfig llmling_agent_config.workers Configuration for team workers.		BaseWorkerConfig

🛈 DocStrings

Models for agent configuration.

AgentsManifest

Bases: Schema

Complete agent configuration manifest defining all available agents.

This is the root configuration that: - Defines available response types (both inline and imported) - Configures all agent instances and their settings - Sets up custom role definitions and capabilities - Manages environment configurations

A single manifest can define multiple agents that can work independently or collaborate through the orchestrator.

Source code in src/llmling_agent/models/manifest.py

class AgentsManifest(Schema):
    """Complete agent configuration manifest defining all available agents.

    This is the root configuration that:
    - Defines available response types (both inline and imported)
    - Configures all agent instances and their settings
    - Sets up custom role definitions and capabilities
    - Manages environment configurations

    A single manifest can define multiple agents that can work independently
    or collaborate through the orchestrator.
    """

    INHERIT: str | list[str] | None = None
    """Inheritance references."""

    resources: dict[str, ResourceConfig | str] = Field(default_factory=dict)
    """Resource configurations defining available filesystems.

    Supports both full config and URI shorthand:
        resources:
          docs: "file://./docs"  # shorthand
          data:  # full config
            type: "source"
            uri: "s3://bucket/data"
            cached: true
    """

    agents: dict[str, AgentConfig] = Field(default_factory=dict)
    """Mapping of agent IDs to their configurations"""

    teams: dict[str, TeamConfig] = Field(default_factory=dict)
    """Mapping of team IDs to their configurations"""

    storage: StorageConfig = Field(default_factory=StorageConfig)
    """Storage provider configuration."""

    observability: ObservabilityConfig = Field(default_factory=ObservabilityConfig)
    """Observability provider configuration."""

    conversion: ConversionConfig = Field(default_factory=ConversionConfig)
    """Document conversion configuration."""

    responses: dict[str, StructuredResponseConfig] = Field(default_factory=dict)
    """Mapping of response names to their definitions"""

    jobs: dict[str, Job] = Field(default_factory=dict)
    """Pre-defined jobs, ready to be used by nodes."""

    mcp_servers: list[str | MCPServerConfig] = Field(default_factory=list)
    """List of MCP server configurations:

    These MCP servers are used to provide tools and other resources to the nodes.
    """
    pool_server: MCPPoolServerConfig = Field(default_factory=MCPPoolServerConfig)
    """Pool server configuration.

    This MCP server configuration is used for the pool MCP server,
    which exposes pool functionality to other applications / clients."""

    prompts: PromptLibraryConfig = Field(default_factory=PromptLibraryConfig)

    model_config = ConfigDict(use_attribute_docstrings=True, extra="forbid")

    @model_validator(mode="before")
    @classmethod
    def normalize_workers(cls, data: dict[str, Any]) -> dict[str, Any]:
        """Convert string workers to appropriate WorkerConfig for all agents."""
        teams = data.get("teams", {})
        agents = data.get("agents", {})

        # Process workers for all agents that have them
        for agent_name, agent_config in agents.items():
            if isinstance(agent_config, dict):
                workers = agent_config.get("workers", [])
            else:
                workers = agent_config.workers

            if workers:
                normalized: list[BaseWorkerConfig] = []

                for worker in workers:
                    match worker:
                        case str() as name if name in teams:
                            # Determine type based on presence in teams/agents
                            normalized.append(TeamWorkerConfig(name=name))
                        case str() as name if name in agents:
                            normalized.append(AgentWorkerConfig(name=name))
                        case str():  # Default to agent if type can't be determined
                            normalized.append(AgentWorkerConfig(name=name))

                        case dict() as config:
                            # If type is explicitly specified, use it
                            if worker_type := config.get("type"):
                                match worker_type:
                                    case "team":
                                        normalized.append(TeamWorkerConfig(**config))
                                    case "agent":
                                        normalized.append(AgentWorkerConfig(**config))
                                    case _:
                                        msg = f"Invalid worker type: {worker_type}"
                                        raise ValueError(msg)
                            else:
                                # Determine type based on worker name
                                worker_name = config.get("name")
                                if not worker_name:
                                    msg = "Worker config missing name"
                                    raise ValueError(msg)

                                if worker_name in teams:
                                    normalized.append(TeamWorkerConfig(**config))
                                else:
                                    normalized.append(AgentWorkerConfig(**config))

                        case BaseWorkerConfig():  # Already normalized
                            normalized.append(worker)

                        case _:
                            msg = f"Invalid worker configuration: {worker}"
                            raise ValueError(msg)

                if isinstance(agent_config, dict):
                    agent_config["workers"] = normalized
                else:
                    # Need to create a new dict with updated workers
                    agent_dict = agent_config.model_dump()
                    agent_dict["workers"] = normalized
                    agents[agent_name] = agent_dict

        return data

    @cached_property
    def resource_registry(self) -> ResourceRegistry:
        """Get registry with all configured resources."""
        registry = ResourceRegistry()
        for name, config in self.resources.items():
            if isinstance(config, str):
                # Convert URI shorthand to SourceResourceConfig
                config = SourceResourceConfig(uri=config)
            registry.register_from_config(name, config)
        return registry

    def clone_agent_config(
        self,
        name: str,
        new_name: str | None = None,
        *,
        template_context: dict[str, Any] | None = None,
        **overrides: Any,
    ) -> str:
        """Create a copy of an agent configuration.

        Args:
            name: Name of agent to clone
            new_name: Optional new name (auto-generated if None)
            template_context: Variables for template rendering
            **overrides: Configuration overrides for the clone

        Returns:
            Name of the new agent

        Raises:
            KeyError: If original agent not found
            ValueError: If new name already exists or if overrides invalid
        """
        if name not in self.agents:
            msg = f"Agent {name} not found"
            raise KeyError(msg)

        actual_name = new_name or f"{name}_copy_{len(self.agents)}"
        if actual_name in self.agents:
            msg = f"Agent {actual_name} already exists"
            raise ValueError(msg)

        # Deep copy the configuration
        config = self.agents[name].model_copy(deep=True)

        # Apply overrides
        for key, value in overrides.items():
            if not hasattr(config, key):
                msg = f"Invalid override: {key}"
                raise ValueError(msg)
            setattr(config, key, value)

        # Handle template rendering if context provided
        if template_context and "name" in template_context and "name" not in overrides:
            config.name = template_context["name"]

        # Note: system_prompts will be rendered during agent creation, not here
        # config.system_prompts remains as PromptConfig objects

        self.agents[actual_name] = config
        return actual_name

    @model_validator(mode="before")
    @classmethod
    def resolve_inheritance(cls, data: dict) -> dict:
        """Resolve agent inheritance chains."""
        nodes = data.get("agents", {})
        resolved: dict[str, dict] = {}
        seen: set[str] = set()

        def resolve_node(name: str) -> dict:
            if name in resolved:
                return resolved[name]

            if name in seen:
                msg = f"Circular inheritance detected: {name}"
                raise ValueError(msg)

            seen.add(name)
            config = (
                nodes[name].model_copy()
                if hasattr(nodes[name], "model_copy")
                else nodes[name].copy()
            )
            inherit = (
                config.get("inherits") if isinstance(config, dict) else config.inherits
            )
            if inherit:
                if inherit not in nodes:
                    msg = f"Parent agent {inherit} not found"
                    raise ValueError(msg)

                # Get resolved parent config
                parent = resolve_node(inherit)
                # Merge parent with child (child overrides parent)
                merged = parent.copy()
                merged.update(config)
                config = merged

            seen.remove(name)
            resolved[name] = config
            return config

        # Resolve all nodes
        for name in nodes:
            resolved[name] = resolve_node(name)

        # Update nodes with resolved configs
        data["agents"] = resolved
        return data

    @property
    def node_names(self) -> list[str]:
        """Get list of all agent and team names."""
        return list(self.agents.keys()) + list(self.teams.keys())

    @property
    def nodes(self) -> dict[str, Any]:
        """Get all agent and team configurations."""
        return {**self.agents, **self.teams}

    def get_mcp_servers(self) -> list[MCPServerConfig]:
        """Get processed MCP server configurations.

        Converts string entries to appropriate MCP server configs based on heuristics:
        - URLs ending with "/sse" -> SSE server
        - URLs starting with http(s):// -> HTTP server
        - Everything else -> stdio command

        Returns:
            List of MCPServerConfig instances

        Raises:
            ValueError: If string entry is empty
        """
        return [
            BaseMCPServerConfig.from_string(cfg) if isinstance(cfg, str) else cfg
            for cfg in self.mcp_servers
        ]

    @cached_property
    def prompt_manager(self) -> PromptManager:
        """Get prompt manager for this manifest."""
        from llmling_agent.prompts.manager import PromptManager

        return PromptManager(self.prompts)

    # @model_validator(mode="after")
    # def validate_response_types(self) -> AgentsManifest:
    #     """Ensure all agent output_types exist in responses or are inline."""
    #     for agent_id, agent in self.agents.items():
    #         if (
    #             isinstance(agent.output_type, str)
    #             and agent.output_type not in self.responses
    #         ):
    #             msg = f"'{agent.output_type=}' for '{agent_id=}' not found in responses"
    #             raise ValueError(msg)
    #     return self

    def get_agent[TAgentDeps](
        self, name: str, deps: TAgentDeps | None = None
    ) -> Agent[TAgentDeps, Any]:
        # TODO: Make this method async to support async function prompts
        from llmling import RuntimeConfig

        from llmling_agent import Agent, AgentContext

        config = self.agents[name]
        cfg = config.get_config()
        runtime = RuntimeConfig.from_config(cfg)  # Create runtime without async context
        # Create context with config path and capabilities
        context = AgentContext[TAgentDeps](
            node_name=name,
            data=deps,
            definition=self,
            config=config,
            runtime=runtime,
            # pool=self,
            # confirmation_callback=confirmation_callback,
        )

        # Resolve system prompts with new PromptConfig types
        from llmling_agent_config.system_prompts import (
            FilePromptConfig,
            FunctionPromptConfig,
            LibraryPromptConfig,
            StaticPromptConfig,
        )

        sys_prompts: list[str] = []
        for prompt in config.system_prompts:
            match prompt:
                case (str() as sys_prompt) | StaticPromptConfig(content=sys_prompt):
                    sys_prompts.append(sys_prompt)
                case FilePromptConfig(path=path, variables=variables):
                    template_path = Path(path)  # Load template from file
                    if not template_path.is_absolute() and config.config_file_path:
                        template_path = Path(config.config_file_path).parent / path

                    template_content = template_path.read_text("utf-8")
                    if variables:  # Apply variables if any
                        from jinja2 import Template

                        template = Template(template_content)
                        content = template.render(**variables)
                    else:
                        content = template_content
                    sys_prompts.append(content)
                case LibraryPromptConfig(reference=reference):
                    try:  # Load from library
                        content = self.prompt_manager.get.sync(reference)
                        sys_prompts.append(content)
                    except Exception as e:
                        msg = (
                            f"Failed to load library prompt {reference!r} "
                            f"for agent {name}"
                        )
                        logger.exception(msg)
                        raise ValueError(msg) from e
                case FunctionPromptConfig(function=function, arguments=arguments):
                    content = function(**arguments)  # Call function to get prompt content
                    sys_prompts.append(content)
        # Create agent with runtime and context
        return Agent(
            runtime=runtime,
            context=context,
            provider=config.get_provider(),
            system_prompt=sys_prompts,
            name=name,
            description=config.description,
            retries=config.retries,
            session=config.get_session_config(),
            output_retries=config.output_retries,
            end_strategy=config.end_strategy,
            debug=config.debug,
            output_type=self.get_output_type(name) or str,
            # name=config.name or name,
        )

    def get_used_providers(self) -> set[str]:
        """Get all providers configured in this manifest."""
        providers = set[str]()

        for agent_config in self.agents.values():
            match agent_config.provider:
                case ("pydantic_ai" as typ) | BaseProviderConfig(type=typ):
                    providers.add(typ)
        return providers

    @classmethod
    def from_file(cls, path: JoinablePathLike) -> Self:
        """Load agent configuration from YAML file.

        Args:
            path: Path to the configuration file

        Returns:
            Loaded agent definition

        Raises:
            ValueError: If loading fails
        """
        import yamling

        try:
            data = yamling.load_yaml_file(path, resolve_inherit=True)
            agent_def = cls.model_validate(data)
            # Update all agents with the config file path and ensure names
            agents = {
                name: config.model_copy(update={"config_file_path": str(path)})
                for name, config in agent_def.agents.items()
            }
            return agent_def.model_copy(update={"agents": agents})
        except Exception as exc:
            msg = f"Failed to load agent config from {path}"
            raise ValueError(msg) from exc

    @cached_property
    def pool(self) -> AgentPool:
        """Create an agent pool from this manifest.

        Returns:
            Configured agent pool
        """
        from llmling_agent import AgentPool

        return AgentPool(manifest=self)

    def get_output_type(self, agent_name: str) -> type[Any] | None:
        """Get the resolved result type for an agent.

        Returns None if no result type is configured.
        """
        agent_config = self.agents[agent_name]
        if not agent_config.output_type:
            return None
        logger.debug("Building response model for %r", agent_config.output_type)
        if isinstance(agent_config.output_type, str):
            response_def = self.responses[agent_config.output_type]
            return response_def.response_schema.get_schema()  # type: ignore
        return agent_config.output_type.response_schema.get_schema()  # type: ignore

INHERIT class-attribute instance-attribute

INHERIT: str | list[str] | None = None

Inheritance references.

agents class-attribute instance-attribute

agents: dict[str, AgentConfig] = Field(default_factory=dict)

Mapping of agent IDs to their configurations

conversion class-attribute instance-attribute

conversion: ConversionConfig = Field(default_factory=ConversionConfig)

Document conversion configuration.

jobs class-attribute instance-attribute

jobs: dict[str, Job] = Field(default_factory=dict)

Pre-defined jobs, ready to be used by nodes.

mcp_servers class-attribute instance-attribute

mcp_servers: list[str | MCPServerConfig] = Field(default_factory=list)

List of MCP server configurations:

These MCP servers are used to provide tools and other resources to the nodes.

node_names property

node_names: list[str]

Get list of all agent and team names.

nodes property

nodes: dict[str, Any]

Get all agent and team configurations.

observability class-attribute instance-attribute

observability: ObservabilityConfig = Field(default_factory=ObservabilityConfig)

Observability provider configuration.

pool cached property

pool: AgentPool

Create an agent pool from this manifest.

Returns:

Type	Description
AgentPool	Configured agent pool

pool_server class-attribute instance-attribute

pool_server: MCPPoolServerConfig = Field(default_factory=MCPPoolServerConfig)

Pool server configuration.

This MCP server configuration is used for the pool MCP server, which exposes pool functionality to other applications / clients.

prompt_manager cached property

prompt_manager: PromptManager

Get prompt manager for this manifest.

resource_registry cached property

resource_registry: ResourceRegistry

Get registry with all configured resources.

resources class-attribute instance-attribute

resources: dict[str, ResourceConfig | str] = Field(default_factory=dict)

Resource configurations defining available filesystems.

Supports both full config and URI shorthand

resources: docs: "file://./docs" # shorthand data: # full config type: "source" uri: "s3://bucket/data" cached: true

responses class-attribute instance-attribute

responses: dict[str, StructuredResponseConfig] = Field(default_factory=dict)

Mapping of response names to their definitions

storage class-attribute instance-attribute

storage: StorageConfig = Field(default_factory=StorageConfig)

Storage provider configuration.

teams class-attribute instance-attribute

teams: dict[str, TeamConfig] = Field(default_factory=dict)

Mapping of team IDs to their configurations

clone_agent_config

clone_agent_config(
    name: str,
    new_name: str | None = None,
    *,
    template_context: dict[str, Any] | None = None,
    **overrides: Any,
) -> str

Create a copy of an agent configuration.

Parameters:

Name	Type	Description	Default
name	str	Name of agent to clone	required
new_name	str | None	Optional new name (auto-generated if None)	None
template_context	dict[str, Any] | None	Variables for template rendering	None
**overrides	Any	Configuration overrides for the clone	{}

Returns:

Type	Description
str	Name of the new agent

Raises:

Type	Description
KeyError	If original agent not found
ValueError	If new name already exists or if overrides invalid

Source code in src/llmling_agent/models/manifest.py

def clone_agent_config(
    self,
    name: str,
    new_name: str | None = None,
    *,
    template_context: dict[str, Any] | None = None,
    **overrides: Any,
) -> str:
    """Create a copy of an agent configuration.

    Args:
        name: Name of agent to clone
        new_name: Optional new name (auto-generated if None)
        template_context: Variables for template rendering
        **overrides: Configuration overrides for the clone

    Returns:
        Name of the new agent

    Raises:
        KeyError: If original agent not found
        ValueError: If new name already exists or if overrides invalid
    """
    if name not in self.agents:
        msg = f"Agent {name} not found"
        raise KeyError(msg)

    actual_name = new_name or f"{name}_copy_{len(self.agents)}"
    if actual_name in self.agents:
        msg = f"Agent {actual_name} already exists"
        raise ValueError(msg)

    # Deep copy the configuration
    config = self.agents[name].model_copy(deep=True)

    # Apply overrides
    for key, value in overrides.items():
        if not hasattr(config, key):
            msg = f"Invalid override: {key}"
            raise ValueError(msg)
        setattr(config, key, value)

    # Handle template rendering if context provided
    if template_context and "name" in template_context and "name" not in overrides:
        config.name = template_context["name"]

    # Note: system_prompts will be rendered during agent creation, not here
    # config.system_prompts remains as PromptConfig objects

    self.agents[actual_name] = config
    return actual_name

from_file classmethod

from_file(path: JoinablePathLike) -> Self

Load agent configuration from YAML file.

Parameters:

Name	Type	Description	Default
path	JoinablePathLike	Path to the configuration file	required

Returns:

Type	Description
Self	Loaded agent definition

Raises:

Type	Description
ValueError	If loading fails

Source code in src/llmling_agent/models/manifest.py

@classmethod
def from_file(cls, path: JoinablePathLike) -> Self:
    """Load agent configuration from YAML file.

    Args:
        path: Path to the configuration file

    Returns:
        Loaded agent definition

    Raises:
        ValueError: If loading fails
    """
    import yamling

    try:
        data = yamling.load_yaml_file(path, resolve_inherit=True)
        agent_def = cls.model_validate(data)
        # Update all agents with the config file path and ensure names
        agents = {
            name: config.model_copy(update={"config_file_path": str(path)})
            for name, config in agent_def.agents.items()
        }
        return agent_def.model_copy(update={"agents": agents})
    except Exception as exc:
        msg = f"Failed to load agent config from {path}"
        raise ValueError(msg) from exc

get_mcp_servers

get_mcp_servers() -> list[MCPServerConfig]

Get processed MCP server configurations.

Converts string entries to appropriate MCP server configs based on heuristics: - URLs ending with "/sse" -> SSE server - URLs starting with http(s):// -> HTTP server - Everything else -> stdio command

Returns:

Type	Description
list[MCPServerConfig]	List of MCPServerConfig instances

Raises:

Type	Description
ValueError	If string entry is empty

Source code in src/llmling_agent/models/manifest.py

def get_mcp_servers(self) -> list[MCPServerConfig]:
    """Get processed MCP server configurations.

    Converts string entries to appropriate MCP server configs based on heuristics:
    - URLs ending with "/sse" -> SSE server
    - URLs starting with http(s):// -> HTTP server
    - Everything else -> stdio command

    Returns:
        List of MCPServerConfig instances

    Raises:
        ValueError: If string entry is empty
    """
    return [
        BaseMCPServerConfig.from_string(cfg) if isinstance(cfg, str) else cfg
        for cfg in self.mcp_servers
    ]

get_output_type

get_output_type(agent_name: str) -> type[Any] | None

Get the resolved result type for an agent.

Returns None if no result type is configured.

Source code in src/llmling_agent/models/manifest.py

def get_output_type(self, agent_name: str) -> type[Any] | None:
    """Get the resolved result type for an agent.

    Returns None if no result type is configured.
    """
    agent_config = self.agents[agent_name]
    if not agent_config.output_type:
        return None
    logger.debug("Building response model for %r", agent_config.output_type)
    if isinstance(agent_config.output_type, str):
        response_def = self.responses[agent_config.output_type]
        return response_def.response_schema.get_schema()  # type: ignore
    return agent_config.output_type.response_schema.get_schema()  # type: ignore

get_used_providers

get_used_providers() -> set[str]

Get all providers configured in this manifest.

Source code in src/llmling_agent/models/manifest.py

def get_used_providers(self) -> set[str]:
    """Get all providers configured in this manifest."""
    providers = set[str]()

    for agent_config in self.agents.values():
        match agent_config.provider:
            case ("pydantic_ai" as typ) | BaseProviderConfig(type=typ):
                providers.add(typ)
    return providers

normalize_workers classmethod

normalize_workers(data: dict[str, Any]) -> dict[str, Any]

Convert string workers to appropriate WorkerConfig for all agents.

Source code in src/llmling_agent/models/manifest.py

@model_validator(mode="before")
@classmethod
def normalize_workers(cls, data: dict[str, Any]) -> dict[str, Any]:
    """Convert string workers to appropriate WorkerConfig for all agents."""
    teams = data.get("teams", {})
    agents = data.get("agents", {})

    # Process workers for all agents that have them
    for agent_name, agent_config in agents.items():
        if isinstance(agent_config, dict):
            workers = agent_config.get("workers", [])
        else:
            workers = agent_config.workers

        if workers:
            normalized: list[BaseWorkerConfig] = []

            for worker in workers:
                match worker:
                    case str() as name if name in teams:
                        # Determine type based on presence in teams/agents
                        normalized.append(TeamWorkerConfig(name=name))
                    case str() as name if name in agents:
                        normalized.append(AgentWorkerConfig(name=name))
                    case str():  # Default to agent if type can't be determined
                        normalized.append(AgentWorkerConfig(name=name))

                    case dict() as config:
                        # If type is explicitly specified, use it
                        if worker_type := config.get("type"):
                            match worker_type:
                                case "team":
                                    normalized.append(TeamWorkerConfig(**config))
                                case "agent":
                                    normalized.append(AgentWorkerConfig(**config))
                                case _:
                                    msg = f"Invalid worker type: {worker_type}"
                                    raise ValueError(msg)
                        else:
                            # Determine type based on worker name
                            worker_name = config.get("name")
                            if not worker_name:
                                msg = "Worker config missing name"
                                raise ValueError(msg)

                            if worker_name in teams:
                                normalized.append(TeamWorkerConfig(**config))
                            else:
                                normalized.append(AgentWorkerConfig(**config))

                    case BaseWorkerConfig():  # Already normalized
                        normalized.append(worker)

                    case _:
                        msg = f"Invalid worker configuration: {worker}"
                        raise ValueError(msg)

            if isinstance(agent_config, dict):
                agent_config["workers"] = normalized
            else:
                # Need to create a new dict with updated workers
                agent_dict = agent_config.model_dump()
                agent_dict["workers"] = normalized
                agents[agent_name] = agent_dict

    return data

resolve_inheritance classmethod

resolve_inheritance(data: dict) -> dict

Resolve agent inheritance chains.

Source code in src/llmling_agent/models/manifest.py

@model_validator(mode="before")
@classmethod
def resolve_inheritance(cls, data: dict) -> dict:
    """Resolve agent inheritance chains."""
    nodes = data.get("agents", {})
    resolved: dict[str, dict] = {}
    seen: set[str] = set()

    def resolve_node(name: str) -> dict:
        if name in resolved:
            return resolved[name]

        if name in seen:
            msg = f"Circular inheritance detected: {name}"
            raise ValueError(msg)

        seen.add(name)
        config = (
            nodes[name].model_copy()
            if hasattr(nodes[name], "model_copy")
            else nodes[name].copy()
        )
        inherit = (
            config.get("inherits") if isinstance(config, dict) else config.inherits
        )
        if inherit:
            if inherit not in nodes:
                msg = f"Parent agent {inherit} not found"
                raise ValueError(msg)

            # Get resolved parent config
            parent = resolve_node(inherit)
            # Merge parent with child (child overrides parent)
            merged = parent.copy()
            merged.update(config)
            config = merged

        seen.remove(name)
        resolved[name] = config
        return config

    # Resolve all nodes
    for name in nodes:
        resolved[name] = resolve_node(name)

    # Update nodes with resolved configs
    data["agents"] = resolved
    return data