execution

Execution runtime models for CrewMaster v2.0.0.

MODULE	DESCRIPTION
drivers	Runtime driver implementations for CrewMaster v2.0.0.
retry	Secure retry execution for CrewMaster v2.0.0.
retry_test	Tests for secure_retry_execution.
runtime	Runtime SPI models for CrewMaster v2.0.0.
runtime_test	Tests for Runtime SPI models.

CLASS	DESCRIPTION
LangChainDriver	Runtime driver backed by LangChain and LangGraph.
PydanticAIDriver	Runtime driver backed by PydanticAI.
RuntimeDriver	Protocol for runtime drivers that execute operations against LLMs.
RuntimeRequest	A request to execute an operation node against a runtime driver.
RuntimeResponse	The response from a runtime driver after executing a node.
RuntimeStreamChunk	A streaming event emitted during multi-node execution.
TokenUsage	Token usage for a single runtime invocation.

FUNCTION	DESCRIPTION
secure_retry_execution	Execute an async driver call with validation retries.

__all__ module-attribute

__all__ = ['LangChainDriver', 'PydanticAIDriver', 'RuntimeDriver', 'RuntimeRequest', 'RuntimeResponse', 'RuntimeStreamChunk', 'TokenUsage', 'secure_retry_execution']

LangChainDriver

LangChainDriver(llm: BaseChatModel)

Runtime driver backed by LangChain and LangGraph.

Encapsulates the brain-muscle loop (BrainBase, MuscleBase, graph nodes start, brain_node, muscle_node, cleaner) behind the CrewMaster Runtime SPI.

The driver takes an LLM directly (not through RunnableConfig) and tool schemas from RuntimeRequest.tools.

ATTRIBUTE	DESCRIPTION
llm	The LangChain chat model instance.

PARAMETER	DESCRIPTION
llm	A LangChain BaseChatModel that will be used for all invocations. Tool binding is handled per-request based on RuntimeRequest.tools. TYPE: BaseChatModel

METHOD	DESCRIPTION
execute	Execute a single node via the brain-muscle loop.
astream	Execute a single node with streaming through the brain-muscle loop.

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node via the brain-muscle loop.

Constructs a LangGraph graph, feeds it the runtime request, runs the brain-muscle loop, and returns a RuntimeResponse.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	A RuntimeResponse with structured output, token usage,
RuntimeResponse	and raw messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming through the brain-muscle loop.

Streams LLM output token-by-token. If tool calls are present in a complete response, the driver executes them and re-enters the loop non-streamed, then resumes streaming for the next brain call.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events (text_delta, tool_call,
AsyncIterator[RuntimeStreamChunk]	tool_result, final).

PydanticAIDriver

PydanticAIDriver(model_name: str = 'openai:gpt-4o', default_model_settings: dict[str, Any] | None = None)

Runtime driver backed by PydanticAI.

Constructs a PydanticAI Agent from each RuntimeRequest and executes it, returning structured output through the CrewMaster Runtime SPI.

ATTRIBUTE	DESCRIPTION
model_name	The model name (or KnownModelName) to use by default.
default_model_settings	Optional default settings (temperature, etc.).

PARAMETER	DESCRIPTION
model_name	Default model to use (can be overridden per-request via RuntimeRequest.runtime_hints["model"]). TYPE: str DEFAULT: 'openai:gpt-4o'
default_model_settings	Default settings like temperature, top_p, etc. TYPE: dict[str, Any] | None DEFAULT: None

METHOD	DESCRIPTION
execute	Execute a single node using PydanticAI.
astream	Execute a single node with streaming output using PydanticAI.

model_name instance-attribute

model_name = model_name

default_model_settings instance-attribute

default_model_settings = default_model_settings or {}

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node using PydanticAI.

Constructs an Agent from the request, runs it, and returns a RuntimeResponse with the structured output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	RuntimeResponse with structured output, token usage, and messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming output using PydanticAI.

Yields RuntimeStreamChunk events (text_delta, tool_call, tool_result, final) as the agent produces output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events during agent execution.

RuntimeDriver

Protocol for runtime drivers that execute operations against LLMs.

Every runtime driver (PydanticAI, LangChain, etc.) must implement this protocol. CrewMaster injects drivers per-agent via AgentConfig.runtime, with a fallback to the default_runtime passed to execute().

Usage::

class PydanticAIDriver:
    async def execute(self, request: RuntimeRequest) -> RuntimeResponse:
        ...

    async def astream(
        self, request: RuntimeRequest
    ) -> AsyncIterator[RuntimeStreamChunk]:
        ...

METHOD	DESCRIPTION
execute	Execute a single node synchronously (from the caller's perspective).
astream	Execute a single node with streaming output.

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node synchronously (from the caller's perspective).

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	A RuntimeResponse with the structured output.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events as the LLM produces output.

RuntimeRequest

A request to execute an operation node against a runtime driver.

ATTRIBUTE	DESCRIPTION
instructions	The fully assembled and rendered prompt text. TYPE: str
tools	The list of tool schemas visible to this node. TYPE: list[Any]
context	Resolved context dictionary for template variables. TYPE: dict[str, Any]
output_schema	The Pydantic model type that the output must conform to. TYPE: type[BaseModel] | None
runtime_hints	Arbitrary driver-specific hints (model name, temp, etc.). TYPE: dict[str, Any]
conversation_history	Previous messages for multi-turn execution. TYPE: list[Any]

instructions instance-attribute

instructions: str

tools class-attribute instance-attribute

tools: list[Any] = Field(default_factory=list)

context class-attribute instance-attribute

context: dict[str, Any] = Field(default_factory=dict)

output_schema class-attribute instance-attribute

output_schema: type[BaseModel] | None = None

runtime_hints class-attribute instance-attribute

runtime_hints: dict[str, Any] = Field(default_factory=dict)

conversation_history class-attribute instance-attribute

conversation_history: list[Any] = Field(default_factory=list)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

RuntimeResponse

The response from a runtime driver after executing a node.

ATTRIBUTE	DESCRIPTION
output	The structured output parsed into the requested schema (or None). TYPE: BaseModel | None
token_usage	Token accounting for this invocation. TYPE: TokenUsage
raw_messages	The raw message list from the LLM provider. TYPE: list[dict[str, Any]]
state	Arbitrary driver state that may be needed downstream. TYPE: dict[str, Any]

output class-attribute instance-attribute

output: BaseModel | None = None

token_usage class-attribute instance-attribute

token_usage: TokenUsage = Field(default_factory=TokenUsage)

raw_messages class-attribute instance-attribute

raw_messages: list[dict[str, Any]] = Field(default_factory=list)

state class-attribute instance-attribute

state: dict[str, Any] = Field(default_factory=dict)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

RuntimeStreamChunk

A streaming event emitted during multi-node execution.

This is a discriminated union keyed on kind. Each kind carries a different set of populated fields.

ATTRIBUTE	DESCRIPTION
kind	The event type. TYPE: Literal['text_delta', 'tool_call', 'tool_result', 'node_complete', 'final']
content	Text delta content (for text_delta). TYPE: str | None
tool_name	Name of the tool being called (for tool_call). TYPE: str | None
tool_args	Arguments for the tool call (for tool_call). TYPE: dict[str, Any] | None
tool_output	Output from a tool execution (for tool_result). TYPE: Any
output	The final parsed output model (for final). TYPE: BaseModel | None
node_name	The name of the completed node (for node_complete). TYPE: str | None
artifact	The intermediate artifact from the node (for node_complete). TYPE: Any

kind instance-attribute

kind: Literal['text_delta', 'tool_call', 'tool_result', 'node_complete', 'final']

content class-attribute instance-attribute

content: str | None = None

tool_name class-attribute instance-attribute

tool_name: str | None = None

tool_args class-attribute instance-attribute

tool_args: dict[str, Any] | None = None

tool_output class-attribute instance-attribute

tool_output: Any = None

output class-attribute instance-attribute

output: BaseModel | None = None

node_name class-attribute instance-attribute

node_name: str | None = None

artifact class-attribute instance-attribute

artifact: Any = None

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

TokenUsage

Token usage for a single runtime invocation.

ATTRIBUTE	DESCRIPTION
input_tokens	Number of tokens consumed by the input (prompt). TYPE: int
output_tokens	Number of tokens generated by the output. TYPE: int

input_tokens class-attribute instance-attribute

input_tokens: int = 0

output_tokens class-attribute instance-attribute

output_tokens: int = 0

secure_retry_execution async

secure_retry_execution(func: RetryFunc, max_attempts: int = 5, delay_seconds: float = 1.0, output_schema: type[BaseModel] | None = None) -> RuntimeResponse

Execute an async driver call with validation retries.

If output_schema is provided, the function validates that the runtime response output conforms to the schema. On validation failure, the error detail is recorded and an exception is raised so the caller (typically the plan executor) can retry with error feedback injected into the next request's instructions.

PARAMETER	DESCRIPTION
func	Async callable that returns a RuntimeResponse (driver call). TYPE: RetryFunc
max_attempts	Maximum number of attempts before giving up. TYPE: int DEFAULT: 5
delay_seconds	Delay (in seconds) between retry attempts. TYPE: float DEFAULT: 1.0
output_schema	Optional Pydantic model to validate the output against. TYPE: type[BaseModel] | None DEFAULT: None

RETURNS	DESCRIPTION
RuntimeResponse	The RuntimeResponse from a successful attempt.

RAISES	DESCRIPTION
RetryExhaustedError	If all attempts are exhausted.
ValidationError	If output_schema validation fails on the last attempt and the error is not wrapped.

drivers

Runtime driver implementations for CrewMaster v2.0.0.

MODULE	DESCRIPTION
langchain	LangChain runtime driver for CrewMaster v2.0.0.
langchain_test	Tests for the LangChain runtime driver.
pydantic_ai	PydanticAI runtime driver for CrewMaster v2.0.0.
pydantic_ai_test	Tests for the PydanticAI runtime driver.

CLASS	DESCRIPTION
LangChainDriver	Runtime driver backed by LangChain and LangGraph.
PydanticAIDriver	Runtime driver backed by PydanticAI.

__all__ module-attribute

__all__ = ['LangChainDriver', 'PydanticAIDriver']

LangChainDriver

LangChainDriver(llm: BaseChatModel)

Runtime driver backed by LangChain and LangGraph.

Encapsulates the brain-muscle loop (BrainBase, MuscleBase, graph nodes start, brain_node, muscle_node, cleaner) behind the CrewMaster Runtime SPI.

The driver takes an LLM directly (not through RunnableConfig) and tool schemas from RuntimeRequest.tools.

ATTRIBUTE	DESCRIPTION
llm	The LangChain chat model instance.

PARAMETER	DESCRIPTION
llm	A LangChain BaseChatModel that will be used for all invocations. Tool binding is handled per-request based on RuntimeRequest.tools. TYPE: BaseChatModel

METHOD	DESCRIPTION
execute	Execute a single node via the brain-muscle loop.
astream	Execute a single node with streaming through the brain-muscle loop.

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node via the brain-muscle loop.

Constructs a LangGraph graph, feeds it the runtime request, runs the brain-muscle loop, and returns a RuntimeResponse.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	A RuntimeResponse with structured output, token usage,
RuntimeResponse	and raw messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming through the brain-muscle loop.

Streams LLM output token-by-token. If tool calls are present in a complete response, the driver executes them and re-enters the loop non-streamed, then resumes streaming for the next brain call.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events (text_delta, tool_call,
AsyncIterator[RuntimeStreamChunk]	tool_result, final).

PydanticAIDriver

PydanticAIDriver(model_name: str = 'openai:gpt-4o', default_model_settings: dict[str, Any] | None = None)

Runtime driver backed by PydanticAI.

Constructs a PydanticAI Agent from each RuntimeRequest and executes it, returning structured output through the CrewMaster Runtime SPI.

ATTRIBUTE	DESCRIPTION
model_name	The model name (or KnownModelName) to use by default.
default_model_settings	Optional default settings (temperature, etc.).

PARAMETER	DESCRIPTION
model_name	Default model to use (can be overridden per-request via RuntimeRequest.runtime_hints["model"]). TYPE: str DEFAULT: 'openai:gpt-4o'
default_model_settings	Default settings like temperature, top_p, etc. TYPE: dict[str, Any] | None DEFAULT: None

METHOD	DESCRIPTION
execute	Execute a single node using PydanticAI.
astream	Execute a single node with streaming output using PydanticAI.

model_name instance-attribute

model_name = model_name

default_model_settings instance-attribute

default_model_settings = default_model_settings or {}

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node using PydanticAI.

Constructs an Agent from the request, runs it, and returns a RuntimeResponse with the structured output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	RuntimeResponse with structured output, token usage, and messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming output using PydanticAI.

Yields RuntimeStreamChunk events (text_delta, tool_call, tool_result, final) as the agent produces output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events during agent execution.

langchain

LangChain runtime driver for CrewMaster v2.0.0.

Implements the RuntimeDriver protocol by refactoring the existing brain-muscle loop into a LangGraph-backed driver. The driver encapsulates the BrainBase, MuscleBase, and graph node logic behind the standard execute(request) -> response interface.

Key adaptations: - _build_messages_for_llm() receives RuntimeRequest instead of BrainInput. - _parse_actions() returns RuntimeResponse instead of BrainOutput. - Helper functions (_convert_to_tool_call, _convert_to_tool_message, _is_skill_available, _is_response_structured, _convert_action_to_computation, _ensure_dict) are internalized as private methods.

CLASS	DESCRIPTION
LangChainDriver	Runtime driver backed by LangChain and LangGraph.

LangChainDriver

LangChainDriver(llm: BaseChatModel)

Runtime driver backed by LangChain and LangGraph.

Encapsulates the brain-muscle loop (BrainBase, MuscleBase, graph nodes start, brain_node, muscle_node, cleaner) behind the CrewMaster Runtime SPI.

The driver takes an LLM directly (not through RunnableConfig) and tool schemas from RuntimeRequest.tools.

ATTRIBUTE	DESCRIPTION
llm	The LangChain chat model instance.

PARAMETER	DESCRIPTION
llm	A LangChain BaseChatModel that will be used for all invocations. Tool binding is handled per-request based on RuntimeRequest.tools. TYPE: BaseChatModel

METHOD	DESCRIPTION
execute	Execute a single node via the brain-muscle loop.
astream	Execute a single node with streaming through the brain-muscle loop.

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node via the brain-muscle loop.

Constructs a LangGraph graph, feeds it the runtime request, runs the brain-muscle loop, and returns a RuntimeResponse.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	A RuntimeResponse with structured output, token usage,
RuntimeResponse	and raw messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming through the brain-muscle loop.

Streams LLM output token-by-token. If tool calls are present in a complete response, the driver executes them and re-enters the loop non-streamed, then resumes streaming for the next brain call.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events (text_delta, tool_call,
AsyncIterator[RuntimeStreamChunk]	tool_result, final).

langchain_test

Tests for the LangChain runtime driver.

Covers: - RuntimeDriver protocol compliance - execute() with FakeLLM (text responses, tool calls, brain-muscle loop) - execute() with structured output - astream() emission of RuntimeStreamChunk events - Internalized helpers (convert_to_tool_call, convert_to_tool_message, is_skill_available, is_response_structured, convert_action_to_computation, ensure_dict) - Migrated tests from brain_base_test.py and muscle_base_test.py

CLASS	DESCRIPTION
FakeLLM	Fake LLM for driver tests.
SimpleOutput
TransferOutput
TestLangChainDriverProtocol	Tests that LangChainDriver conforms to the RuntimeDriver protocol.
TestLangChainDriverExecute	Tests for LangChainDriver.execute() — migrated brain_base_test.
TestLangChainDriverMuscleLoop	Tests for the brain-muscle loop via execute().
TestLangChainDriverAstream	Tests for LangChainDriver.astream().
TestInternalizedHelpers	Tests for helper methods internalized in LangChainDriver.
TestLangChainDriverTokenUsage	Tests for token usage in response.
TestLangChainDriverEdgeCases	Tests for edge case handling.

FakeLLM

Fake LLM for driver tests.

Supports: - bind_tools (no-op passthrough) - streaming via _astream - sequential responses from messages iterator

METHOD	DESCRIPTION
bind_tools

bind_tools

bind_tools(tools: Any, **kwargs: Any) -> 'FakeLLM'

SimpleOutput

ATTRIBUTE	DESCRIPTION
answer	TYPE: str

answer instance-attribute

answer: str

TransferOutput

ATTRIBUTE	DESCRIPTION
from_account	TYPE: str
to_account	TYPE: str
amount	TYPE: float
result	TYPE: str

from_account instance-attribute

from_account: str

to_account instance-attribute

to_account: str

amount instance-attribute

amount: float

result instance-attribute

result: str

TestLangChainDriverProtocol

Tests that LangChainDriver conforms to the RuntimeDriver protocol.

METHOD	DESCRIPTION
test_is_runtime_driver	LangChainDriver should be recognized as a RuntimeDriver.
test_has_execute_method	LangChainDriver should have an async execute method.
test_has_astream_method	LangChainDriver should have an async astream method.

test_is_runtime_driver

test_is_runtime_driver()

LangChainDriver should be recognized as a RuntimeDriver.

test_has_execute_method

test_has_execute_method()

LangChainDriver should have an async execute method.

test_has_astream_method

test_has_astream_method()

LangChainDriver should have an async astream method.

TestLangChainDriverExecute

Tests for LangChainDriver.execute() — migrated brain_base_test.

METHOD	DESCRIPTION
test_hello_text_response	Text response from LLM → RuntimeResponse with raw_messages.
test_returns_runtime_response_type	execute() should always return a RuntimeResponse.
test_execute_with_output_schema_json_match	execute() should parse structured output when LLM returns valid JSON.
test_skill_non_valid_raises_error	Tool not available should raise ValueError (legacy behavior).
test_computation_required_loop	Brain returns tool call → muscle executes → brain returns text.
test_execute_with_context_injection	Context should be injected into system instructions.
test_execute_with_multiple_messages_in_history	Conversation history should be passed to the LLM.

test_hello_text_response async

test_hello_text_response()

Text response from LLM → RuntimeResponse with raw_messages.

test_returns_runtime_response_type async

test_returns_runtime_response_type()

execute() should always return a RuntimeResponse.

test_execute_with_output_schema_json_match async

test_execute_with_output_schema_json_match()

execute() should parse structured output when LLM returns valid JSON.

test_skill_non_valid_raises_error async

test_skill_non_valid_raises_error()

Tool not available should raise ValueError (legacy behavior).

test_computation_required_loop async

test_computation_required_loop()

Brain returns tool call → muscle executes → brain returns text.

test_execute_with_context_injection async

test_execute_with_context_injection()

Context should be injected into system instructions.

test_execute_with_multiple_messages_in_history async

test_execute_with_multiple_messages_in_history()

Conversation history should be passed to the LLM.

TestLangChainDriverMuscleLoop

Tests for the brain-muscle loop via execute().

METHOD	DESCRIPTION
test_multiple_tool_rounds	Multiple brain-muscle rounds: request tool A → feedback → request tool B → final.
test_direct_response_no_tools	LLM responds directly without tool calls.

test_multiple_tool_rounds async

test_multiple_tool_rounds()

Multiple brain-muscle rounds: request tool A → feedback → request tool B → final.

test_direct_response_no_tools async

test_direct_response_no_tools()

LLM responds directly without tool calls.

TestLangChainDriverAstream

Tests for LangChainDriver.astream().

METHOD	DESCRIPTION
test_astream_yields_chunks	astream() should yield RuntimeStreamChunk events.
test_astream_final_chunk	astream() should end with a final chunk.
test_astream_with_tool_calls	astream() should yield tool_call and tool_result chunks.
test_astream_multiple_tool_rounds	astream() handles multiple tool rounds.

test_astream_yields_chunks async

test_astream_yields_chunks()

astream() should yield RuntimeStreamChunk events.

test_astream_final_chunk async

test_astream_final_chunk()

astream() should end with a final chunk.

test_astream_with_tool_calls async

test_astream_with_tool_calls()

astream() should yield tool_call and tool_result chunks.

test_astream_multiple_tool_rounds async

test_astream_multiple_tool_rounds()

astream() handles multiple tool rounds.

TestInternalizedHelpers

Tests for helper methods internalized in LangChainDriver.

METHOD	DESCRIPTION
test_convert_to_tool_call_basic	_convert_to_tool_call should produce a valid ToolCall.
test_convert_to_tool_call_empty_args	_convert_to_tool_call should handle empty args.
test_convert_to_tool_call_complex_args	_convert_to_tool_call should handle complex nested args.
test_convert_to_tool_message_basic	_convert_to_tool_message should produce a valid ToolMessage.
test_convert_to_tool_message_dict	_convert_to_tool_message should stringify dict results.
test_convert_to_tool_message_numeric	_convert_to_tool_message should stringify numeric results.
test_is_skill_available_present	_is_skill_available should find a skill by name.
test_is_skill_available_not_present	_is_skill_available should return False when not found.
test_is_skill_available_empty_list	_is_skill_available should handle empty list.
test_is_skill_available_case_sensitive	_is_skill_available should be case sensitive.
test_is_response_structured_true	_is_response_structured should find structured skills.
test_is_response_structured_not_structured	_is_response_structured should return False for non-structured.
test_is_response_structured_not_found	_is_response_structured should return False when not found.
test_is_response_structured_empty	_is_response_structured should handle empty list.
test_convert_action_to_computation_with_dict	_convert_action_to_computation should handle ToolAgentAction.
test_convert_action_to_computation_with_string	_convert_action_to_computation should wrap non-dict input.
test_convert_action_to_computation_generic_dict	_convert_action_to_computation should handle AgentAction (no ID).
test_convert_action_to_computation_generic_non_dict	_convert_action_to_computation with AgentAction non-dict.
test_ensure_dict_passes_through	_ensure_dict should return dicts unchanged.
test_ensure_dict_wraps_string	_ensure_dict should wrap strings with default key.
test_ensure_dict_wraps_string_custom_key	_ensure_dict should use custom key when provided.
test_ensure_dict_wraps_int	_ensure_dict should wrap integers.
test_ensure_dict_empty_dict	_ensure_dict should return empty dict as-is.
test_ensure_dict_empty_string	_ensure_dict should wrap empty string.

test_convert_to_tool_call_basic

test_convert_to_tool_call_basic()

_convert_to_tool_call should produce a valid ToolCall.

test_convert_to_tool_call_empty_args

test_convert_to_tool_call_empty_args()

_convert_to_tool_call should handle empty args.

test_convert_to_tool_call_complex_args

test_convert_to_tool_call_complex_args()

_convert_to_tool_call should handle complex nested args.

test_convert_to_tool_message_basic

test_convert_to_tool_message_basic()

_convert_to_tool_message should produce a valid ToolMessage.

test_convert_to_tool_message_dict

test_convert_to_tool_message_dict()

_convert_to_tool_message should stringify dict results.

test_convert_to_tool_message_numeric

test_convert_to_tool_message_numeric()

_convert_to_tool_message should stringify numeric results.

test_is_skill_available_present

test_is_skill_available_present()

_is_skill_available should find a skill by name.

test_is_skill_available_not_present

test_is_skill_available_not_present()

_is_skill_available should return False when not found.

test_is_skill_available_empty_list

test_is_skill_available_empty_list()

_is_skill_available should handle empty list.

test_is_skill_available_case_sensitive

test_is_skill_available_case_sensitive()

_is_skill_available should be case sensitive.

test_is_response_structured_true

test_is_response_structured_true()

_is_response_structured should find structured skills.

test_is_response_structured_not_structured

test_is_response_structured_not_structured()

_is_response_structured should return False for non-structured.

test_is_response_structured_not_found

test_is_response_structured_not_found()

_is_response_structured should return False when not found.

test_is_response_structured_empty

test_is_response_structured_empty()

_is_response_structured should handle empty list.

test_convert_action_to_computation_with_dict

test_convert_action_to_computation_with_dict()

_convert_action_to_computation should handle ToolAgentAction.

test_convert_action_to_computation_with_string

test_convert_action_to_computation_with_string()

_convert_action_to_computation should wrap non-dict input.

test_convert_action_to_computation_generic_dict

test_convert_action_to_computation_generic_dict()

_convert_action_to_computation should handle AgentAction (no ID).

test_convert_action_to_computation_generic_non_dict

test_convert_action_to_computation_generic_non_dict()

_convert_action_to_computation with AgentAction non-dict.

test_ensure_dict_passes_through

test_ensure_dict_passes_through()

_ensure_dict should return dicts unchanged.

test_ensure_dict_wraps_string

test_ensure_dict_wraps_string()

_ensure_dict should wrap strings with default key.

test_ensure_dict_wraps_string_custom_key

test_ensure_dict_wraps_string_custom_key()

_ensure_dict should use custom key when provided.

test_ensure_dict_wraps_int

test_ensure_dict_wraps_int()

_ensure_dict should wrap integers.

test_ensure_dict_empty_dict

test_ensure_dict_empty_dict()

_ensure_dict should return empty dict as-is.

test_ensure_dict_empty_string

test_ensure_dict_empty_string()

_ensure_dict should wrap empty string.

TestLangChainDriverTokenUsage

Tests for token usage in response.

METHOD	DESCRIPTION
test_execute_returns_token_usage	execute() should return a TokenUsage in the response.
test_execute_token_usage_zero_when_no_metadata	Token usage should be zero when no usage_metadata.

test_execute_returns_token_usage async

test_execute_returns_token_usage()

execute() should return a TokenUsage in the response.

test_execute_token_usage_zero_when_no_metadata async

test_execute_token_usage_zero_when_no_metadata()

Token usage should be zero when no usage_metadata.

TestLangChainDriverEdgeCases

Tests for edge case handling.

METHOD	DESCRIPTION
test_empty_tools_list	Driver should work with empty tools list.
test_no_conversation_history	Driver should work without conversation history.
test_agent_finish_with_non_string_return	AgentFinish with non-string output should be handled.
test_no_output_schema_fallback	Without output_schema, should still produce a RuntimeResponse.

test_empty_tools_list async

test_empty_tools_list()

Driver should work with empty tools list.

test_no_conversation_history async

test_no_conversation_history()

Driver should work without conversation history.

test_agent_finish_with_non_string_return async

test_agent_finish_with_non_string_return()

AgentFinish with non-string output should be handled.

test_no_output_schema_fallback async

test_no_output_schema_fallback()

Without output_schema, should still produce a RuntimeResponse.

pydantic_ai

PydanticAI runtime driver for CrewMaster v2.0.0.

Implements the RuntimeDriver protocol using PydanticAI, providing async execution and streaming against any LLM supported by PydanticAI.

CLASS	DESCRIPTION
PydanticAIDriver	Runtime driver backed by PydanticAI.

PydanticAIDriver

PydanticAIDriver(model_name: str = 'openai:gpt-4o', default_model_settings: dict[str, Any] | None = None)

Runtime driver backed by PydanticAI.

Constructs a PydanticAI Agent from each RuntimeRequest and executes it, returning structured output through the CrewMaster Runtime SPI.

ATTRIBUTE	DESCRIPTION
model_name	The model name (or KnownModelName) to use by default.
default_model_settings	Optional default settings (temperature, etc.).

PARAMETER	DESCRIPTION
model_name	Default model to use (can be overridden per-request via RuntimeRequest.runtime_hints["model"]). TYPE: str DEFAULT: 'openai:gpt-4o'
default_model_settings	Default settings like temperature, top_p, etc. TYPE: dict[str, Any] | None DEFAULT: None

METHOD	DESCRIPTION
execute	Execute a single node using PydanticAI.
astream	Execute a single node with streaming output using PydanticAI.

model_name instance-attribute

model_name = model_name

default_model_settings instance-attribute

default_model_settings = default_model_settings or {}

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node using PydanticAI.

Constructs an Agent from the request, runs it, and returns a RuntimeResponse with the structured output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	RuntimeResponse with structured output, token usage, and messages.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming output using PydanticAI.

Yields RuntimeStreamChunk events (text_delta, tool_call, tool_result, final) as the agent produces output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events during agent execution.

pydantic_ai_test

Tests for the PydanticAI runtime driver.

Covers: - Driver construction with default model - execute() against a mock (no real LLM call) - execute() against a real model (integration test) - astream() emission of chunks - RuntimeDriver protocol compliance

CLASS	DESCRIPTION
TestPydanticAIDriverProtocol	Tests that PydanticAIDriver conforms to RuntimeDriver protocol.
TestPydanticAIDriverConstruction	Tests for driver initialization.
TestPydanticAIDriverExecute	Tests for the execute() method.
TestPydanticAIDriverAstream	Tests for the astream() method.

ATTRIBUTE	DESCRIPTION
needs_openai
pytestmark

needs_openai module-attribute

needs_openai = skipif(not get('OPENAI_API_KEY'), reason='OPENAI_API_KEY not set — skipping integration test that needs a real LLM')

pytestmark module-attribute

pytestmark = skipif(PydanticAIDriver is None, reason='pydantic_ai not installed — skipping all PydanticAIDriver tests')

TestPydanticAIDriverProtocol

Tests that PydanticAIDriver conforms to RuntimeDriver protocol.

METHOD	DESCRIPTION
test_is_runtime_driver	PydanticAIDriver should be recognized as a RuntimeDriver.
test_has_execute_method	PydanticAIDriver should have an execute method.
test_has_astream_method	PydanticAIDriver should have an astream method.

test_is_runtime_driver

test_is_runtime_driver()

PydanticAIDriver should be recognized as a RuntimeDriver.

test_has_execute_method

test_has_execute_method()

PydanticAIDriver should have an execute method.

test_has_astream_method

test_has_astream_method()

PydanticAIDriver should have an astream method.

TestPydanticAIDriverConstruction

Tests for driver initialization.

METHOD	DESCRIPTION
test_default_model_name	PydanticAIDriver should default to gpt-4o.
test_custom_model_name	PydanticAIDriver should accept a custom model name.
test_default_model_settings	PydanticAIDriver should accept default model settings.
test_no_default_settings_by_default	PydanticAIDriver should have empty default settings.

test_default_model_name

test_default_model_name()

PydanticAIDriver should default to gpt-4o.

test_custom_model_name

test_custom_model_name()

PydanticAIDriver should accept a custom model name.

test_default_model_settings

test_default_model_settings()

PydanticAIDriver should accept default model settings.

test_no_default_settings_by_default

test_no_default_settings_by_default()

PydanticAIDriver should have empty default settings.

TestPydanticAIDriverExecute

Tests for the execute() method.

METHOD	DESCRIPTION
test_execute_returns_runtime_response	execute() should return a RuntimeResponse.
test_execute_with_simple_prompt	execute() should process a simple prompt and return output.
test_execute_returns_token_usage	execute() should return token usage information.
test_execute_with_context	execute() should incorporate context into instructions.
test_execute_with_model_override	execute() should use model from runtime_hints.
test_execute_without_output_schema	execute() should work without an output schema (plain text).

test_execute_returns_runtime_response async

test_execute_returns_runtime_response()

execute() should return a RuntimeResponse.

test_execute_with_simple_prompt async

test_execute_with_simple_prompt()

execute() should process a simple prompt and return output.

test_execute_returns_token_usage async

test_execute_returns_token_usage()

execute() should return token usage information.

test_execute_with_context async

test_execute_with_context()

execute() should incorporate context into instructions.

test_execute_with_model_override async

test_execute_with_model_override()

execute() should use model from runtime_hints.

test_execute_without_output_schema async

test_execute_without_output_schema()

execute() should work without an output schema (plain text).

TestPydanticAIDriverAstream

Tests for the astream() method.

METHOD	DESCRIPTION
test_astream_yields_chunks	astream() should yield RuntimeStreamChunk events.
test_astream_returns_final_chunk	astream() should end with a final chunk.
test_astream_without_output_schema	astream() should work without an output schema.

test_astream_yields_chunks async

test_astream_yields_chunks()

astream() should yield RuntimeStreamChunk events.

test_astream_returns_final_chunk async

test_astream_returns_final_chunk()

astream() should end with a final chunk.

test_astream_without_output_schema async

test_astream_without_output_schema()

astream() should work without an output schema.

retry

Secure retry execution for CrewMaster v2.0.0.

Provides secure_retry_execution — a retry wrapper that wraps a driver call, validates the output against the expected schema, and retries with error feedback injected into the next attempt.

FUNCTION	DESCRIPTION
secure_retry_execution	Execute an async driver call with validation retries.

ATTRIBUTE	DESCRIPTION
RetryFunc

RetryFunc module-attribute

RetryFunc = Callable[[], Awaitable[RuntimeResponse]]

secure_retry_execution async

secure_retry_execution(func: RetryFunc, max_attempts: int = 5, delay_seconds: float = 1.0, output_schema: type[BaseModel] | None = None) -> RuntimeResponse

Execute an async driver call with validation retries.

If output_schema is provided, the function validates that the runtime response output conforms to the schema. On validation failure, the error detail is recorded and an exception is raised so the caller (typically the plan executor) can retry with error feedback injected into the next request's instructions.

PARAMETER	DESCRIPTION
func	Async callable that returns a RuntimeResponse (driver call). TYPE: RetryFunc
max_attempts	Maximum number of attempts before giving up. TYPE: int DEFAULT: 5
delay_seconds	Delay (in seconds) between retry attempts. TYPE: float DEFAULT: 1.0
output_schema	Optional Pydantic model to validate the output against. TYPE: type[BaseModel] | None DEFAULT: None

RETURNS	DESCRIPTION
RuntimeResponse	The RuntimeResponse from a successful attempt.

RAISES	DESCRIPTION
RetryExhaustedError	If all attempts are exhausted.
ValidationError	If output_schema validation fails on the last attempt and the error is not wrapped.

retry_test

Tests for secure_retry_execution.

Covers: - Successful execution on first attempt - Retry after transient failures - Exhaustion of all attempts - Output schema validation retries - Validation error feedback

CLASS	DESCRIPTION
MockOutput
TestSecureRetryExecution	Tests for the secure_retry_execution wrapper.

MockOutput

ATTRIBUTE	DESCRIPTION
value	TYPE: str

value instance-attribute

value: str

TestSecureRetryExecution

Tests for the secure_retry_execution wrapper.

METHOD	DESCRIPTION
test_succeeds_on_first_attempt	Should return result on first successful attempt.
test_retries_on_exception	Should retry after an exception and succeed.
test_exhausts_all_attempts	Should raise after exhausting all attempts.
test_retries_on_schema_mismatch	Should retry when output type doesn't match schema.
test_raises_validation_error_on_exhaustion	Should raise ValidationError if output schema never matches.
test_skips_validation_when_schema_is_none	Should not validate output when no schema is provided.
test_skips_validation_when_output_is_none	Should not validate when response output is None.
test_uses_custom_max_attempts	Should respect custom max_attempts.
test_uses_custom_delay	Should respect custom delay between retries.
test_default_parameters	Should work with default max_attempts and delay.

test_succeeds_on_first_attempt async

test_succeeds_on_first_attempt()

Should return result on first successful attempt.

test_retries_on_exception async

test_retries_on_exception()

Should retry after an exception and succeed.

test_exhausts_all_attempts async

test_exhausts_all_attempts()

Should raise after exhausting all attempts.

test_retries_on_schema_mismatch async

test_retries_on_schema_mismatch()

Should retry when output type doesn't match schema.

test_raises_validation_error_on_exhaustion async

test_raises_validation_error_on_exhaustion()

Should raise ValidationError if output schema never matches.

test_skips_validation_when_schema_is_none async

test_skips_validation_when_schema_is_none()

Should not validate output when no schema is provided.

test_skips_validation_when_output_is_none async

test_skips_validation_when_output_is_none()

Should not validate when response output is None.

test_uses_custom_max_attempts async

test_uses_custom_max_attempts()

Should respect custom max_attempts.

test_uses_custom_delay async

test_uses_custom_delay()

Should respect custom delay between retries.

test_default_parameters async

test_default_parameters()

Should work with default max_attempts and delay.

runtime

Runtime SPI models for CrewMaster v2.0.0.

The Runtime SPI defines the contract between CrewMaster's execution layer and concrete LLM runtimes (PydanticAI, LangChain, etc.). Every driver must implement the RuntimeDriver protocol.

Key types:

• RuntimeRequest: The instruction, tools, and context sent to a runtime.
• RuntimeResponse: The structured output, token usage, and raw messages returned by a runtime.
• RuntimeStreamChunk: A discriminated union of streaming event types.
• RuntimeDriver: The async protocol that all runtime drivers must implement.
• TokenUsage: Token accounting for a single runtime invocation.

CLASS	DESCRIPTION
TokenUsage	Token usage for a single runtime invocation.
RuntimeRequest	A request to execute an operation node against a runtime driver.
RuntimeResponse	The response from a runtime driver after executing a node.
RuntimeStreamChunk	A streaming event emitted during multi-node execution.
RuntimeDriver	Protocol for runtime drivers that execute operations against LLMs.

TokenUsage

Token usage for a single runtime invocation.

ATTRIBUTE	DESCRIPTION
input_tokens	Number of tokens consumed by the input (prompt). TYPE: int
output_tokens	Number of tokens generated by the output. TYPE: int

input_tokens class-attribute instance-attribute

input_tokens: int = 0

output_tokens class-attribute instance-attribute

output_tokens: int = 0

RuntimeRequest

A request to execute an operation node against a runtime driver.

ATTRIBUTE	DESCRIPTION
instructions	The fully assembled and rendered prompt text. TYPE: str
tools	The list of tool schemas visible to this node. TYPE: list[Any]
context	Resolved context dictionary for template variables. TYPE: dict[str, Any]
output_schema	The Pydantic model type that the output must conform to. TYPE: type[BaseModel] | None
runtime_hints	Arbitrary driver-specific hints (model name, temp, etc.). TYPE: dict[str, Any]
conversation_history	Previous messages for multi-turn execution. TYPE: list[Any]

instructions instance-attribute

instructions: str

tools class-attribute instance-attribute

tools: list[Any] = Field(default_factory=list)

context class-attribute instance-attribute

context: dict[str, Any] = Field(default_factory=dict)

output_schema class-attribute instance-attribute

output_schema: type[BaseModel] | None = None

runtime_hints class-attribute instance-attribute

runtime_hints: dict[str, Any] = Field(default_factory=dict)

conversation_history class-attribute instance-attribute

conversation_history: list[Any] = Field(default_factory=list)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

RuntimeResponse

The response from a runtime driver after executing a node.

ATTRIBUTE	DESCRIPTION
output	The structured output parsed into the requested schema (or None). TYPE: BaseModel | None
token_usage	Token accounting for this invocation. TYPE: TokenUsage
raw_messages	The raw message list from the LLM provider. TYPE: list[dict[str, Any]]
state	Arbitrary driver state that may be needed downstream. TYPE: dict[str, Any]

output class-attribute instance-attribute

output: BaseModel | None = None

token_usage class-attribute instance-attribute

token_usage: TokenUsage = Field(default_factory=TokenUsage)

raw_messages class-attribute instance-attribute

raw_messages: list[dict[str, Any]] = Field(default_factory=list)

state class-attribute instance-attribute

state: dict[str, Any] = Field(default_factory=dict)

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

RuntimeStreamChunk

A streaming event emitted during multi-node execution.

This is a discriminated union keyed on kind. Each kind carries a different set of populated fields.

ATTRIBUTE	DESCRIPTION
kind	The event type. TYPE: Literal['text_delta', 'tool_call', 'tool_result', 'node_complete', 'final']
content	Text delta content (for text_delta). TYPE: str | None
tool_name	Name of the tool being called (for tool_call). TYPE: str | None
tool_args	Arguments for the tool call (for tool_call). TYPE: dict[str, Any] | None
tool_output	Output from a tool execution (for tool_result). TYPE: Any
output	The final parsed output model (for final). TYPE: BaseModel | None
node_name	The name of the completed node (for node_complete). TYPE: str | None
artifact	The intermediate artifact from the node (for node_complete). TYPE: Any

kind instance-attribute

kind: Literal['text_delta', 'tool_call', 'tool_result', 'node_complete', 'final']

content class-attribute instance-attribute

content: str | None = None

tool_name class-attribute instance-attribute

tool_name: str | None = None

tool_args class-attribute instance-attribute

tool_args: dict[str, Any] | None = None

tool_output class-attribute instance-attribute

tool_output: Any = None

output class-attribute instance-attribute

output: BaseModel | None = None

node_name class-attribute instance-attribute

node_name: str | None = None

artifact class-attribute instance-attribute

artifact: Any = None

model_config class-attribute instance-attribute

model_config = {'arbitrary_types_allowed': True}

RuntimeDriver

Protocol for runtime drivers that execute operations against LLMs.

Every runtime driver (PydanticAI, LangChain, etc.) must implement this protocol. CrewMaster injects drivers per-agent via AgentConfig.runtime, with a fallback to the default_runtime passed to execute().

Usage::

class PydanticAIDriver:
    async def execute(self, request: RuntimeRequest) -> RuntimeResponse:
        ...

    async def astream(
        self, request: RuntimeRequest
    ) -> AsyncIterator[RuntimeStreamChunk]:
        ...

METHOD	DESCRIPTION
execute	Execute a single node synchronously (from the caller's perspective).
astream	Execute a single node with streaming output.

execute async

execute(request: RuntimeRequest) -> RuntimeResponse

Execute a single node synchronously (from the caller's perspective).

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

RETURNS	DESCRIPTION
RuntimeResponse	A RuntimeResponse with the structured output.

astream async

astream(request: RuntimeRequest) -> AsyncIterator[RuntimeStreamChunk]

Execute a single node with streaming output.

PARAMETER	DESCRIPTION
request	The fully assembled runtime request. TYPE: RuntimeRequest

YIELDS	DESCRIPTION
AsyncIterator[RuntimeStreamChunk]	RuntimeStreamChunk events as the LLM produces output.

runtime_test

Tests for Runtime SPI models.

CLASS	DESCRIPTION
TestTokenUsage
TestRuntimeRequest
TestRuntimeResponse
TestRuntimeStreamChunk
TestRuntimeDriverProtocol

TestTokenUsage

METHOD	DESCRIPTION
test_defaults	TokenUsage should default to zero tokens.
test_full_construction	TokenUsage should accept token counts.
test_serialization_roundtrip	TokenUsage should serialize and deserialize.

test_defaults

test_defaults()

TokenUsage should default to zero tokens.

test_full_construction

test_full_construction()

TokenUsage should accept token counts.

test_serialization_roundtrip

test_serialization_roundtrip()

TokenUsage should serialize and deserialize.

TestRuntimeRequest

METHOD	DESCRIPTION
test_defaults	RuntimeRequest should have sensible defaults for optional fields.
test_full_construction	RuntimeRequest should accept all fields.
test_instructions_required	RuntimeRequest should require instructions.
test_accepts_arbitrary_output_schema_type	RuntimeRequest should accept Pydantic model types.

test_defaults

test_defaults()

RuntimeRequest should have sensible defaults for optional fields.

test_full_construction

test_full_construction()

RuntimeRequest should accept all fields.

test_instructions_required

test_instructions_required()

RuntimeRequest should require instructions.

test_accepts_arbitrary_output_schema_type

test_accepts_arbitrary_output_schema_type()

RuntimeRequest should accept Pydantic model types.

TestRuntimeResponse

METHOD	DESCRIPTION
test_defaults	RuntimeResponse should have sensible defaults.
test_full_construction	RuntimeResponse should accept all fields.
test_output_accepts_none	RuntimeResponse.output should accept None.
test_accepts_pydantic_output	RuntimeResponse should accept Pydantic model instances.
test_serialization_roundtrip	RuntimeResponse should serialize and deserialize.

test_defaults

test_defaults()

RuntimeResponse should have sensible defaults.

test_full_construction

test_full_construction()

RuntimeResponse should accept all fields.

test_output_accepts_none

test_output_accepts_none()

RuntimeResponse.output should accept None.

test_accepts_pydantic_output

test_accepts_pydantic_output()

RuntimeResponse should accept Pydantic model instances.

test_serialization_roundtrip

test_serialization_roundtrip()

RuntimeResponse should serialize and deserialize.

TestRuntimeStreamChunk

METHOD	DESCRIPTION
test_text_delta_kind	RuntimeStreamChunk with kind='text_delta' should carry content.
test_tool_call_kind	RuntimeStreamChunk with kind='tool_call' should carry tool info.
test_tool_result_kind	RuntimeStreamChunk with kind='tool_result' should carry tool output.
test_node_complete_kind	RuntimeStreamChunk with kind='node_complete' should carry node info.
test_final_kind	RuntimeStreamChunk with kind='final' should carry output.
test_invalid_kind_rejected	RuntimeStreamChunk should reject invalid kinds.
test_defaults_on_other_fields	RuntimeStreamChunk should default unused fields to None.
test_serialization_roundtrip	RuntimeStreamChunk should serialize and deserialize.

test_text_delta_kind

test_text_delta_kind()

RuntimeStreamChunk with kind='text_delta' should carry content.

test_tool_call_kind

test_tool_call_kind()

RuntimeStreamChunk with kind='tool_call' should carry tool info.

test_tool_result_kind

test_tool_result_kind()

RuntimeStreamChunk with kind='tool_result' should carry tool output.

test_node_complete_kind

test_node_complete_kind()

RuntimeStreamChunk with kind='node_complete' should carry node info.

test_final_kind

test_final_kind()

RuntimeStreamChunk with kind='final' should carry output.

test_invalid_kind_rejected

test_invalid_kind_rejected()

RuntimeStreamChunk should reject invalid kinds.

test_defaults_on_other_fields

test_defaults_on_other_fields()

RuntimeStreamChunk should default unused fields to None.

test_serialization_roundtrip

test_serialization_roundtrip()

RuntimeStreamChunk should serialize and deserialize.

TestRuntimeDriverProtocol

METHOD	DESCRIPTION
test_is_runtime_checkable	RuntimeDriver should be a runtime-checkable protocol.
test_conforming_class_passes_isinstance	A class implementing execute() and astream() should be recognized.
test_non_conforming_class_fails_isinstance	A class without execute/astream should not be recognized.
test_missing_astream_fails	A class with execute but no astream should not be recognized.
test_missing_execute_fails	A class with astream but no execute should not be recognized.

test_is_runtime_checkable

test_is_runtime_checkable()

RuntimeDriver should be a runtime-checkable protocol.

test_conforming_class_passes_isinstance

test_conforming_class_passes_isinstance()

A class implementing execute() and astream() should be recognized.

test_non_conforming_class_fails_isinstance

test_non_conforming_class_fails_isinstance()

A class without execute/astream should not be recognized.

test_missing_astream_fails

test_missing_astream_fails()

A class with execute but no astream should not be recognized.

test_missing_execute_fails

test_missing_execute_fails()

A class with astream but no execute should not be recognized.