Tool API

Tool function

def my_tool(input_data: Any, params: dict) -> Any:

Parameter	Description
`input_data`	Output of the previous step, or raw input for the first step
`params`	Dict from the workflow step's `params` field
Return	Processed data, becomes `input_data` for the next step

Condition function

def my_condition(context: dict) -> bool:

Parameter	Description
`context`	Dict containing `last_output` from the previous step
Return	`True` to run the step, `False` to skip it

Registration

Every tool file must define a register function:

def register(manager: LocalToolManager):
    manager.register_tool("tool_name", tool_function)
    manager.register_condition("condition_name", condition_function)

`manager.register_tool(name, func)`

Register a tool function under the given name. The name is what workflow JSON files reference.

`manager.register_condition(name, func)`

Register a condition function under the given name.

`manager.run_tool_chain(input_data, chain, context=None)`

Execute a sequence of tool steps. Called internally by the engine. You generally don't call this directly.

`manager.load_project_tools(tools_dir)`

Discover and load all tool files from a directory. Called internally by the engine.

File discovery rules

Scans tools/ for .py files
Files starting with _ are skipped
Files without a register function are warned and skipped
Files with import errors or exceptions during loading are logged and skipped. One bad file doesn't break everything
Built-in tools load first, then user tools
User tools can override builtins by registering the same name (logged as a warning)

Validation

Tool functions must accept at least 2 parameters (input_data, params). Registration will raise TypeError if the signature doesn't match.

Condition and tool names must be registered before the tool chain runs. Missing names raise ValueError with a clear message.