Load from config/key.yaml, config/config.yaml, and env (descending priority).

Observe the environment, gather relevant information, and add to memory.

Parse the generated execution plan from context.

Parse role definitions to be added from text.

Process response from SerpAPI.

Process response from SerpAPI.

If the role belongs to an env, broadcast role messages to the env.

Use LLM to coerce output into the exact schema defined by mapping. Strategy: ask the model to output ONLY a strict JSON object whose keys mat

Decide what to do next and choose the action/state.

Sequential questioning

directly traverse the `properties` in the first level. schema structure likes ``` { "title":"prd", "type":"object",

FIXME: Currently does not refresh the persisted store after add.

Add multiple Roles to the current environment.

Check if a command exists on PATH. :param command: Command to check :return: 0 if exists; non-zero otherwise

Return the number of tokens used by a list of messages. Falls back to a reasonable ChatML heuristic for unknown models instead of raising, so

Returns the number of tokens in a text string. Args: string (str): The text string. model_name (str): The name of the encodi

Create role(s) based on the plan and args.

Return all messages triggered by a specified Action

Return all messages triggered by specified Actions

Get parameters for SerpAPI.

Get payloads for Serper.

Find project root by walking up directories.

Get a specific role in the environment by name.

Return the results of a Google search using the official Google API Args: query (str): The search query. num_results (int): The n