← ClaudeAtlas

inquiry_botlisted

Instructions and architectural analysis of the Universal Inquiry Bot.
ulf1/mas-inquiry · ★ 0 · AI & Automation · score 47
Install: claude install-skill ulf1/mas-inquiry
# Universal Inquiry Bot The Universal Inquiry Bot is designed to dissect a human inquiry across multiple analytical dimensions (e.g., Causal, Temporal, Structural) using a sparsely-connected multi-agent architecture inspired by biological neural networks. ## 1. LangGraph Architecture The system is orchestrated using `langgraph` as a state machine (`StateGraph`). ### Shared State (`AgentState`) - **`inquiry`**: The original user string or question. - **`active_workers`**: List of currently active dimension workers. - **`deactivated_workers`**: Workers that failed to yield improved metrics and were deactivated. - **`loop_count`**: Counter controlling the recursion/skip connections depth (up to 2 iterations). - **`stop`**: Boolean flag indicating if the iteration should stop. - **`worker_replies`**: A composite dictionary tracking answers and connections mapped to each dimension. - **`summary`**: The final synthesized output. ### Graph Nodes & Workflow 1. **Input Layer (`prelim_nodes`)**: The system broadcasts the user `inquiry` to all `active_workers` concurrently via `ThreadPoolExecutor`. Each worker agent produces an initial set of answers related to its dimension's focus and suggests links (skip connections) to other dimensions. 2. **Hidden Layers (`cross_nodes`)**: Using the connections defined in the input layer, the overarching graph feeds answers from source dimensions as `additional_context` into the LLM prompts of target dimensions. - The target workers gener