Implement a ReAct-style Agent Loop with Given APIs

Problem: Implement a ReAct-Style Agent Loop with Given APIs You need to implement a simplified ReAct (Reason + Act) agent execution loop. The system provides several APIs (as functions). Your job is to wire them into a loop so the agent can iteratively complete a task via a “think / act / observe” cycle. Given APIs (provided as pseudocode/function signatures in the interview) llm(prompt: str) -> str Returns the model output string for the current prompt. parse(output: str) -> dict Parses the model output into a structured dict. Possible keys: "thought": str "action": str "action_input": str "final": str (if present, the loop should stop) tool(action: str, action_input: str) -> str Executes the tool call and returns an observation. Task Implement react_loop(task: str, max_steps: int) -> str: Initialize a prompt containing the task. Loop up to max_steps times: Call llm(prompt). Parse with parse(...). If final exists, return it. Otherwise require action and action_input: Call tool(action, action_input) to get an observation. Append thought/action/action_input/observation to the prompt in a consistent format. If max_steps is reached without a final, return a failure message (e.g., "Max steps reached"). Constraints / Edge Cases Handle parse errors or missing fields. You may either fail fast or append an error observation to request a retry. Focus: loop structure, state/prompt accumulation, termination conditions, robustness. I/O (abstract for local testing) Input: task and max_steps. Output: final answer string. Example Input task=Find the capital of France; max_steps=3 Output Paris