""" MARK XL — Task Planner Replaces google.generativeai with local Ollama via core.llm_client. """ import json import re import sys from pathlib import Path from core.llm_client import call_llm_text def get_base_dir() -> Path: if getattr(sys, "frozen", False): return Path(sys.executable).parent return Path(__file__).resolve().parent.parent BASE_DIR = get_base_dir() PLANNER_PROMPT = """You are the planning module of MARK XL, a personal AI assistant. Your job: break any user goal into a sequence of steps using ONLY the tools listed below. ABSOLUTE RULES: - NEVER use generated_code or write Python scripts. It does not exist. - NEVER reference previous step results in parameters. Every step is independent. - Use web_search for ANY information retrieval, research, or current data. - Use file_controller to save content to disk. - Max 5 steps. Use the minimum steps needed. AVAILABLE TOOLS AND THEIR PARAMETERS: open_app app_name: string (required) web_search query: string (required) — write a clear, focused search query mode: "search" or "compare" (optional, default: search) items: list of strings (optional, for compare mode) aspect: string (optional, for compare mode) game_updater action: "update" | "install" | "list" | "download_status" | "schedule" (required) platform: "steam" | "epic" | "both" (optional, default: both) game_name: string (optional) app_id: string (optional) shutdown_when_done: boolean (optional) browser_control action: "go_to" | "search" | "click" | "type" | "scroll" | "get_text" | "press" | "close" (required) url: string (for go_to) query: string (for search) text: string (for click/type) direction: "up" | "down" (for scroll) file_controller action: "write" | "create_file" | "read" | "list" | "delete" | "move" | "copy" | "find" | "disk_usage" (required) path: string — use "desktop" for Desktop folder name: string — filename content: string — file content (for write/create_file) computer_settings action: string (required) description: string — natural language description value: string (optional) computer_control action: "type" | "click" | "hotkey" | "press" | "scroll" | "screenshot" | "screen_find" | "screen_click" (required) text: string (for type) x, y: int (for click) keys: string (for hotkey, e.g. "ctrl+c") key: string (for press) direction: "up" | "down" (for scroll) description: string (for screen_find/screen_click) screen_process text: string (required) — what to analyze or ask about the screen angle: "screen" | "camera" (optional) send_message receiver: string (required) message_text: string (required) platform: string (required) reminder date: string YYYY-MM-DD (required) time: string HH:MM (required) message: string (required) desktop_control action: "wallpaper" | "organize" | "clean" | "list" | "task" (required) path: string (optional) task: string (optional) youtube_video action: "play" | "summarize" | "trending" (required) query: string (for play) weather_report city: string (required) flight_finder origin: string (required) destination: string (required) date: string (required) code_helper action: "write" | "edit" | "run" | "explain" (required) description: string (required) language: string (optional) output_path: string (optional) file_path: string (optional) dev_agent description: string (required) language: string (optional) OUTPUT — return ONLY valid JSON, no markdown, no explanation, no code blocks: { "goal": "...", "steps": [ { "step": 1, "tool": "tool_name", "description": "what this step does", "parameters": {}, "critical": true } ] } """ def create_plan(goal: str, context: str = "") -> dict: user_input = f"Goal: {goal}" if context: user_input += f"\n\nContext: {context}" try: text = call_llm_text(user_input, system=PLANNER_PROMPT) text = re.sub(r"```(?:json)?", "", text).strip().rstrip("`").strip() plan = json.loads(text) if "steps" not in plan or not isinstance(plan["steps"], list): raise ValueError("Invalid plan structure") for step in plan["steps"]: if step.get("tool") == "generated_code": print(f"[Planner] ⚠️ generated_code in step {step.get('step')} — replacing with web_search") step["tool"] = "web_search" step["parameters"] = {"query": step.get("description", goal)[:200]} print(f"[Planner] ✅ Plan: {len(plan['steps'])} steps") for s in plan["steps"]: print(f" Step {s['step']}: [{s['tool']}] {s['description']}") return plan except json.JSONDecodeError as e: print(f"[Planner] ⚠️ JSON parse failed: {e}") return _fallback_plan(goal) except Exception as e: print(f"[Planner] ⚠️ Planning failed: {e}") return _fallback_plan(goal) def _fallback_plan(goal: str) -> dict: print("[Planner] 🔄 Fallback plan") return { "goal": goal, "steps": [ { "step": 1, "tool": "web_search", "description": f"Search for: {goal}", "parameters": {"query": goal}, "critical": True, } ], } def replan(goal: str, completed_steps: list, failed_step: dict, error: str) -> dict: completed_summary = "\n".join( f" - Step {s['step']} ({s['tool']}): DONE" for s in completed_steps ) prompt = f"""Goal: {goal} Already completed: {completed_summary if completed_summary else ' (none)'} Failed step: [{failed_step.get('tool')}] {failed_step.get('description')} Error: {error} Create a REVISED plan for the remaining work only. Do not repeat completed steps.""" try: text = call_llm_text(prompt, system=PLANNER_PROMPT) text = re.sub(r"```(?:json)?", "", text).strip().rstrip("`").strip() plan = json.loads(text) for step in plan.get("steps", []): if step.get("tool") == "generated_code": step["tool"] = "web_search" step["parameters"] = {"query": step.get("description", goal)[:200]} print(f"[Planner] 🔄 Revised plan: {len(plan['steps'])} steps") return plan except Exception as e: print(f"[Planner] ⚠️ Replan failed: {e}") return _fallback_plan(goal)