update tools; add ChecklistTool for task planning and integrate into agent

Author: dcloud347

argus/__init__.py

@@ -11,6 +11,7 @@
 from argus.llm.base import LLMClient
 from argus.tools.ask_web_agent import AskWebAgentTool
 from argus.tools.base import Tool
+from argus.tools.checklist import ChecklistTool
 from argus.tools.shell import ShellTool
 from argus.utils.evermind import EverMindMemory
 from argus.web_agent import WebAgent
@@ -19,6 +20,7 @@
     "Agent",
     "WebAgent",
     "AskWebAgentTool",
+    "ChecklistTool",
     "Content",
     "SystemMessage",
     "UserMessage",

config.example.yaml

@@ -10,6 +10,12 @@ agent:
     After each tool call, wait for the result before deciding the next step.
     Please briefly explain your reasoning before each tool call.
 
+    ## Task Planning (MANDATORY FIRST STEP)
+    Before doing anything else, call the `checklist` tool with action='init' to make a detailed plan listing every step you plan to take.
+    After completing each step, you have to call `checklist` with action='complete' and the step's index to mark this step is done.
+    If you discover new steps mid-task, use action='add'.
+    If you find your plan is not working, you may make a new plan by calling `checklist` with action='init' again.
+
     ## Recommended Workflow Phases (follow in order)
     You have a limited number of steps. Spend them wisely across these phases:
       1. **Explore** (≤12 steps): Understand the codebase and locate the relevant code.

run_swebench_multimodal.py

@@ -22,6 +22,7 @@
 from argus.llm.anthropic import AnthropicClient
 from argus.llm.base import LLMClient
 from argus.llm.openai import OpenAIClient
+from argus.tools.checklist import ChecklistTool
 from argus.tools.shell import ShellTool
 
 logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(name)s: %(message)s")
@@ -146,7 +147,7 @@ def _build_agent(cfg: AgentConfig, instance: dict, port: int, web_agent: WebAgen
 
     agent = Agent(
         llm=llm,
-        tools=[shell, AskWebAgentTool(web_agent)],
+        tools=[ChecklistTool(), shell, AskWebAgentTool(web_agent)],
         system_prompt=cfg.system_prompt,
         max_steps=cfg.max_steps,
         log_dir=log_dir,