Merge branch 'mgx_ops' into json_repair

# Conflicts:
#	metagpt/prompts/di/role_zero.py
#	metagpt/roles/di/data_analyst.py

metagpt/roles/architect.py

@@ -8,6 +8,8 @@
 from metagpt.actions import WritePRD
 from metagpt.actions.design_api import WriteDesign
 from metagpt.roles.di.role_zero import RoleZero
+from metagpt.tools.libs.software_development import write_trd_and_framework
+from metagpt.utils.common import tool2name
 
 
 class Architect(RoleZero):
@@ -29,9 +31,14 @@ class Architect(RoleZero):
         "libraries. Use same language as user requirement"
     )
 
-    instruction: str = """Use WriteDesign tool to write a system design document"""
+    instruction: str = """Use WriteDesign tool to write a system design document if a system design is required; Use `write_trd_and_framework` tool to write a software framework if a software framework is required;"""
     max_react_loop: int = 1  # FIXME: Read and edit files requires more steps, consider later
-    tools: list[str] = ["Editor:write,read,write_content", "RoleZero", "WriteDesign"]
+    tools: list[str] = [
+        "Editor:write,read,write_content",
+        "RoleZero",
+        "WriteDesign",
+        write_trd_and_framework.__name__,
+    ]
 
     def __init__(self, **kwargs) -> None:
         super().__init__(**kwargs)
@@ -45,11 +52,11 @@ class Architect(RoleZero):
         self._watch({WritePRD})
 
     def _update_tool_execution(self):
-        wd = WriteDesign()
+        write_design = WriteDesign()
+        self.tool_execution_map.update(tool2name(WriteDesign, ["run"], write_design.run))
         self.tool_execution_map.update(
             {
-                "WriteDesign.run": wd.run,
-                "WriteDesign": wd.run,  # alias
-                "run": wd.run,  # alias
+                write_trd_and_framework.__name__: write_trd_and_framework,
+                "run": write_design.run,  # alias
             }
         )

metagpt/roles/di/engineer2.py

@@ -12,7 +12,7 @@ class Engineer2(RoleZero):
     goal: str = "Take on game, app, and web development"
     instruction: str = ENGINEER2_INSTRUCTION
 
-    tools: str = ["Plan", "Editor:write,read", "RoleZero", "ReviewAndRewriteCode"]
+    tools: list[str] = ["Plan", "Editor:write,read", "RoleZero", "ReviewAndRewriteCode"]
 
     def _update_tool_execution(self):
         review = ReviewAndRewriteCode()

metagpt/roles/di/role_zero.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import inspect
 import json
+import re
 import traceback
 from typing import Callable, Literal, Tuple
 
@@ -10,7 +11,11 @@ from pydantic import model_validator
 from metagpt.actions import Action
 from metagpt.actions.di.run_command import RunCommand
 from metagpt.logs import logger
-from metagpt.prompts.di.role_zero import CMD_PROMPT, ROLE_INSTRUCTION, JSON_REPAIR_PROMPT
+from metagpt.prompts.di.role_zero import (
+    CMD_PROMPT,
+    JSON_REPAIR_PROMPT,
+    ROLE_INSTRUCTION,
+)
 from metagpt.roles import Role
 from metagpt.schema import AIMessage, Message, UserMessage
 from metagpt.strategy.experience_retriever import DummyExpRetriever, ExpRetriever
@@ -20,8 +25,8 @@ from metagpt.tools.libs.editor import Editor
 from metagpt.tools.tool_recommend import BM25ToolRecommender, ToolRecommender
 from metagpt.tools.tool_registry import register_tool
 from metagpt.utils.common import CodeParser
+from metagpt.utils.repair_llm_raw_output import RepairType, repair_llm_raw_output
 from metagpt.utils.report import ThoughtReporter
-from metagpt.utils.repair_llm_raw_output import repair_llm_raw_output, RepairType
 
 
 @register_tool(include_functions=["ask_human", "reply_to_human"])
@@ -88,6 +93,23 @@ class RoleZero(Role):
             "RoleZero.ask_human": self.ask_human,
             "RoleZero.reply_to_human": self.reply_to_human,
         }
+        self.tool_execution_map.update(
+            {
+                f"Browser.{i}": getattr(self.browser, i)
+                for i in [
+                    "click",
+                    "close_tab",
+                    "go_back",
+                    "go_forward",
+                    "goto",
+                    "hover",
+                    "press",
+                    "scroll",
+                    "tab_focus",
+                    "type",
+                ]
+            }
+        )
         # can be updated by subclass
         self._update_tool_execution()
         return self
@@ -126,7 +148,14 @@ class RoleZero(Role):
             available_commands=tool_info,
             instruction=self.instruction.strip(),
         )
-        context = self.llm.format_msg(self.rc.memory.get(self.memory_k) + [UserMessage(content=prompt)])
+        memory = self.rc.memory.get(self.memory_k)
+        if not self.browser.is_empty_page:
+            pattern = re.compile(r"Command Browser\.(\w+) executed")
+            for index, msg in zip(range(len(memory), 0, -1), memory[::-1]):
+                if pattern.match(msg.content):
+                    memory.insert(index, UserMessage(cause_by="browser", content=await self.browser.view()))
+                    break
+        context = self.llm.format_msg(memory + [UserMessage(content=prompt)])
         # print(*context, sep="\n" + "*" * 5 + "\n")
         async with ThoughtReporter(enable_llm_stream=True):
             self.command_rsp = await self.llm.aask(context, system_msgs=self.system_msg)
@@ -141,7 +170,7 @@ class RoleZero(Role):
         try:
             commands = CodeParser.parse_code(block=None, lang="json", text=self.command_rsp)
             commands = json.loads(repair_llm_raw_output(output=commands, req_keys=[None], repair_type=RepairType.JSON))
-        except json.JSONDecodeError as e:
+        except json.JSONDecodeError:
             commands = await self.llm.aask(msg=JSON_REPAIR_PROMPT.format(json_data=self.command_rsp))
             commands = json.loads(CodeParser.parse_code(block=None, lang="json", text=commands))
         except Exception as e:

metagpt/roles/product_manager.py

@@ -9,9 +9,10 @@
 
 from metagpt.actions import UserRequirement, WritePRD
 from metagpt.actions.prepare_documents import PrepareDocuments
+from metagpt.actions.requirement_analysis.requirement.pic2txt import Pic2Txt
 from metagpt.roles.di.role_zero import RoleZero
 from metagpt.roles.role import RoleReactMode
-from metagpt.utils.common import any_to_name, any_to_str
+from metagpt.utils.common import any_to_name, any_to_str, tool2name
 from metagpt.utils.git_repository import GitRepository
 
 
@@ -32,9 +33,9 @@ class ProductManager(RoleZero):
     constraints: str = "utilize the same language as the user requirements for seamless communication"
     todo_action: str = any_to_name(WritePRD)
 
-    instruction: str = """Use WritePRD tool to write PRD"""
+    instruction: str = """Use WritePRD tool to write PRD if a PRD is required; Use `Pic2Txt` tool to write out an intact textual user requirements if an intact textual user requiremnt is required given some images alongside the contextual textual descriptions;"""
     max_react_loop: int = 1  # FIXME: Read and edit files requires more steps, consider later
-    tools: list[str] = ["Editor:write,read,write_content", "RoleZero", "WritePRD"]
+    tools: list[str] = ["Editor:write,read,write_content", "RoleZero", "WritePRD", Pic2Txt.__name__]
 
     def __init__(self, **kwargs) -> None:
         super().__init__(**kwargs)
@@ -47,12 +48,9 @@ class ProductManager(RoleZero):
 
     def _update_tool_execution(self):
         wp = WritePRD()
-        self.tool_execution_map.update(
-            {
-                "WritePRD.run": wp.run,
-                "WritePRD": wp.run,  # alias
-            }
-        )
+        self.tool_execution_map.update(tool2name(WritePRD, ["run"], wp.run))
+        pic2txt = Pic2Txt()
+        self.tool_execution_map.update(tool2name(Pic2Txt, ["run"], pic2txt.run))
 
     async def _think(self) -> bool:
         """Decide what to do"""