MetaGPT writen-report

User description

Features

Combine the open source large language model and the Metagpt framework to achieve intelligent writing. In order to verify the feasibility of the technical route, "accident reports" are used as the experimental subjects, and will be expanded to paper writing or project bidding (document level) writing.（将开源大语言模型和 Metagpt框架相互结合，实现智能写作工作。为验证技术路线的可行性，以“事故报告”做实验对象，后续将拓展到论文写作，或者项目招投标(文档级)写作上)

Feature Docs

This function is document-level writing, which imitates human writing thinking. You write a first draft first, and then submit it to the teacher for evaluation and feedback. Later, based on the feedback, the first draft was revised and improved. (该功能是文档级别写作，模仿人类的写作思维，先写出初稿，后交给老师评估，得到反馈意见。后根据反馈意见，进行对初稿修订改进)

Influence

Liberate the writing ability of public officials. （解放公职人员写作能力） -Implement generation of long document levels（实现长文档级别的生成） Result

Generate papers, bidding documents, official documents, reports, etc.（生成论文，招投标文件，公文，报告等）例如：输入： “5.18”金达花园三期项目坍塌事故调查报告.json

输出： (https://github.com/geekan/MetaGPT/files/15218678/5.18._metagpt.txt)

Other

Type

enhancement

Description

Introduced a comprehensive system for generating, evaluating, and refining accident reports using the MetaGPT framework.
Developed a new RewriteReport class with methods for planning, executing tasks, and handling tools for intelligent report creation.
Added test scripts and actions for evaluating and refining the generated reports.
Updated model configuration to "qwen-14B-Chat" for enhanced performance.
Provided detailed guidance and structured task planning for different sections of an accident report.

Changes walkthrough

Relevant files

Enhancement

core_report.py `Implement RewriteReport Class for Accident Report Generation` tests/write_roport/core_report.py Introduced a new class `RewriteReport` to handle the process of writing, evaluating, and refining accident reports. Implemented methods for planning, acting on tasks, and handling tool recommendations. Added functionality to write initial drafts, evaluate them, and refine based on feedback. Integrated with the MetaGPT framework for intelligent document creation.	+151/-0
write_evaluator_refine.py `Define Actions for Report Writing, Evaluation, and Refinement` tests/write_roport/write_evaluator_refine.py Defined actions for writing analysis, evaluation, and refinement of reports. Utilized structured prompts and evaluation criteria for generating and refining text.	+110/-0
write_report_planner.py `Implement Planner for Structured Report Writing Tasks` tests/write_roport/write_report_planner.py Implemented a planner for report writing tasks based on user requirements and SOP. Added functionality to update and manage plans for report writing.	+102/-0

Tests

test_write_report.py

Test Script for RewriteReport Functionality

tests/write_roport/test_write_report.py

Created a test script to instantiate and run the RewriteReport class
with specified parameters.

+13/-0

Configuration changes

config2.yaml `Update Model Configuration in YAML` config/config2.yaml Updated the model configuration from "gpt-4-turbo" to "qwen-14B-Chat".	+1/-1
sop.json `Structured Order of Tasks for Report Writing in JSON` tests/data/report_writer/sop.json Defined a structured order of tasks for writing an accident report based on user-provided information.	+38/-0

Documentation

domain.yaml

Define Domain-Specific Guidance for Report Writing

tests/data/report_writer/domain.yaml

Added detailed descriptions and guidance for writing various sections
of an accident report.

+98/-0

✨ PR-Agent usage: Comment /help on the PR to get a list of all available PR-Agent tools and their descriptions

Apr 29 '24 03:04 wangwenju269

PR Description updated to latest commit (https://github.com/geekan/MetaGPT/commit/81ab4a974f3ee430c72bd1f8fc637a8f6432ab29)

[ ] Copy walkthrough table to "Files Changed" Tab

Apr 29 '24 03:04 codiumai-pr-agent-pro[bot]

PR Review

(Review updated until commit https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8)

⏱️ Estimated effort to review [1-5]	4, because the PR introduces a complex system involving multiple components such as planning, executing, evaluating, and refining tasks within the MetaGPT framework. The code is extensive and integrates various functionalities, which requires a thorough understanding of the system's architecture and the interactions between components.
🧪 Relevant tests	Yes
🔍 Possible issues	Possible Bug: The method `_check_data` in `RewriteReport` might not handle cases where the `custom` key is not present in `working_memory.index`, potentially leading to unhandled exceptions or logical errors.
🔍 Possible issues	Performance Concern: The `run` method in `RewriteReport` and other asynchronous methods might face performance issues due to multiple await calls and complex logic, especially when handling large data or multiple requests concurrently.
🔒 Security concerns	No

✨ Review tool usage guide:

Overview: The review tool scans the PR code changes, and generates a PR review which includes several types of feedbacks, such as possible PR issues, security threats and relevant test in the PR. More feedbacks can be added by configuring the tool.

The tool can be triggered automatically every time a new PR is opened, or can be invoked manually by commenting on any PR.

When commenting, to edit configurations related to the review tool (pr_reviewer section), use the following template:

/review --pr_reviewer.some_config1=... --pr_reviewer.some_config2=...

With a configuration file, use the following template:

[pr_reviewer]
some_config1=...
some_config2=...

See the review usage page for a comprehensive guide on using this tool.

Apr 29 '24 03:04 codiumai-pr-agent-pro[bot]

PR Code Suggestions

Category	Suggestions
Enhancement	Convert the prompt text into a structured JSON format for better manageability and processing. Consider using a more structured data format like JSON or YAML for the prompts and templates. This would make it easier to manage and modify the content, as well as potentially simplify the code that processes these templates. user_develop/prompt/write_report_paragraph.py [2-7] -FIRST_PROMPT = """ -请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。在段落的结尾处，加入一段官方声明，声明应体现事故调查的严谨性和全面性。 -* 注意事项 * - 1. 在段落的结尾，请务必补充一段官方声明。eg：<事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下> - 2. 事件描述简短扼要。 -请确保您的文本不仅简要记录了事故概要，而且语言表达准确、连贯，体现出调查报告的权威性和专业性。字数要求 300 - 500 之间。 -""" +FIRST_PROMPT = { + "description": "请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。", + "requirements": [ + "在段落的结尾，请务必补充一段官方声明。eg：<事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下>", + "事件描述简短扼要。" + ], + "note": "请确保您的文本不仅简要记录了事故概要，而且语言表达准确、连贯，体现出调查报告的权威性和专业性。字数要求 300 - 500 之间。" +}
Enhancement	Simplify the conversion of `messages` to a list using Python's built-in `list()` function. Instead of manually checking and converting the type of `messages` into a list, use Python's built-in `list()` function which can handle various iterable types and simplifies the code. user_develop/core_report.py [139-140] `-if not isinstance(messages, list): - messages = [messages] +messages = list(messages) if not isinstance(messages, list) else messages`
Performance	Optimize string concatenation by using `str.join()` directly with a generator expression. Replace the repeated string concatenation with a more efficient method such as using `str.join()` which is generally faster and more readable when concatenating multiple strings. user_develop/action/evaluator_report.py [79-82] `-info = ( - "\n".join([str(ct) for ct in working_memory if str(ct).startswith("user")]) - if working_memory - else "" -) +info = "\n".join(str(ct) for ct in working_memory if str(ct).startswith("user")) if working_memory else ""`
Performance	Use asynchronous file operations to prevent blocking in async functions. Replace the synchronous file reading and JSON operations with asynchronous equivalents to avoid blocking the event loop in an async function. user_develop/plan/report_planner.py [50-51] `-with open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: - rsp = json.dumps(json.loads(file.read()), ensure_ascii=False) +import aiofiles +import orjson +async with aiofiles.open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: + data = await file.read() + rsp = orjson.dumps(orjson.loads(data), ensure_ascii=False).decode()`
Best practice	Use a context manager for file operations to ensure proper resource management. Use a context manager when opening files to ensure that the file is properly closed after its suite finishes, even if an exception is raised. user_develop/core_report.py [126-127] `with open(file_path, "r", encoding="utf-8") as file: - data_info = json.loads(file.read()) + data_info = json.load(file)`
	Improve file path handling for cross-platform compatibility. Replace the direct file path concatenation with `os.path.join` for better compatibility across different operating systems. user_develop/plan/report_planner.py [50] `-with open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: +with open(os.path.join(DATA_PATH, "sop.json"), "r", encoding="utf-8") as file:`
	Ensure files are properly closed after operations. Use a context manager when opening files to ensure that the file is properly closed after its suite finishes. user_develop/plan/report_planner.py [51] `-rsp = json.dumps(json.loads(file.read()), ensure_ascii=False) +with open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: + rsp = json.dumps(json.loads(file.read()), ensure_ascii=False)`
Maintainability	Refactor the message type handling using a dictionary mapping to improve scalability and manageability. Instead of using multiple if-else conditions to handle different types of messages, consider using a dictionary mapping to streamline the type handling and make the code more scalable and easier to manage. user_develop/core_report.py [145-153] -if isinstance(msg, str): - processed_messages.append({"role": "user", "content": msg}) -elif isinstance(msg, dict): - assert set(msg.keys()) == set(["role", "content"]) - processed_messages.append(msg) -elif isinstance(msg, Message): - processed_messages.append(msg.to_dict()) -else: - raise ValueError( - f"Only support message type are: str, Message, dict, but got {type(messages).__name__}!" - ) +message_type_handlers = { + str: lambda m: {"role": "user", "content": m}, + dict: lambda m: m, + Message: lambda m: m.to_dict() +} +try: + processed_messages.append(message_type_handlers[type(msg)](msg)) +except KeyError: + raise ValueError(f"Unsupported message type: {type(msg).__name__}")
Maintainability	Refactor method to improve readability and separation of concerns. Refactor the loop in `update_plan` to separate concerns and improve readability by extracting methods. user_develop/plan/report_planner.py [59-75] while not plan_confirmed: context = self.get_useful_memories() rsp = await WriteReportPlan().run(context, max_tasks, self.human_design_sop) + if not self.process_response(rsp): + continue + _, plan_confirmed = await self.ask_review(trigger=ReviewConst.TASK_REVIEW_TRIGGER) + +def process_response(self, rsp): self.working_memory.add(Message(content=rsp, role="assistant", cause_by=WritePlan)) is_plan_valid, error = precheck_update_plan_from_rsp(rsp, self.plan) - if not is_plan_valid and max_retries > 0: + if not is_plan_valid and self.max_retries > 0: error_msg = f"The generated plan is not valid with error: {error}, try regenerating, remember to generate either the whole plan or the single changed task only" logger.warning(error_msg) self.working_memory.add(Message(content=error_msg, role="assistant", cause_by=WritePlan)) - max_retries -= 1 - continue - _, plan_confirmed = await self.ask_review(trigger=ReviewConst.TASK_REVIEW_TRIGGER) + self.max_retries -= 1 + return False + return True
Error handling	Add error handling for file operations to enhance robustness. Handle potential exceptions that may occur during file operations to prevent the program from crashing. user_develop/plan/report_planner.py [50-51] `-with open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: - rsp = json.dumps(json.loads(file.read()), ensure_ascii=False) +try: + with open(DATA_PATH / "sop.json", "r", encoding="utf-8") as file: + rsp = json.dumps(json.loads(file.read()), ensure_ascii=False) +except IOError as e: + logger.error(f"Failed to read file: {e}") + return ""`

✨ Improve tool usage guide:

Overview: The improve tool scans the PR code changes, and automatically generates suggestions for improving the PR code. The tool can be triggered automatically every time a new PR is opened, or can be invoked manually by commenting on a PR.

When commenting, to edit configurations related to the improve tool (pr_code_suggestions section), use the following template:

/improve --pr_code_suggestions.some_config1=... --pr_code_suggestions.some_config2=...

With a configuration file, use the following template:

[pr_code_suggestions]
some_config1=...
some_config2=...

See the improve usage page for a comprehensive guide on using this tool.

Apr 29 '24 03:04 codiumai-pr-agent-pro[bot]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 70.19%. Comparing base (f201b2f) to head (c4f6370). Report is 110 commits behind head on main.

:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1235      +/-   ##
==========================================
- Coverage   70.22%   70.19%   -0.04%     
==========================================
  Files         316      316              
  Lines       18860    18860              
==========================================
- Hits        13245    13239       -6     
- Misses       5615     5621       +6

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

Apr 29 '24 04:04 codecov-commenter

按照返回意见，重新修改代码，清晰。对原 Metagpt 结构未作任何改变，复用源基层代码，模仿datainterpreter 类进行改造，实现与源代码高耦合。

Apr 30 '24 03:04 wangwenju269

Great! Let's review

Apr 30 '24 03:04 geekan

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Description updated to latest commit (https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8)

[ ] Copy walkthrough table to "Files Changed" Tab

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

Persistent review updated to latest commit https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Description updated to latest commit (https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8)

[ ] Copy walkthrough table to "Files Changed" Tab

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

Persistent review updated to latest commit https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Code Suggestions

Category	Suggestions
Bug	Initialize `tool_recommender` with a default instance to prevent potential AttributeError. The `tool_recommender` attribute is initialized to `None` and later conditionally assigned to `BM25ToolRecommender` based on the `tools` attribute. This could lead to a scenario where `tool_recommender` remains `None` and is used before assignment, causing an AttributeError. To avoid this, initialize `tool_recommender` with a default instance of `ToolRecommender`. tests/write_roport/core_report.py [29] `-tool_recommender: ToolRecommender = None +tool_recommender: ToolRecommender = ToolRecommender()`
Enhancement	Provide a meaningful default value for `upload_file` to prevent file path errors. The `upload_file` attribute is used to determine the file path for reading and writing operations but is initialized as an empty string, which could lead to errors if not properly set before use. Ensure that `upload_file` is validated or has a meaningful default value that points to a valid file path. tests/write_roport/core_report.py [34] `-upload_file: str = "" +upload_file: str = "/default/path/to/file.txt" # Update this path as necessary`
Enhancement	Handle the result of the superclass `run` method to ensure proper flow and data handling. The `run` method in `RewriteReport` class uses `await super().run(with_message)` without returning its result. If the superclass `run` method returns a value that should be used or checked, ensure to handle or return it appropriately. tests/write_roport/core_report.py [60] `-await super().run(with_message) +result = await super().run(with_message) +return result # Ensure to handle or return the result appropriately`
Possible issue	Add a break condition to prevent potential infinite loops in retry logic. The method `_ready_write_report` uses a while loop for retries which could potentially lead to an infinite loop if `success` is never set to `True`. Implement a maximum retry limit to prevent infinite loops. tests/write_roport/core_report.py [91] `while not success and counter < max_retry: + if counter == max_retry - 1: + break # Prevent infinite loop by breaking out after max retries`
Possible issue	Add a check to ensure `working_memory.index` exists to prevent AttributeError. The method `_check_data` checks if "custom" is in `working_memory.index` but does not handle the case where `working_memory.index` might not exist or be accessible, which could raise an AttributeError. Ensure that `working_memory.index` is initialized and accessible before checking its contents. tests/write_roport/core_report.py [147] `-if "custom" not in self.working_memory.index: +if hasattr(self.working_memory, 'index') and "custom" not in self.working_memory.index:`
Maintainability	Improve YAML file consistency by using uniform indentation. Consider using consistent indentation for YAML files to avoid parsing errors. The indentation for the `name`, `desc`, and `guidance` fields under `first_paragraph` is inconsistent with the rest of the document. tests/data/report_writer/domain.yaml [2-10] first_paragraph: - name: first_paragraph - desc: 请帮我依据用户所提供事故要点信息进行简要扩写, 完整事故报告的第一自然段。 - guidance: \| - 请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。在段落的结尾处，加入一段官方声明，声明应体现事故调查的严谨性和全面性。 + name: first_paragraph + desc: 请帮我依据用户所提供事故要点信息进行简要扩写, 完整事故报告的第一自然段。 + guidance: \| + 请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。在段落的结尾处，加入一段官方声明，声明应体现事故调查的严谨性和全面性。
Best practice	Replace special formatting in guidance text with standard YAML comments or fields. Ensure that the guidance text does not contain unnecessary special characters or formatting issues that might confuse the users or parsers. The use of * 注意事项 * could be replaced with a more standard YAML comment or a separate field if needed. tests/data/report_writer/domain.yaml [7-9] `guidance: \| - * 注意事项 * - 1. 在段落的结尾,请务必补充一段官方声明。eg:<事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下> + # 注意事项 + - 在段落的结尾,请务必补充一段官方声明。例如: "事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下"`

✨ Improve tool usage guide:

Overview: The improve tool scans the PR code changes, and automatically generates suggestions for improving the PR code. The tool can be triggered automatically every time a new PR is opened, or can be invoked manually by commenting on a PR.

When commenting, to edit configurations related to the improve tool (pr_code_suggestions section), use the following template:

/improve --pr_code_suggestions.some_config1=... --pr_code_suggestions.some_config2=...

With a configuration file, use the following template:

[pr_code_suggestions]
some_config1=...
some_config2=...

See the improve usage page for a comprehensive guide on using this tool.

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

PR-Agent was enabled for this repository. To continue using it, please link your git user with your CodiumAI identity here.

PR Code Suggestions

Category	Suggestions
Enhancement	Improve the method signature by removing the default value for `upload_file` to prevent potential misuse. The `RewriteReport` class has a method `run` that uses the `upload_file` parameter to set the instance variable `self.upload_file`. However, this parameter is also provided as a default argument in the method signature, which can lead to confusion and potential bugs if the method is called without specifying the `upload_file` parameter. To improve clarity and avoid potential issues, remove the default value from the method signature and handle the case where `upload_file` is not provided within the method. tests/write_roport/core_report.py [58-59] `-async def run(self, with_message=None, upload_file="") -> Message \| None: +async def run(self, with_message=None, upload_file: str) -> Message \| None: + if not upload_file: + raise ValueError("upload_file must be provided") self.upload_file = upload_file await super().run(with_message)`
Performance	Add a break statement to exit the loop early once a successful evaluation is achieved. The `RewriteReport` class uses a method `_ready_write_report` that has a loop for retrying the evaluation of a report. The loop uses a counter to limit retries, but it does not break out of the loop once a successful evaluation is achieved. This can lead to unnecessary iterations even after a successful evaluation. To optimize performance and avoid unnecessary operations, add a break statement once a successful evaluation is achieved. tests/write_roport/core_report.py [91-97] `while not success and counter < max_retry: suggestion, success = await self.evaluator.run(working_memory=self.working_memory.get()) if not success: report = await self.refine.run(suggestion=suggestion, working_memory=self.working_memory.get()) self.working_memory.delete(self.working_memory.get()[-1]) self.working_memory.add(Message(content=report, role="assistant", cause_by=RefineReport)) + else: + break counter += 1`
Possible issue	Add checks to ensure `self.rc` and `self.rc.todo` are initialized before accessing them. The `RewriteReport` class method `_write_report` uses a local variable `todo` which is assigned from `self.rc.todo`, assuming it always exists. This can lead to AttributeError if `self.rc` or `self.rc.todo` is not properly initialized. To make the code more robust, add a check to ensure that `self.rc` and `self.rc.todo` are not None before accessing them. tests/write_roport/core_report.py [107-108] `+if not hasattr(self, 'rc') or not hasattr(self.rc, 'todo'): + raise AttributeError("Expected 'rc' and 'rc.todo' to be initialized") todo = self.rc.todo # todo is WriteAnalysisCode logger.info(f"ready to {todo.name}")`
	Add a check to ensure `self.working_memory.index` is not None before accessing it. The method `_check_data` in the `RewriteReport` class checks if "custom" is in `self.working_memory.index` but does not handle the case where `self.working_memory.index` might be uninitialized or None. This can lead to an AttributeError. To prevent this, add a check to ensure `self.working_memory.index` is not None before checking for "custom" in it. tests/write_roport/core_report.py [147-149] `-if "custom" not in self.working_memory.index: +if self.working_memory.index is None or "custom" not in self.working_memory.index: logger.info("add user additional data into working_memory") self.read_data_info()`
	Add a check to ensure `self.planner` is initialized before accessing its methods. The method `_ready_write_report` in the `RewriteReport` class uses a conditional to check if `self.use_plan` is True to get the plan status. However, this method does not handle the case where `self.planner` might be None, which can lead to AttributeError. To make the code more robust, add a check to ensure that `self.planner` is not None before accessing its methods. tests/write_roport/core_report.py [74] `-plan_status = self.planner.get_plan_status() if self.use_plan else "" +if self.use_plan: + if self.planner is None: + raise AttributeError("Planner is not initialized") + plan_status = self.planner.get_plan_status() +else: + plan_status = ""`
Maintainability	Standardize YAML indentation for better readability and to avoid parsing issues. Consider using a consistent indentation style for YAML files to improve readability and prevent parsing errors. The indentation levels in the `first_paragraph` and `second_paragraph` sections are inconsistent. tests/data/report_writer/domain.yaml [2-10] first_paragraph: - name: first_paragraph - desc: 请帮我依据用户所提供事故要点信息进行简要扩写, 完整事故报告的第一自然段。 - guidance: \| - 请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。在段落的结尾处，加入一段官方声明，声明应体现事故调查的严谨性和全面性。 - * 注意事项 * - 1. 在段落的结尾,请务必补充一段官方声明。eg:<事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下> - 2. 事件描述简短扼要。 - 请确保您的文本不仅简要记录了事故概要，而且语言表达准确、连贯，体现出调查报告的权威性和专业性。字数要求 300 - 500 之间。 + name: first_paragraph + desc: 请帮我依据用户所提供事故要点信息进行简要扩写, 完整事故报告的第一自然段。 + guidance: \| + 请根据以下关键信息撰写一个合理的事故报告开头的第一自然段,主要介绍事故概要。在段落的结尾处，加入一段官方声明，声明应体现事故调查的严谨性和全面性。 + * 注意事项 * + 1. 在段落的结尾,请务必补充一段官方声明。eg:<事故调查组按照“实事求是、客观公正、尊重科学”的原则，..., ... ,现将有关情况报告如下> + 2. 事件描述简短扼要。 + 请确保您的文本不仅简要记录了事故概要，而且语言表达准确、连贯，体现出调查报告的权威性和专业性。字数要求 300 - 500 之间。

✨ Improve tool usage guide:

Overview: The improve tool scans the PR code changes, and automatically generates suggestions for improving the PR code. The tool can be triggered automatically every time a new PR is opened, or can be invoked manually by commenting on a PR.

When commenting, to edit configurations related to the improve tool (pr_code_suggestions section), use the following template:

/improve --pr_code_suggestions.some_config1=... --pr_code_suggestions.some_config2=...

With a configuration file, use the following template:

[pr_code_suggestions]
some_config1=...
some_config2=...

See the improve usage page for a comprehensive guide on using this tool.

Apr 30 '24 03:04 codiumai-pr-agent-pro[bot]

I developed an official document writing artifact, which can be expanded to the fields of paper writing, official document generation, bidding documents and other fields according to user needs. In the future, tools such as networking and literature retrieval will continue to be integrated. The code repository is at: METAGPT\tests\metagpt\ext\write_report\report.py (我开发一款公文写作神器，根据用户的需求可以拓展到论文写作，公文生成，招投标文档等领域。后续会继续集成联网，文献检索等工具。代码仓库在： METAGPT\tests\metagpt\ext\write_report\report.py)

May 06 '24 06:05 wangwenju269

@wangwenju269 哈喽，有一些问题先请问下

为什么代码是写在tests目录下？
我看大部分的提示词都是中文，是必须使用中文提示词，还是什么原因呢？

May 17 '24 03:05 shenchucheng

您好：

答复如下

   问题1 ： 放在 test 路径下的目的：自己开发的项目代码脚本 和 源metagpt框架 能更好解耦，本项目是复用 metagpt 现有的类和方法，代码做到高内聚，低耦合。

   问题2： 提示词是可以拓展的，已将提示词单独放在一个文件里。后续更换提示词文件，实现实现招投标、论文写作等应用场景。 提示词是可插拔替换的，根据具体业务场景，更换提示词

祝好！

在 2024-05-17 11:06:18，"shenchucheng" @.***> 写道：

@wangwenju269 哈喽，有一些问题先请问下

为什么代码是写在tests目录下？我看大部分的提示词都是中文，是必须使用中文提示词，还是什么原因呢？

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

May 17 '24 03:05 wangwenju269

有修改意见，请及时反馈给我，谢谢

May 17 '24 04:05 wangwenju269

您好：答复如下问题1 ：放在 test 路径下的目的：自己开发的项目代码脚本和源metagpt框架能更好解耦，本项目是复用 metagpt 现有的类和方法，代码做到高内聚，低耦合。问题2：提示词是可以拓展的，已将提示词单独放在一个文件里。后续更换提示词文件，实现实现招投标、论文写作等应用场景。提示词是可插拔替换的，根据具体业务场景，更换提示词祝好！在 2024-05-17 11:06:18，"shenchucheng" @.> 写道： @wangwenju269 哈喽，有一些问题先请问下为什么代码是写在tests目录下？我看大部分的提示词都是中文，是必须使用中文提示词，还是什么原因呢？ — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.>

metagpt下有ext文件夹（external, extension之意），可将代码放入metagpt/ext下，参考stanford_town的放置方法，可与metagpt核心代码相分开

May 17 '24 09:05 garylin2099

好的，收到，我尽快修改

---- 回复的原邮件 ---- | 发件人 | @.> | | 日期 | 2024年05月17日 17:32 | | 收件人 | @.> | | 抄送至 | @.>@.> | | 主题 | Re: [geekan/MetaGPT] writen-report (PR #1235) |

@garylin2099 commented on this pull request.

代码虽然继承了MG内DataInterpreter, WritePlan, Planner等类，但仍不免需要改写大量函数，子类函数与父类函数有较多重复。这并非此PR的问题，是原来DI及DI相关类未留出足够的可配置项。若从此PR出发，可以再仔细考虑，哪些地方是可以直接使用原DI代码，哪些是必须要自定义，尽量减少些重复

In tests/metagpt/ext/write_report/core_report.py:

+from pydantic import Field, model_validator + +from metagpt.logs import logger +from metagpt.roles import Role +from metagpt.schema import Message, Task, TaskResult +from metagpt.tools.tool_recommend import BM25ToolRecommender, ToolRecommender +from tests.metagpt.ext.write_report.write_evaluator_refine import (

EvaluatorReport,
RefineReport,
WriteAnalysisReport, +) +from tests.metagpt.ext.write_report.write_report_planner import WritePlanner

+class RewriteReport(Role):

应为ReportRewriter，RewriteReport是动作的意思

In config/config2.yaml:

\ No newline at end of file

model: "qwen-14B-Chat" # or gpt-3.5-turbo

这个改动不应该提交

In tests/metagpt/ext/write_report/report.py:

+# -- coding: utf-8 -- +""" @.*** : 2024/5/06 13:54 @.*** : wangwenju269 @.*** : report.py +"""

+import fire + +from tests.metagpt.ext.write_report.core_report import RewriteReport + + +async def main(auto_run: bool = True):

requirement = "写一份事故报告"
di = RewriteReport(auto_run=auto_run, human_design_sop=True, use_evaluator=True)
await di.run(f"{requirement}", upload_file="E:\workspace\MetaGPT\data\“5.18”金达花园三期项目坍塌事故调查报告.json")

建议去掉具体的路径名和文件名

In tests/metagpt/ext/write_report/write_evaluator_refine.py:

       "reason": str = "评分的依据和解释",

       "critique": str = "提供针对生成文本的具体改进建议"

```
   }},
```
```
   ...
```
]

+""" + +REFINE = """请根据 user 的具体任务要求（Plan Status）以及以下的反馈建议，对 assistant 之前的回应进行深入审查，并在需要的地方进行适当的改进。 +用户反馈建议：{suggestion} + +在修订时，请保持与 assistant 原始回应的格式和风格一致，并确保输出的内容充分体现了用户的具体任务要求。 +""" + + +def process_message(messages: Union[str, Message, list[dict], list[Message], list[str]]) -> list[dict]:

可使用llm.format_msg，见具体provider文件

In tests/metagpt/ext/write_report/write_evaluator_refine.py:

@@ -0,0 +1,170 @@ +import json +from typing import Union

+from metagpt.actions import Action +from metagpt.schema import Message +from metagpt.utils.common import CodeParser + +STRUCTUAL_PROMPT = """

可跟write_analysis_code.py保持一致，将prompt放入单独文件

In tests/metagpt/ext/write_report/write_report_planner.py:

       self.PROMPT_TEMPLATE = f"{self.PROMPT_TEMPLATE}\n\n{self.CONSTRAINTS}"

       task_type_desc = "\n".join([f"- **{v['name']}**: {v['desc']}" for k, v in usertasktype.items()])

       prompt = self.PROMPT_TEMPLATE.format(

           context="\n".join([str(ct) for ct in context]), max_tasks=max_tasks, task_type_desc=task_type_desc

```
       )
```
```
       rsp = await self._aask(prompt)
```

       rsp = CodeParser.parse_code(block=None, text=rsp)

可改写PROMPT_TEMPLATE，这段使用super().run

In tests/metagpt/ext/write_report/write_report_planner.py:

       rsp = await WriteReportPlan().run(context, max_tasks, self.human_design_sop)

       self.working_memory.add(Message(content=rsp, role="assistant", cause_by=WritePlan))

       # precheck plan before asking reviews

       is_plan_valid, error = precheck_update_plan_from_rsp(rsp, self.plan)

       if not is_plan_valid and max_retries > 0:

           error_msg = f"The generated plan is not valid with error: {error}, try regenerating, remember to generate either the whole plan or the single changed task only"

```
           logger.warning(error_msg)
```

           self.working_memory.add(Message(content=error_msg, role="assistant", cause_by=WritePlan))

```
           max_retries -= 1
```
```
           continue
```

       _, plan_confirmed = await self.ask_review(trigger=ReviewConst.TASK_REVIEW_TRIGGER)

   update_plan_from_rsp(rsp=rsp, current_plan=self.plan)

```
   # working_memory
```
```
   self.working_memory.clear()
```
def get_plan_status(self) -> str:

这个函数是否需要改写？PLAN_STATUS待填入的变量名可以和原PLAN_STATUS保持一致

In tests/metagpt/ext/write_report/write_report_planner.py:

async def run(self, context: list[Message], max_tasks: int = 7, human_design_sop=True) -> str:

```
   if not human_design_sop:
```

       self.PROMPT_TEMPLATE = f"{self.PROMPT_TEMPLATE}\n\n{self.CONSTRAINTS}"

       task_type_desc = "\n".join([f"- **{v['name']}**: {v['desc']}" for k, v in usertasktype.items()])

       prompt = self.PROMPT_TEMPLATE.format(

           context="\n".join([str(ct) for ct in context]), max_tasks=max_tasks, task_type_desc=task_type_desc

```
       )
```
```
       rsp = await self._aask(prompt)
```

       rsp = CodeParser.parse_code(block=None, text=rsp)

```
   else:
```

       with open(TEST_DATA_PATH / "report_writer/sop.json", "r", encoding="utf-8") as file:

           rsp = json.dumps(json.loads(file.read()), ensure_ascii=False)  # 读取人类设计 `sop` 流程

```
   return rsp
```

+class WritePlanner(Planner):

WriteReportPlanner

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>

May 23 '24 13:05 wangwenju269

答复：文件放置在 test 目录下，方便更好测试代码是否独立执行成功。我已遵循评审大佬的要求和反馈建议，做出相应修改，清再次查收。祝好！

May 24 '24 08:05 wangwenju269

tests

答复：文件放置在 test 目录下，方便更好测试代码是否独立执行成功。我已遵循评审大佬的要求和反馈建议，做出相应修改，清再次查收。祝好！

tests目录是单测目录，tests/metagpt/ext/是metagpt/ext下的代码对应的单测文件

May 29 '24 08:05 shenchucheng

MetaGPT MetaGPT copied to clipboard

writen-report

User description

Type

Description

Changes walkthrough

PR Review

(Review updated until commit https://github.com/geekan/MetaGPT/commit/ad8eeb7460e60392d381a1171900c347eaf8a8c8)

PR Code Suggestions

Codecov Report

PR Code Suggestions

PR Code Suggestions

MetaGPT
MetaGPT copied to clipboard