Star type dynamic coding Agent system and method for solving complex task
By adopting star topology and dynamic tool generation methods in the Agent system, the problem of single Agent handling complex tasks and tool fixation is solved, and more efficient task completion and tool utilization is achieved.
Patent Information
- Application Number
- CN202411784335.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-06
- Publication Date
- 2025-04-22
AI Technical Summary
A single agent performs poorly when handling complex tasks, and the tools are fixed and difficult to implement, resulting in the inability to effectively complete complex tasks.
Agents are organized using a star topology, split and schedule tasks through the central agent, and dynamically generate tools by encoding agents to complete tasks.
It effectively solves the problem of a single agent handling complex tasks and tool fixation, and improves the performance of the agent in complex tasks and the reusability of tools.
Smart Images
Figure CN119862028A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to a star-shaped dynamic coding Agent system and method for solving complex tasks, belonging to the field of software technology. Background Art
[0002] The emergence of the Large Language Model (LLM) has undoubtedly made many ideas about realizing AI Agents possible. In the context of the gradual development and maturity of LLM, the implementation of LLM-based applications is an important link. As we all know, to realize an agent, it is mainly to give LLM more capabilities, including: thinking, memory, perception, and action. LLM is used as the brain of the agent to think and plan, the agent is given memory through the database, and the agent is given the ability to perceive and act by providing tools. This model is usually completed by only one LLM, like realizing an independent individual, then the disadvantages of being an independent individual appear.
[0003] First, LLM usually has a system prompt word to define the role, scenario, and tone of the LLM in the current context window. This means that the conversation content in the same context dialog window is dedicated and special. In order to allow LLM to focus on handling problems in one scenario and get better answers, this may cause a single agent to perform poorly when handling complex tasks. Complex tasks usually consist of tasks in multiple different scenarios. When completing one of the tasks, the context content of other tasks is noise for the task, which affects the attention mechanism of LLM and makes it unable to focus on completing any one task.
[0004] Secondly, first, because there are too many types of tools mixed together and LLM is allowed to decide on their own, this undoubtedly increases the difficulty for LLM to use the tools accurately; second, how to make LLM understand the role of the tools and when to use them through descriptions of the tools is also a difficulty; third, tools are usually provided by people, so if tools are to be provided, they must be implemented in advance, and LLM can only use the provided tools. The tools are relatively fixed, and in complex task scenarios, people certainly cannot implement tools that cover all scenarios, which is even more difficult.
[0005] At present, in addition to independent single agents, multiple single agents are also communicated and collaborated to form an agent community. This can indeed allow each agent to focus on processing a type of task, but each agent needs to understand the capabilities of all other agents in the community in order to communicate accurately, and the mesh communication topology is relatively chaotic. Summary of the invention
[0006] In view of at least one of the above technical problems, the present application provides a star-shaped dynamic coding agent system and method for solving complex tasks. By adopting a star topology structure to organize agents, the problem of poor performance of a single agent when handling complex tasks and mixed use of tools is solved; by dynamically generating tools through coding agents, the problem of fixed tools and difficult implementation is solved.
[0007] Technical solution: To solve the above technical problems, the technical solution adopted in this application is:
[0008] In a first aspect, the present application provides a star-shaped dynamic encoding Agent system for solving complex tasks, the system comprising a central Agent and multiple encoding Agents;
[0009] The central agent is used to: receive complex tasks input by users; split and refine the complex tasks according to the task types after thinking, organize the scheduling order of the distributed tasks, and output the task list; distribute the tasks to the corresponding encoding agents according to the scheduling order in the task list and the task types; receive the task completion status returned by the encoding agent, further think about it according to the task completion status, adjust the subsequent tasks, and then issue the next task until all tasks in the task list are completed;
[0010] The encoding agent is used to: receive tasks distributed by the central agent; generate reusable tools through encoding according to task requirements and add them to the tool list, or call corresponding tools from the tool list; use the tools to execute the tasks and return the task completion status to the central agent.
[0011] In a second aspect, the present application provides a star-shaped dynamic coding agent method for solving complex tasks, the method comprising:
[0012] The central agent performs the following steps:
[0013] Complex tasks that receive user input;
[0014] After thinking about it, the complex tasks are split and refined according to the task types, and the scheduling order of the distributed tasks is organized and the task list is output;
[0015] According to the scheduling order in the task list, the tasks are distributed to the corresponding encoding agents according to the task types;
[0016] Receive the task completion status returned by the encoding agent, further think about the task completion status, adjust the subsequent tasks, and then issue the next task until all tasks in the task list are completed;
[0017] The coding agent performs the following steps:
[0018] Receive tasks distributed by the central agent;
[0019] Generate reusable tools through coding according to task requirements and add them to the tool list, or call corresponding tools from the tool list;
[0020] The tool is used to execute the task and the task completion status is returned to the central agent.
[0021] In a third aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in the second aspect.
[0022] In a fourth aspect, the present application provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in the second aspect when executing the computer program.
[0023] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which implements the steps of the method described in the second aspect when executed by a processor.
[0024] Beneficial effects: The star-shaped dynamic coding Agent system and method for solving complex tasks provided by the present application have the following advantages: the system consists of a central Agent and multiple coding Agents. The central Agent is mainly responsible for the splitting and refinement of complex tasks, task distribution, and thinking and decision-making, with the goal of completing complex tasks, and is a unified entrance for users to use the Agent system; the coding Agent is mainly responsible for completing the received tasks, dynamically generating tools to complete the tasks according to task requirements, and finally returning the task completion status, with the goal of completing tasks in specific fields; the star topology structure is used to organize Agents to solve the problem of poor performance of a single Agent when handling complex tasks and mixed use of tools; the problem of fixed tools and difficult implementation is solved by dynamically generating tools through coding Agents. BRIEF DESCRIPTION OF THE DRAWINGS
[0025] Figure 1 A schematic diagram of a star-shaped dynamic coding Agent system for solving complex tasks according to an embodiment of the present application;
[0026] Figure 2 A schematic diagram of an application scenario of a method according to an embodiment of the present application;
[0027] Figure 3 A schematic diagram of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION
[0028] The present application is further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present application, and cannot be used to limit the protection scope of the present application.
[0029] This solution proposes a star-shaped dynamic coding agent system for solving complex tasks. It is an agent system composed of a central agent and multiple coding agents. More specifically, the central agent is mainly responsible for the splitting and refinement of complex tasks, task distribution, and decision-making, with the goal of completing complex tasks. It is the unified entrance for users to use the agent system; the coding agent is mainly responsible for completing the received tasks, dynamically generating tools to complete the tasks according to the task requirements, and finally returning the task completion status. It aims to complete the tasks in a specific field and is the actual executor of the task.
[0030] The professional terms involved in this application are as follows:
[0031] Large Language Model (LLM): A deep learning model that is trained using large amounts of text data, has billions of parameters, and can understand and process natural language.
[0032] Prompt words: text content used to prompt LLM to give targeted answers, usually including system prompts, which are used to prompt LLM to generate answers in the current context window in what role, in what scenario, and with what tone. User prompts generally refer to the conversation content entered by the user, which are used to prompt LLM to complete tasks, including but not limited to daily conversations, answering questions, translation, writing, and coding.
[0033] Context: LLM itself does not record the content of the conversation. The context is used to record all the current conversation content, including: prompt words, and the answer content generated by LLM. LLM infers and generates content based on the current context content. Of course, the context window size is limited. Currently, taking gpt-4o as an example, the context window size can reach 128k tokens.
[0034] Agent: An intelligent entity that uses LLM as its brain and can think, remember, perceive, act, and complete tasks autonomously.
[0035] Tool: This is implemented through the function_calling function of LLM. By providing the tool name and tool description, LLM calls the tool by replying with formatted JSON, executes the tool in the external operating environment, and returns the operating results to LLM. Agents can perceive external information or take actions through tools.
[0036] Chain of Thought (CoT) is a prompt word technology that guides LLM to think, which can effectively improve LLM's reasoning ability. The principle is that because LLM is generated token by token, the previously generated content also serves as a prompt for subsequent generation.
[0037] Robotic process automation (RPA) is a program that simulates the manual operations of users on the computer desktop through RPA process code to achieve process automation.
[0038] Example 1: Figure 1 As shown, a star-shaped dynamic encoding Agent system for solving complex tasks, the system includes: a central Agent and multiple encoding Agents;
[0039] Central Agent, used for:
[0040] Complex tasks that receive user input;
[0041] After thinking about it, the complex tasks are split and refined according to the task types, and the scheduling order of the distributed tasks is organized and the task list is output;
[0042] According to the scheduling order in the task list, the tasks are distributed to the corresponding encoding agents according to the task types;
[0043] Receive the task completion status returned by the encoding agent, further think about the task completion status, adjust the subsequent tasks, and then issue the next task until all tasks in the task list are completed;
[0044] Encoding Agent, used to:
[0045] Receive tasks distributed by the central agent;
[0046] Generate reusable tools through coding according to task requirements and add them to the tool list, or call corresponding tools from the tool list;
[0047] The tool is used to execute the task and the task completion status is returned to the central agent.
[0048] It should be further explained that the central agent is the unified entrance for users to interact with the system, so the complex tasks input by users are uniformly sent to the central agent. The central agent thinks and plans, splits and refines the complex tasks according to the task type, and organizes the scheduling order of the distributed tasks. The encoding agent completes the task and waits for the encoding agent to return the task completion status. It further thinks and adjusts the plan based on the task completion status, interacts with the user when necessary, prompts the user to supplement or confirm information, etc., until the complex task is completed.
[0049] Specifically, the prompt word technology of the thinking chain CoT is used to make the central agent always think before acting, which is conducive to better reasoning of LLM. For LLM, the so-called thinking is to let LLM generate some content of the nature of the thinking process, which is helpful for the final result of subsequent reasoning. When receiving a complex task from the user, first think about it, and talk to yourself about what different types of tasks can be divided and refined, the detailed description of the task, and the order of task scheduling, and then output the task list in the required JSON format. In the order of the task list, the tasks are distributed to specific encoding agents according to the task type. After a task is distributed to the specified encoding agent, wait for the encoding agent to execute the task and return the task completion status. According to the task completion status, further think about it, adjust the subsequent tasks, and then issue the next task. Until all tasks in the task list are completed, the overall task completion status is evaluated, and interact with the user. In the process of completing the task, if the user's cooperation is required to execute the task, a prompt message is sent to the user through the central agent to prompt the user to supplement or confirm the information.
[0050] In some specific instances, the complex task of generating RPA process code is that the RPA process usually consists of desktop operation tasks and data processing tasks.
[0051] Assume that the requirements of a complex task are: monitor air tickets for a certain time period on a certain booking platform, continuously obtain all flight information, save it to a local log file, automatically book tickets when the price is lower than a certain threshold, and send the booking information and log files to a specified mailbox.
[0052] Assume that the encoding agents in special fields are:
[0053]
[0054] Assume that the user input task is: monitor Ctrip's flight tickets from Nanjing to Beijing, continuously obtain all flight information, save it to a local log file, automatically book a ticket when the price is less than 500, and send the booking information and log file to a specified mailbox. Then the central agent performs the following steps:
[0055] 1. After thinking, split the tasks into smaller parts and output a JSON task list.
[0056] The tasks are initially split and refined. The agent field indicates which coding agent completes the task, and the task field indicates a detailed description of the task to be completed. [
[0058] {
[0059] "agent":"web_search_agent",
[0060] "task":"Explore the location information of the departure, destination, time, search button and other web page elements on Ctrip's flight page"
[0061] },
[0062] {
[0063] "agent":"web_code_agent",
[0064] "task":"Generate operation code for searching flight information based on web page element location information"
[0065] },
[0066] {
[0067] "agent":"python_code_agent",
[0068] "task":"Generate a python script to search for flights first, continuously obtain all flight information, and save it to a local log file. When the price is lower than 500, it will automatically book a ticket and send the booking information and log file to the user."
[0069] },
[0070] {
[0071] "agent":"web_search_agent",
[0072] "task":"Automatically explore the ticket booking process and return the location information of the web page elements"
[0073] },
[0074] {
[0075] "agent":"web_code_agent",
[0076] "task":"Generate ticket booking operation code based on web page element positioning information"
[0077] },
[0078] {
[0079] "agent":"python_code_agent",
[0080] "task":"Add booking code to python script"
[0081] },
[0082] {
[0083] "agent":"exec_code_agent",
[0084] "task":"Execute Python script"
[0085] }, ]
[0087] 2. Task scheduling and adjustment process Figure 2 As shown;
[0088] First, the web_search_agent is scheduled to explore the web page element location information of the operation steps for searching flight information, and wait for the location information to be returned;
[0089] Then, the web_code_agent is dispatched to generate an operation code for searching flights based on the positioning information. During the process, the user will be asked which day's flights to monitor, for example, December 1st.
[0090] Then dispatch python_code_agent to continue to generate subsequent processing code based on the searched flight code. The code will be tested and optimized in the agent that specializes in processing python code to ensure that the code can run correctly;
[0091] During this period, it may be necessary to return to inform the central agent that it needs to ask the user for information such as the email address and password. After the central agent obtains the information from the user, it will reschedule python_code_agent with the information to continue completing the task;
[0092] If the generated code is tested at this time and a library is found to be missing, the central agent is informed, and the central agent adjusts the task list, schedules system_env_agent to install the missing library, and then reschedules python_code_agent to continue completing the task;
[0093] The task scheduling and adjustment process of the central agent is roughly the same, until the complete code is generated and handed over to exec_code_agent to run the script.
[0094] It should be further explained that the coding agent is responsible for completing the tasks assigned by the central agent. It only needs to focus on the tasks in the specific field it is responsible for and generate reusable tools in the field when necessary. There is no need to generate them repeatedly. Limited tools are provided in advance by humans, or none are provided. The coding agent is left to dynamically generate tools to improve the list of reusable tools. When difficulties are encountered during the execution of the task, the task will be temporarily interrupted, the task completion status will be returned, and the central agent will be asked for help and required information, such as: user supplementary information is required, system information is required, etc. Until the task is completed, the task completion status will be returned.
[0095] Specifically, in the process of executing a task, the encoding agent will generate reusable tools through encoding according to the task requirements. Usually, the tool is a python function, and the generated content includes: python function code, function description, input and output parameters. The corresponding python function code is dynamically run when the tool is used. Each encoding agent manages its own tool list. The encoding agent can expand this tool list infinitely, and vectorize each tool to maintain a vector library for retrieval. When the number of tools reaches a certain number, RAG or Rerank can be used to retrieve the tools that may be used in the current task to ensure that it does not exceed the context window size of LLM, while increasing the probability of using the tool correctly. If RAG or Rerank does not retrieve a suitable tool, the encoding agent can also generate a temporary tool for use and append it to the tool list.
[0096] In some specific embodiments, for example, scheduling system_env_agent to complete the task of installing a python library, the encoding agent will dynamically generate the python function code of the installation library and abstract the code into a reusable tool. The generated content includes: the tool name is
[0097] install_python_package, the python function code for installing libraries, the function description is used to install python libraries, the input parameter is the library name. And the tool is appended to the tool list. When any task to install libraries arrives next time, system_env_agent only needs to use the tool and pass in the library name, without generating code again.
[0098] In summary, the star-shaped dynamic coding agent system for solving complex tasks provided by the present application solves the problem of poor performance of a single agent when handling complex tasks and mixed use of tools by adopting a star topology structure to organize agents; and solves the problem of fixed tools and difficult implementation by dynamically generating tools through coding agents.
[0099] Embodiment 2: Based on Embodiment 1, this embodiment provides a star-shaped dynamic coding Agent method for solving complex tasks, including:
[0100] The central agent performs the following steps: receiving the complex tasks input by the user; splitting and refining the complex tasks according to the task types after thinking, organizing the scheduling order of the distributed tasks, and outputting the task list;
[0101] Distribute tasks to corresponding encoding agents according to the scheduling order in the task list and task types; receive task completion status returned by the encoding agent, further consider the task completion status, adjust subsequent tasks, and then issue the next task until all tasks in the task list are completed;
[0102] The encoding agent performs the following steps: receiving the task distributed by the central agent; generating a reusable tool through encoding according to the task requirements and adding it to the tool list, or calling the corresponding tool from the tool list; executing the task using the tool, and returning the task completion status to the central agent.
[0103] Embodiment 3: Based on Embodiment 2, this embodiment provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in Embodiment 2 are implemented.
[0104] Embodiment 4: Based on embodiment 2, as Figure 3 As shown, this embodiment provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in Example 2 when executing the computer program.
[0105] Embodiment 5: Based on Embodiment 2, this embodiment proposes a hardware system that applies a star-shaped dynamic coding Agent method for solving complex tasks in Embodiment 2. The method is applied on the server side.
[0106] The terminal may be an electronic device with network access capability, for example, a desktop computer, a tablet computer, a laptop computer, a smart phone, a digital assistant, a shopping guide terminal, a television, etc.
[0107] The server can be an electronic device with certain computing and processing capabilities. For example, the server can be a server of a distributed system, or a system with multiple processors, memories, network communication modules, etc. operating in collaboration. The server can also be a cloud server, or an intelligent cloud computing server or intelligent cloud host with artificial intelligence technology. The server can also be a server cluster formed by several servers. Alternatively, with the development of science and technology, the server can also be a new technical means that can realize the corresponding functions of the implementation method of the specification. For example, it can be a new form of "server" based on quantum computing.
[0108] Embodiment 6: Based on Embodiment 2, this embodiment further provides a computer program product comprising instructions, which, when executed by a computer, enables the computer to execute a star-shaped dynamic coding Agent method for solving complex tasks in Embodiment 2.
[0109] It should be understood that the specific examples in this article are only intended to help those skilled in the art better understand the implementation methods of this specification, rather than limiting the scope of this application.
[0110] It can be understood that in the various implementations of this specification, the size of the sequence number of each process does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the implementation methods of this specification.
[0111] It can be understood that the various embodiments described in this specification can be implemented individually or in combination, and the embodiments of this specification are not limited to this.
[0112] Unless otherwise specified, all technical and scientific terms used in the embodiments of this specification have the same meaning as those generally understood by those skilled in the art of the technical field of this specification. The terms used in this specification are only for the purpose of describing specific embodiments and are not intended to limit the scope of this specification. The term "and / or" used in this specification includes any and all combinations of one or more related listed items. The singular forms of "a", "above", and "the" used in the embodiments of this specification and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.
[0113] It can be understood that the processor of the embodiment of this specification can be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method implementation can be completed by the hardware integrated logic circuit or software instructions in the processor. The above processor can be a general processor, a digital signal processor (Digital Signal Processor, DSP), an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), a field programmable gate array (Field Programmable Gate Array, FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiment of this specification can be implemented or executed. The general processor can be a microprocessor or the processor can also be any conventional processor, etc. The steps of the method disclosed in the embodiment of this specification can be directly embodied as a hardware decoding processor to perform, or the hardware and software modules in the decoding processor are combined and executed. The software module can be located in a mature storage medium in the field such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor reads the information in the memory and completes the steps of the above method in combination with its hardware.
[0114] It is understood that the memory in the embodiments of this specification may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memory. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (programmable ROM, PROM), an erasable programmable read-only memory (erasablePROM, EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory may be a random access memory (RAM). It should be noted that the memory of the systems and methods described herein is intended to include, but is not limited to, these and any other suitable types of memory.
[0115] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this specification.
[0116] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method implementation methods and will not be repeated here.
[0117] In the several embodiments provided in this specification, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device implementation described above is only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0118] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.
[0119] In addition, each functional unit in each embodiment of the present specification may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
[0120] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this specification, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of this specification. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc., various media that can store program codes.
[0121] The above is only a preferred implementation of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the technical principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.
Claims
1. A star-shaped dynamic coding Agent system, characterized in that: It includes a central agent and multiple encoding agents; The central agent is used to: receive complex tasks input by users; split and refine the complex tasks according to task types after thinking, organize the scheduling order of distributed tasks, and output a task list; Distribute tasks to corresponding encoding agents according to the scheduling order in the task list and task types; receive task completion status returned by the encoding agent, further consider the task completion status, adjust subsequent tasks, and then issue the next task until all tasks in the task list are completed; The encoding agent is used to: receive tasks distributed by the central agent; generate reusable tools through encoding according to task requirements and add them to the tool list, or call corresponding tools from the tool list; use the tools to execute the tasks and return the task completion status to the central agent.
2. The star-shaped dynamic coding Agent system according to claim 1 is characterized in that: After thinking about it, the complex tasks are split and refined according to the task types, and the scheduling order of the distributed tasks is organized and the task list is output, including: The prompt word technology of thinking chain CoT is used to make the central agent always think first and then split and refine.
3. The star-shaped dynamic coding Agent system according to claim 1 or 2, characterized in that: The task list is in JSON format.
4. The star-shaped dynamic coding Agent system according to claim 2 is characterized in that: The task list includes an agent field and a task field, wherein the agent field indicates which coding agent completes the task, and the task field indicates a detailed description of the task to be completed.
5. The star-shaped dynamic coding Agent system according to claim 1 is characterized in that: If the task requires user cooperation, the central agent is also used to: According to the task completion status, prompt information is sent to the user to prompt the user to supplement or confirm the information, and the user's supplemented or confirmed information is sent to the corresponding encoding agent to continue to complete the task.
6. The star-shaped dynamic coding Agent system according to claim 1 is characterized in that: The tool list is composed of pre-provided tools, or all tools are given to the coding agent to dynamically generate tools to improve the reusable tool list.
7. The star-shaped dynamic coding Agent system according to claim 1 or 6, characterized in that: The process of building the tool list includes: Each encoding agent manages its own tool list. The encoding agent can expand this tool list infinitely, vectorize each tool, and maintain a vector library for retrieval. When the number of tools in the tool list reaches a certain number, RAG or Rerank is used to first retrieve the tools that can be used for the current task to ensure that it does not exceed the context window size of LLM, while increasing the probability of correct use of tools; if RAG or Rerank does not retrieve a suitable tool, the encoding agent generates a reusable tool through encoding according to task requirements and appends it to the tool list.
8. The star-shaped dynamic coding Agent system according to claim 1 is characterized in that: The multiple encoding agents specifically include: web_search_agent, used to automatically explore all possible location information of web page elements; web_code_agent, used to generate web page operation code based on positioning information; python_code_agent, for generating any regular python code; exec_code_agent, which is responsible for running and managing code; system_env_agent is used to specifically handle system environment issues.
9. A star-shaped dynamic coding Agent method, characterized in that: The method comprises: The central agent performs the following steps: Complex tasks that receive user input; After thinking about it, the complex tasks are split and refined according to the task types, and the scheduling order of the distributed tasks is organized and the task list is output; According to the scheduling order in the task list, the tasks are distributed to the corresponding encoding agents according to the task types; Receive the task completion status returned by the encoding agent, further think about the task completion status, adjust the subsequent tasks, and then issue the next task until all tasks in the task list are completed; The coding agent performs the following steps: Receive tasks distributed by the central agent; Generate reusable tools through coding according to task requirements and add them to the tool list, or call corresponding tools from the tool list; The tool is used to execute the task and the task completion status is returned to the central agent.