An agent capability integration and reuse system and method based on cross-task memory

Abstract

The application provides an agent capability integration and reuse system and method based on cross-task memory. The system comprises: a cross-task memory module, configured to store and index historical task experience and capability metadata of an agent, and provide a historical task experience retrieval service to a task processing module; a capability encapsulation module, configured to standardize and encapsulate the agent capability, publish a capability interface, provide an available capability interface query service to the task processing module, and provide a capability calling interface to a communication execution module; the task processing module, configured to retrieve corresponding historical task experience after receiving a task request, acquire a corresponding available capability interface, and generate a capability calling scheme; the communication execution module, configured to call the corresponding capability encapsulated by the capability encapsulation module according to the capability calling scheme, and return an execution result to a memory updating module; and the memory updating module, configured to update the cross-task memory module according to the execution result and user feedback.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a system and method for integrating and reusing intelligent agent capabilities based on cross-task memory. Background Technology

[0002] With the rapid development of artificial intelligence technology, intelligent agents are being used more and more widely in various scenarios, covering areas such as multi-agent collaboration, tool invocation ecosystem building, and cross-platform intelligent system deployment. Users have placed higher demands on the diversity of capabilities, collaborative efficiency, and adaptability of intelligent agents, requiring them to be able to integrate tools and model capabilities from different sources and achieve efficient reuse.

[0003] In the current process of integrating and reusing intelligent agent capabilities, the communication protocols offered by different technology providers lack a unified standard, such as OpenAI's Function Calls and Google's Tool APIs, leading to incompatibility between models and tools from different ecosystems. Simultaneously, the memory mechanisms of intelligent agents are mostly limited to single-task implementations, lacking the ability to accumulate and transfer experience across tasks. This prevents the optimization of similar task processing using historical data, severely impacting the operational efficiency and scalable application of intelligent agents. Existing technologies mainly include: RAG technology: It is mainly used to retrieve and extract information from external documents. It can only support agents to passively receive external document information, and cannot realize active capability reuse and experience transfer. It does not involve the standardized integration of multiple models and tools. The ReAct framework achieves tool call command parsing and execution through prompt engineering, but the stability of command parsing depends on the prompt design, lacks flexibility when adapting to different tools and models, and does not establish a cross-task memory management mechanism. Traditional Function Call protocol: It adopts a static tool definition method, which has poor compatibility with different models. When adding or replacing tools, adaptation code needs to be rewritten. The capability reusability is low, and the lack of a unified interface specification makes system expansion difficult. None of the aforementioned existing technologies simultaneously solve the three core problems of protocol compatibility, cross-task memory reuse, and system scalability, and their main shortcomings include the following: 1. The communication protocols of different models and tools in the existing intelligent agent field are incompatible with each other, and there are language barriers. This forces developers to write exclusive adaptation code for different combinations, resulting in low development efficiency and the problem of "reinventing the wheel". 2. The intelligent agent lacks an effective cross-task memory mechanism. Experience data and ability calling schemes from past tasks cannot be efficiently transferred and reused. In similar task scenarios, the entire reasoning process needs to be re-executed, resulting in low task processing efficiency. 3. In traditional intelligent agent capability integration methods, the high degree of coupling between modules and the lack of unified interfaces make system expansion difficult and make it hard to adapt to dynamically changing task requirements, which restricts the large-scale implementation of intelligent agent technology. 4. Some capabilities rely on native model support, and small models have difficulty adapting to complex capability calling scenarios, which limits the application scope of intelligent agent technology. Summary of the Invention

[0004] The purpose of this invention is to provide a system and method for integrating and reusing intelligent agent capabilities based on cross-task memory, in order to solve the above-mentioned problems in the prior art.

[0005] This invention provides a system for integrating and reusing intelligent agent capabilities based on cross-task memory, comprising: A cross-task memory module, connected to the task processing module and the memory update module, is used to store and index the agent's historical task experience and capability metadata, and to provide historical task experience retrieval services to the task processing module. The capability encapsulation module, connected to the task processing module and the communication execution module, is used to standardize and encapsulate the capabilities of the intelligent agent and publish capability interfaces, provide the task processing module with an available capability interface query service, and provide the communication execution module with a capability call interface; The task processing module, connected to the cross-task memory module, the capability encapsulation module, and the communication execution module, is used to receive task requests, retrieve corresponding historical task experience from the cross-task memory module according to the task request, obtain the corresponding available capability interface from the capability encapsulation module, generate a capability invocation scheme, and send it to the communication execution module. The communication execution module, connected to the capability encapsulation module, task processing module, and memory update module, is used to receive the capability invocation scheme, invoke the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and return the execution result to the memory update module. The memory update module, connected to the communication execution module and the cross-task memory module, is used to update the cross-task memory module based on the execution results and user feedback.

[0006] This invention provides a method for integrating and reusing intelligent agent capabilities based on cross-task memory, comprising: The cross-task memory module stores and indexes the agent's historical task experience and capability metadata, and provides historical task experience retrieval services to the task processing module. The capability encapsulation module standardizes and encapsulates the capabilities of the intelligent agent and publishes capability interfaces, providing the task processing module with a query service for available capability interfaces and the communication execution module with a capability call interface. The task processing module receives task requests, retrieves corresponding historical task experience from the cross-task memory module based on the task requests, obtains corresponding available capability interfaces from the capability encapsulation module, generates a capability invocation scheme, and sends it to the communication execution module. The communication execution module receives the capability invocation scheme, invokes the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and returns the execution result to the memory update module. The cross-task memory module is updated based on the execution results and user feedback through the memory update module.

[0007] This invention also provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of the above-described method for integrating and reusing intelligent agent capabilities based on cross-task memory.

[0008] This invention also provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, implements the steps of the above-described method for integrating and reusing intelligent agent capabilities based on cross-task memory.

[0009] The embodiments of the present invention can include the following beneficial effects: The embodiments of the present invention solve the core pain points of intelligent agent capability integration and reuse through the dual paths of "breaking compatibility barriers through standardized protocols" and "achieving experience reuse through hierarchical memory". The two work together to achieve efficiency improvement and optimization of expansion capabilities. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in one or more embodiments of this specification or in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a schematic diagram of an intelligent agent capability integration and reuse system based on cross-task memory, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the hierarchical structure of the cross-task memory module according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the ternary architecture of the MCP protocol component according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the capability integration and reuse process according to an embodiment of the present invention; Figure 5This is a flowchart of the intelligent agent capability integration and reuse method based on cross-task memory according to an embodiment of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the technical solutions in one or more embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this specification, and not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of this document.

[0013] System Implementation Examples According to embodiments of the present invention, a system for integrating and reusing intelligent agent capabilities based on cross-task memory is provided. Figure 1 This is a schematic diagram of an intelligent agent capability integration and reuse system based on cross-task memory, as an embodiment of the present invention. Figure 1 As shown, the intelligent agent capability integration and reuse system based on cross-task memory according to an embodiment of the present invention specifically includes: A cross-task memory module 10, connected to the task processing module and the memory update module, is used to store and index the agent's historical task experience and capability metadata, and to provide historical task experience retrieval services to the task processing module; specifically including: The task experience layer, connected to the memory update module, is used to store process data, result feedback, and optimization strategies of the agent in executing historical tasks. The capability metadata layer, connected to the task processing module, is used to store the capability metadata of the intelligent agent; wherein, the capability metadata includes capability name, input parameter format, output data type, performance index, and applicable scenarios; The index layer is reused and connected to the task experience layer, capability metadata layer and task processing module to establish an association index between task features and capability metadata, and to provide the task processing module with a fast retrieval service based on task features. The negative experience base, connected to the memory update module and the reuse index layer, is used to independently store the task call chain, error parameter combination, timeout record and user negative feedback data of failed tasks, and to provide the reuse index layer with the marking information of failed tasks; The memory evolution engine is connected to the task experience layer, capability metadata layer, reuse index layer, and negative experience library. It is used to periodically mine and analyze the positive experience data stored in the task experience layer and the negative experience data stored in the negative experience library. It dynamically adjusts the weight of different capability combinations in the reuse index layer based on historical execution success rate and efficiency, and generates corresponding avoidance strategies by combining the failed task data in the negative experience library.

[0014] Capability encapsulation module 12, connected to the task processing module and the communication execution module, is used to standardize and encapsulate the capabilities of the intelligent agent and publish capability interfaces, provide an available capability interface query service to the task processing module, and provide a capability call interface to the communication execution module; specifically including: The MCP protocol server, connected to the communication execution module and the dynamic capability registration center, is used to encapsulate the capabilities of agents from different sources into standardized capability interfaces according to the MCP protocol specification; wherein, the capability interface includes capability name, parameter format, return type and calling constraints. A dynamic capability registration center, connected to the MCP protocol server, task processing module, and communication execution module, is used to publish, update, and deregister capability interfaces encapsulated by the MCP protocol server, provide available capability interface query services to the task processing module, and provide capability call routing information to the communication execution module; specifically including: The registration and publishing unit is used to broadcast its encapsulated capability interfaces when the MCP protocol server goes online and to automatically deregister the corresponding capability interfaces when it goes offline. The version management unit is used to manage the versions of capability interfaces and generate update notifications when capability interfaces change. The notification subscription unit is used to receive capability subscription requests from MCP protocol clients and push update notifications to subscribed MCP protocol clients in real time when the capability interface changes. The protocol negotiation submodule, integrated into the MCP protocol server and connected to the communication execution module, is used to negotiate the communication data format and compression algorithm when establishing a connection with the communication execution module.

[0015] Task processing module 14, connected to the cross-task memory module, capability encapsulation module, and communication execution module, is used to receive task requests, retrieve corresponding historical task experience from the cross-task memory module according to the task request, obtain corresponding available capability interfaces from the capability encapsulation module, generate a capability invocation scheme, and send it to the communication execution module; specifically including: The task parsing unit is used to receive task requests from users, parse and extract the task features and constraints of the task requests; The memory retrieval unit, connected to the task parsing unit and the cross-task memory module, is used to retrieve the capability invocation scheme of historical similar tasks from the reuse index layer of the cross-task memory module according to the task characteristics, and to obtain the failure task marking information to be avoided from the negative experience base. The capability query unit, connected to the capability encapsulation module, is used to query the dynamic capability registry center for a list of currently available standardized capability interfaces. The scheme generation unit, connected to the memory retrieval unit, the capability query unit, and the communication execution module, is used to combine the capability invocation schemes of similar historical tasks, the failed task marking information, and the currently available standardized capability interface list to generate the optimal capability invocation scheme for the current task through a matching algorithm, and send it to the communication execution module. The matching algorithm is a dynamic weighted scoring mechanism, and its scoring factors include suitability, execution efficiency, historical success rate, and negative avoidance coefficient.

[0016] The communication execution module 16, connected to the capability encapsulation module, task processing module, and memory update module, is used to receive the capability invocation scheme, invoke the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and return the execution result to the memory update module; specifically including: The MCP protocol client is connected to the task processing module, capability encapsulation module, and memory update module. It is used to receive the capability invocation scheme generated by the task processing module, convert the capability invocation scheme into an invocation instruction conforming to the MCP protocol format and send it to the capability encapsulation module, and receive the execution result and return it to the memory update module. A communication channel is used to establish a communication connection between the MCP protocol client and the MCP protocol server. The MCP protocol client integrates a protocol negotiation submodule, which is used to negotiate the communication data format and compression algorithm when establishing a connection with the MCP protocol server; the MCP protocol client also integrates a capability discovery unit, which is used to subscribe to the list of available capability interfaces from the dynamic capability registry center and receive real-time update notifications when capability interfaces change.

[0017] The memory update module 18 is connected to the communication execution module and the cross-task memory module, and is used to update the cross-task memory module according to the execution results and user feedback.

[0018] The following describes in detail the above-mentioned technical solutions of the present invention with reference to the specific circumstances of the intelligent agent capability integration and reuse system based on cross-task memory in the embodiments of the present invention.

[0019] The technical terms are explained below: Protocol compatibility: refers to the communication compatibility between different models (such as Claude, Qwen2.5, etc.) and tools (such as map query tools, message push tools, etc.). Due to the lack of unified protocols in existing technologies, it is necessary to develop specific adaptation logic. Cross-task memory: refers to the ability of an intelligent agent to store, index, and reuse data such as task experience, capability call parameters, and execution result feedback during the execution of different tasks. Existing technologies lack a dedicated hierarchical storage and indexing mechanism to support this function. Module coupling: refers to the degree of dependency between various functional modules of an agent (such as capability invocation module and task processing module). High coupling means that the overall architecture needs to be adjusted when modifying or adding modules, which affects the efficiency of expansion. MCP protocol: Model and Component Communication Protocol, is a standardized communication protocol defined in this invention to unify the interface specifications of models and tools; Capability metadata: Standardized data describing the core attributes of an agent's capabilities, which forms the basis for capability matching and invocation.

[0020] This invention achieves efficient integration and reuse of agent capabilities through a combination of a "cross-task memory module + enhanced MCP protocol component + adaptive capability integration and reuse process," as detailed below: 1. Cross-task memory module design (including task experience layer, capability metadata layer, reuse index layer, negative experience base, memory evolution engine, distributed storage architecture, semantic clustering algorithm): A. Structural Composition: It adopts a hierarchical memory storage structure, comprising three core layers: a task experience layer, a capability metadata layer, and a reusable index layer. The overall deployment is based on a distributed storage architecture; specifically as follows... Figure 2 As shown, it includes a task experience layer (time-series database, storing task process data, result feedback, and optimization strategies), a capability metadata layer (structured database, storing capability names, parameter formats, performance indicators, and applicable scenarios), a reuse index layer (semantic clustering index, establishing a mapping between tasks and capabilities), and distributed storage nodes (supporting distributed storage and access of the three layers of data).

[0021] B. Implementation at each level: Task Experience Layer: A time-series database is used to store process data (such as task triggering conditions, ability call order, parameter configuration), result feedback (such as execution success rate, user evaluation), and optimization strategies (such as parameter adjustment scheme) of the agent in executing various tasks, to ensure the timeliness and traceability of the data; Capability metadata layer: A structured database is used to store standardized capability description information, including: capability name, input parameter format, output data type, performance indicators (such as response time and accuracy), applicable scenario scope, etc., to provide basic data support for capability matching; Reusable index layer: Based on semantic clustering algorithm, it performs correlation analysis on task features and capability metadata, establishes a correlation index between tasks and adaptation capabilities, and supports quick retrieval of the optimal capability combination scheme through task keywords and scene features; Negative experience database: Independent of positive experience storage, it specifically records failed task call chains, parameter combinations that cause errors, timeout capability call records, and user feedback data that explicitly rejects the data. By constructing a taboo matrix, it proactively avoids similar error paths when generating subsequent solutions, preventing repeated pitfalls. Memory Evolution Engine: Regularly performs deep learning mining on the data in the task experience layer and negative experience library. Using reinforcement learning algorithms (such as Q-learning), it automatically adjusts the weights of different capability combinations in the reuse index layer based on the success rate and efficiency of historical execution, thereby achieving self-optimization and evolution of the memory library.

[0022] C. Working Principle: The module receives various types of data in real time during task execution, stores them hierarchically, and updates the index. The evolution engine periodically mines historical data (including negative experience) and optimizes index weights. When a new task is triggered, it quickly matches similar historical tasks and corresponding capability invocation schemes through the index and automatically filters out failed schemes in the negative experience library.

[0023] 2. Capability encapsulation mechanism based on the enhanced MCP protocol (including host, client, server, WebSocket communication channel, Python protocol interface, protocol negotiation module, and dynamic capability registry center): A. Architecture Composition: A three-tier architecture of "host-client-server" is adopted, with all three components establishing a long-connection communication channel via the WebSocket protocol to ensure the real-time performance and stability of data transmission; specifically as follows... Figure 3 As shown, it includes a host (containing a task scheduling engine, capability matching algorithm, and result integration module, which receives user requests and coordinates the overall process), a client (containing a protocol parsing module and a format conversion module, responsible for tool discovery, instruction conversion, and result feedback), a server (containing an MCP protocol capability encapsulation interface, which implements standardized capability encapsulation and registration), and a WebSocket communication channel (connecting the client and server to achieve real-time data transmission).

[0024] B. Implementation of each component: Server (Capability Encapsulator): Implements the capability encapsulation interface of the MCP protocol using Python. It describes the names, parameter formats, return types and calling constraints of intelligent agent capabilities from different sources (such as tool invocation, inference calculation, data processing, etc.) in a unified manner according to the MCP protocol specification. After completing the standardized encapsulation of capabilities, it publishes them through the dynamic capability registry center. Dynamic Capability Registry: Replacing static registration, it adopts a publish-subscribe model. When the server goes online, it broadcasts its own capabilities and automatically deregisters when it goes offline. It supports capability version management. When a capability is upgraded (such as parameter changes), it notifies the client to update the cache in real time to ensure the accuracy of the call. The client (protocol translator) integrates a protocol parsing and format conversion module, responsible for tool discovery (subscribing to a list of available capabilities from the registry center), command conversion (converting host-generated call commands into the MCP protocol format), and result feedback (converting server-returned execution results into a host-recognizable format). It also integrates a protocol negotiation submodule, negotiating the data format (such as JSON or ProtoBuf) and compression algorithm with the server during connection establishment to adapt to different bandwidth environments. Host (interaction core): Deploys the task scheduling engine, capability matching algorithm and result integration module, and is responsible for receiving user requirements, coordinating the collaboration process between the client and the server, and generating the optimal capability combination solution; C. Working principle: The server publishes standardized capability interfaces through a dynamic registry center, the client implements protocol adaptation and negotiation, and the host coordinates the entire task processing process. The three work together to achieve seamless communication between different models and tools.

[0025] 3. Capability integration and reuse processes, such as Figure 4 As shown: Step 1: Initialization Phase: Complete the database construction of the cross-task memory module (time-series database, structured database, negative experience base) and initialize the reuse index; the MCP server completes the standardized encapsulation of the target capabilities and registers with the dynamic registry center; the client and server establish a WebSocket long connection and confirm the transmission format through the protocol negotiation module; Step 2: Demand Triggering Stage: The host receives user requests and extracts core task objectives (such as planning a home control + message push collaboration solution) and constraints (such as response time and platform compatibility) through the task parsing module. Step 3: Memory Retrieval Stage: The host calls the cross-task memory module, and based on the task characteristics, retrieves the capability combination schemes, execution effects and optimization strategies of similar historical tasks through the reuse index layer. At the same time, it queries the negative experience base and marks the erroneous schemes that need to be avoided this time. Step 4: Capability Query Phase: The client subscribes to / queries the currently available standardized capability list (including capability name, parameter format, performance indicators, etc.) from the dynamic registry center of the MCP server and synchronizes it to the host; Step 5: Solution Generation Stage: The host's capability matching algorithm combines the memory retrieval results (positive + negative) with the real-time capability list, and uses a dynamic weighted scoring mechanism to generate the optimal capability combination solution. The weighting factors of this mechanism include not only adaptability, execution efficiency, and historical success rate, but also a negative avoidance coefficient, which reduces the weight of similar calls that have failed in the past, ensuring the robustness of the solution; Step 6: Protocol Conversion and Execution Phase: The host converts the capability combination scheme into a call instruction, the client parses the instruction into the MCP protocol format and sends it to the server; the server executes the capability call, completes the collaborative operation of the tool / model, and returns the execution result. Step 7: Result Feedback Stage: The client converts the results returned by the server into a format that the host can recognize. The host then integrates and processes the results and sends them back to the user. Step 8: Memory Update Phase: The cross-task memory module records the capability combination scheme, execution effect, user feedback, and optimization adjustment information for this task. If the execution fails or the user evaluation is poor, the data is stored in the negative experience base; if the execution is successful, the data is stored in the task experience layer; then the memory evolution engine is triggered to perform incremental learning, dynamically adjusting the reuse index layer and the negative avoidance coefficient, forming a closed loop of "experience-reuse-evolution".

[0026] In summary, the key to the embodiments of the present invention lies in: 1. The hierarchical storage structure design of the cross-task memory module achieves efficient storage, associated indexing, and fast retrieval of task experience data and capability information through the collaboration of the task experience layer, capability metadata layer, and reuse index layer; 2. The negative experience library and memory evolution engine introduced in the cross-task memory module achieve self-optimization of the memory library and avoidance of error paths by storing failure cases and combining them with reinforcement learning algorithms, which is different from the traditional single positive memory mechanism. 3. The “host-client-server” ternary architecture based on the MCP protocol, and the dynamic capability registration center and protocol negotiation module introduced on this basis, realize the standardized encapsulation, dynamic discovery and adaptive communication of different models and tool capabilities, and break the protocol compatibility barrier. 4. It includes a closed-loop adaptive capability integration and reuse process that includes "requirement triggering - memory retrieval (including negative filtering) - capability query - solution generation (dynamic weighting) - protocol conversion - execution feedback - memory evolution", ensuring effective transfer of experience across tasks and efficient reuse of capabilities; 5. Reuse the semantic clustering algorithm used in the index layer and the dynamic weighted scoring ability matching algorithm with negative avoidance coefficient introduced by the host to improve the matching accuracy of tasks and capabilities and the calling efficiency.

[0027] Preferably, in addition to a distributed storage architecture, the cross-task memory module of this embodiment of the invention can also adopt a centralized storage architecture, which realizes centralized storage and management of three-layer data through a high-performance server, and is suitable for application scenarios with small task scale and limited data volume; In addition to the WebSocket protocol, the communication protocols in this embodiment of the invention can also use HTTP / 2 or gRPC protocols to build a communication channel between the client and the server, which can also achieve efficient data transmission and real-time interaction, adapting to different network environment requirements; In addition to the weighted scoring mechanism, the capability matching algorithm in this embodiment of the invention can also use neural network models (such as CNN, RNN) or Transformer architecture to achieve deep matching of task features and capability metadata, thereby improving the accuracy of capability combination schemes in complex scenarios. In addition to semantic clustering algorithms, the memory indexing algorithm in this embodiment of the invention can also use K-nearest neighbor algorithm or decision tree algorithm to establish an association index between tasks and capabilities, so as to meet the fast retrieval needs under different data characteristics. In addition to reinforcement learning, the memory evolution engine in this embodiment of the invention can also use genetic algorithms to simulate population evolution of ability combination schemes and find the global optimal solution, which is suitable for scenarios with high requirements for scheme exploration.

[0028] In smart home control scenarios, by integrating Baidu Maps route query, Slack message push, and home device control capabilities through the embodiments of the present invention, test results show that the cross-model call success rate reaches 98%, and the task completion efficiency is improved by 55% compared with traditional solutions, which can intuitively demonstrate the technical effect of the embodiments of the present invention.

[0029] Method Implementation Examples According to embodiments of the present invention, a method for integrating and reusing intelligent agent capabilities based on cross-task memory is provided. Figure 5 This is a flowchart of the intelligent agent capability integration and reuse method based on cross-task memory according to an embodiment of the present invention, as follows: Figure 5 As shown, the method for integrating and reusing agent capabilities based on cross-task memory according to an embodiment of the present invention specifically includes: Step S501: Store and index the agent's historical task experience and capability metadata through the cross-task memory module, and provide historical task experience retrieval service to the task processing module; Step S502: The capability encapsulation module standardizes and encapsulates the capabilities of the intelligent agent and publishes the capability interface, providing the task processing module with an available capability interface query service and the communication execution module with a capability call interface. Step S503: Receive a task request through the task processing module, retrieve the corresponding historical task experience from the cross-task memory module according to the task request, obtain the corresponding available capability interface from the capability encapsulation module, generate a capability invocation scheme, and send it to the communication execution module. Step S504: Receive the capability invocation scheme through the communication execution module, invoke the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and return the execution result to the memory update module; Step S505: The cross-task memory module is updated by the memory update module based on the execution results and user feedback.

[0030] The embodiments of the present invention are method embodiments corresponding to the system embodiments described above. The specific operations of each step can be understood by referring to the description of the system embodiments, and will not be repeated here.

[0031] In summary, compared with the prior art, the beneficial effects of the embodiments of the present invention include: 1. Enhanced compatibility: As a unified interaction standard, the MCP protocol, combined with dynamic registration and protocol negotiation mechanisms, supports seamless integration with various models and tools. It has stronger adaptability when the network fluctuates or the capability version changes. There is no need to develop adaptation code for specific combinations, which solves the problem of protocol incompatibility in existing technologies and greatly reduces integration costs. 2. Higher reuse efficiency: The cross-task memory module effectively accumulates task experience and capability data. In particular, it avoids repeated trial and error through the negative experience library, which significantly improves the success rate of the first call. In similar task scenarios, existing capability solutions can be directly reused without re-executing the complete reasoning process. Tests show that the task processing efficiency is improved by more than 55% compared with the traditional solution, and the reasoning time is shortened by more than 40%. 3. Superior intelligence and robustness: The memory evolution engine enables the system to learn independently, continuously optimizing the matching strategy with increasing usage; the negative experience base effectively prevents the recurrence of past errors, improving the system's stability by 30% in complex dynamic environments; 4. Enhanced scalability: The modular design and dynamic capability registry center enable the system to support plug-and-play and hot-swappable capability modules. Adding or replacing tools / models does not require restarting the service or reconstructing the overall architecture, solving the problems of high module coupling and difficulty in expansion in existing technologies. 5. Wider range of applications: Without relying on native model support, through client protocol conversion and prompt engineering adaptation, even small models can realize the invocation of complex capabilities, expanding the application scenarios of intelligent agent technology.

[0032] Device Example 1 This invention provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it performs the steps described in the method embodiment.

[0033] Device Example 2 This invention provides a computer-readable storage medium storing an information transmission implementation program, which, when executed by a processor, performs the steps described in the method embodiment.

[0034] The computer-readable storage media described in this embodiment include, but are not limited to, ROM, RAM, disk, or optical disk.

[0035] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for integrating and reusing intelligent agent capabilities based on cross-task memory, characterized in that, include: A cross-task memory module, connected to the task processing module and the memory update module, is used to store and index the agent's historical task experience and capability metadata, and to provide historical task experience retrieval services to the task processing module. The capability encapsulation module, connected to the task processing module and the communication execution module, is used to standardize and encapsulate the capabilities of the intelligent agent and publish capability interfaces, provide the task processing module with an available capability interface query service, and provide the communication execution module with a capability call interface; The task processing module, connected to the cross-task memory module, the capability encapsulation module, and the communication execution module, is used to receive task requests, retrieve corresponding historical task experience from the cross-task memory module according to the task request, obtain the corresponding available capability interface from the capability encapsulation module, generate a capability invocation scheme, and send it to the communication execution module. The communication execution module, connected to the capability encapsulation module, task processing module, and memory update module, is used to receive the capability invocation scheme, invoke the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and return the execution result to the memory update module. The memory update module, connected to the communication execution module and the cross-task memory module, is used to update the cross-task memory module based on the execution results and user feedback.

2. The system according to claim 1, characterized in that, The cross-task memory module specifically includes: The task experience layer, connected to the memory update module, is used to store process data, result feedback, and optimization strategies of the agent in executing historical tasks. The capability metadata layer, connected to the task processing module, is used to store the capability metadata of the intelligent agent; wherein, the capability metadata includes capability name, input parameter format, output data type, performance index, and applicable scenarios; The index layer is reused and connected to the task experience layer, capability metadata layer and task processing module to establish an association index between task features and capability metadata, and to provide the task processing module with a fast retrieval service based on task features.

3. The system according to claim 2, characterized in that, The cross-task memory module also includes: The negative experience base, connected to the memory update module and the reuse index layer, is used to independently store the task call chain, error parameter combination, timeout record and user negative feedback data of failed tasks, and to provide the reuse index layer with the marking information of failed tasks; The memory evolution engine is connected to the task experience layer, capability metadata layer, reuse index layer, and negative experience library. It is used to periodically mine and analyze the positive experience data stored in the task experience layer and the negative experience data stored in the negative experience library. It dynamically adjusts the weight of different capability combinations in the reuse index layer based on historical execution success rate and efficiency, and generates corresponding avoidance strategies by combining the failed task data in the negative experience library.

4. The system according to claim 3, characterized in that, The capability encapsulation module specifically includes: The MCP protocol server, connected to the communication execution module and the dynamic capability registration center, is used to encapsulate the capabilities of agents from different sources into standardized capability interfaces according to the MCP protocol specification; wherein, the capability interface includes capability name, parameter format, return type and calling constraints. The dynamic capability registration center is connected to the MCP protocol server, task processing module, and communication execution module. It is used to publish, update, and deregister the capability interfaces encapsulated by the MCP protocol server, provide available capability interface query services to the task processing module, and provide capability call routing information to the communication execution module. The protocol negotiation submodule, integrated into the MCP protocol server and connected to the communication execution module, is used to negotiate the communication data format and compression algorithm when establishing a connection with the communication execution module.

5. The system according to claim 4, characterized in that, The dynamic capability registry specifically includes: The registration and publishing unit is used to broadcast its encapsulated capability interfaces when the MCP protocol server goes online and to automatically deregister the corresponding capability interfaces when it goes offline. The version management unit is used to manage the versions of capability interfaces and generate update notifications when capability interfaces change. The notification subscription unit is used to receive capability subscription requests from MCP protocol clients and push update notifications to subscribed MCP protocol clients in real time when capability interfaces change.

6. The system according to claim 4, characterized in that, The task processing module specifically includes: The task parsing unit is used to receive task requests from users, parse and extract the task features and constraints of the task requests; The memory retrieval unit, connected to the task parsing unit and the cross-task memory module, is used to retrieve the capability invocation scheme of historical similar tasks from the reuse index layer of the cross-task memory module according to the task characteristics, and to obtain the failure task marking information to be avoided from the negative experience base. The capability query unit, connected to the capability encapsulation module, is used to query the dynamic capability registry center for a list of currently available standardized capability interfaces. The scheme generation unit, connected to the memory retrieval unit, the capability query unit, and the communication execution module, is used to combine the capability invocation schemes of similar historical tasks, the failed task marking information, and the currently available standardized capability interface list to generate the optimal capability invocation scheme for the current task through a matching algorithm, and send it to the communication execution module. The matching algorithm is a dynamic weighted scoring mechanism, and its scoring factors include suitability, execution efficiency, historical success rate, and negative avoidance coefficient.

7. The system according to claim 4, characterized in that, The communication execution module specifically includes: The MCP protocol client is connected to the task processing module, capability encapsulation module, and memory update module. It is used to receive the capability invocation scheme generated by the task processing module, convert the capability invocation scheme into an invocation instruction conforming to the MCP protocol format and send it to the capability encapsulation module, and receive the execution result and return it to the memory update module. A communication channel is used to establish a communication connection between the MCP protocol client and the MCP protocol server. The MCP protocol client integrates a protocol negotiation submodule, which is used to negotiate the communication data format and compression algorithm when establishing a connection with the MCP protocol server; the MCP protocol client also integrates a capability discovery unit, which is used to subscribe to the list of available capability interfaces from the dynamic capability registry center and receive real-time update notifications when capability interfaces change.

8. A method for integrating and reusing intelligent agent capabilities based on cross-task memory, characterized in that, include: The cross-task memory module stores and indexes the agent's historical task experience and capability metadata, and provides historical task experience retrieval services to the task processing module. The capability encapsulation module standardizes and encapsulates the capabilities of the intelligent agent and publishes capability interfaces, providing the task processing module with a query service for available capability interfaces and the communication execution module with a capability call interface. The task processing module receives task requests, retrieves corresponding historical task experience from the cross-task memory module based on the task requests, obtains corresponding available capability interfaces from the capability encapsulation module, generates a capability invocation scheme, and sends it to the communication execution module. The communication execution module receives the capability invocation scheme, invokes the corresponding capability encapsulated by the capability encapsulation module according to the capability invocation scheme, and returns the execution result to the memory update module. The cross-task memory module is updated based on the execution results and user feedback through the memory update module.

9. An electronic device, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the method for integrating and reusing intelligent agent capabilities based on cross-task memory as described in claim 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores an implementation program for information transmission, which, when executed by a processor, implements the steps of the method for integrating and reusing intelligent agent capabilities based on cross-task memory as described in claim 8.

Images