Task processing method, system, medium, and electronic device

Abstract

The present disclosure relates to a task processing method, system, medium and electronic device, the method is applied to a first agent end, comprising: generating a task list according to a task request input by a user, the task list comprising a plurality of subtasks, each subtask corresponding to different task constraints, the task constraints being used to represent the task conditions required for executing the subtask; sending a consumption request to a second agent end, so that the second agent end matches a target subtask from the plurality of subtasks of the task list according to a task processing type pre-configured by itself, checks the task constraints of the target subtask, determines a checking result, and processes the target subtask according to the checking result, thereby improving the cloud resource management and control capability.

Description

Technical Field

[0001] This disclosure relates to the fields of cloud computing and artificial intelligence technologies, and more specifically, to a task processing method, system, medium, and electronic device. Background Technology

[0002] As cloud computing continues to expand, the management and control of cloud resource platforms are becoming increasingly complex. The industry is gradually introducing AI (Artificial Intelligence) technologies, such as large-scale models, to simplify operations and maintenance, hoping to leverage natural language for resource scheduling. Traditional cloud resource management architectures are no longer adequate for the demands of intelligent operations and maintenance, and the industry urgently needs to optimize its cloud platform task scheduling and management models.

[0003] In related technologies, cloud resource management often relies on native APIs (Application Programming Interfaces), custom scripts, or fixed automated processes for scheduling. Such solutions have fixed automated processes, which cannot adapt to complex tasks that cross resource categories. In addition, they usually directly issue the output instructions of large models for execution. Due to the influence of model illusion, unauthorized and erroneous operations will occur during task execution, making it difficult to guarantee the operational stability of the cloud computing production environment. Summary of the Invention

[0004] The purpose of this disclosure is to provide a task processing method, system, medium, and electronic device to improve cloud resource management capabilities.

[0005] To achieve the above objectives, in a first aspect, this disclosure provides a task processing method applied to a first agent, the method comprising: Based on the task request input by the user, a task list is generated. The task list includes multiple sub-tasks, and each sub-task corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the sub-task. A consumption request is sent to the second agent so that the second agent can match the target subtask from multiple subtasks in the task list according to its own pre-configured task processing type, verify the task constraints of the target subtask, determine the verification result, and process the target subtask according to the verification result.

[0006] Optionally, the task constraints include at least one of resource type constraints, permission and identity constraints, prerequisite dependency constraints, and rollback requirement constraints.

[0007] Optionally, the method further includes: The completion status of each subtask in the task list is checked according to a preset cycle, and the completion rate of each subtask is determined based on its completion status. If the completion rate of the subtask is greater than or equal to the preset completion rate, the global task is determined to be completed and the task result is output. The task result is determined based on the completed subtasks. If the completion rate of the subtask is less than the preset completion rate, the global task is determined to have failed.

[0008] Secondly, this disclosure also provides a task processing method applied to a second agent, the method comprising: Upon receiving a consumption request from the first agent, the target subtask is matched from multiple subtasks in the task list according to the task processing type pre-configured by the second agent. The task list is generated by the first agent based on the task request input by the user. The task list includes multiple subtasks, each of which corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the subtask. The task constraints of the target subtask are verified, and the verification result is determined. The target subtask is processed based on the verification results.

[0009] Optionally, the second agent is pre-configured with a skill library, which includes multiple skills, each corresponding to a different skill identifier and skill constraints. The skill constraints characterize the conditions required to execute the skill. The step of validating the task constraints of the target sub-task and determining the validation result includes: The skill identifier of the target subtask is obtained by reading it; The skill identifiers of multiple skills in the skill library are matched with the skill identifiers of the target sub-task to determine the first verification result; If the first verification result indicates failure, the verification result is determined to be unsuccessful. If the first verification result indicates success, a target skill that successfully matches the target subtask is determined, and the task constraints of the target subtask are verified according to the skill constraints of the target skill to determine a second verification result. If the second verification result indicates success, the verification result is determined to be successful; if the second verification result indicates failure, the verification result is determined to be unsuccessful.

[0010] Optionally, processing the target subtask based on the verification result includes: If the verification result is successful, the cloud resource platform is invoked to execute the target subtask, and the task status of the target subtask in the task list is updated to complete. If the verification result is unsuccessful, a task failure message is sent to the first agent.

[0011] Optionally, the skill constraints of each skill in the skill library include at least one of the following: task type constraints, permission and identity constraints, prerequisite dependency constraints, rollback capability constraints, and data structure constraints.

[0012] Thirdly, this disclosure also provides a task processing system, which includes at least one first agent, multiple second agents, and a cloud resource platform; The first agent is used to execute the task processing method described in the first aspect of this disclosure; The second agent is used to execute the task processing method described in the second aspect of this disclosure; The cloud resource platform is used to execute the target sub-tasks assigned by the second agent.

[0013] Fourthly, this disclosure also provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the steps of the task processing method provided in this disclosure.

[0014] Fifthly, this disclosure also provides an electronic device, the electronic device comprising: processor; Memory used to store processor-executable instructions; The processor is configured as follows: The steps of performing the task processing method provided in this disclosure.

[0015] Based on the above technical solution, diverse task requests from users can be received and broken down into multiple subtasks with corresponding configuration constraints. Users can initiate operations without needing to be familiar with the underlying APIs, lowering the barrier to entry for cloud computing. Furthermore, relying on the second agent's own task processing type to automatically match subtasks breaks free from fixed orchestration constraints, enabling flexible handling of cross-resource composite tasks. Then, by implementing pre-execution constraint checks to intercept non-compliant instructions, the system avoids the risks of unauthorized access and misoperation that can result from direct instruction issuance by the agent, thus improving the intelligence and security of cloud resource computing and management.

[0016] Other features and advantages of this disclosure will be described in detail in the following detailed description section. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the present disclosure and form part of the specification. They are used together with the following detailed description to explain the present disclosure, but do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart illustrating a task processing method applied to a first agent according to an exemplary embodiment.

[0018] Figure 2 This is a schematic diagram illustrating a task flow process according to an exemplary embodiment.

[0019] Figure 3 This is a flowchart illustrating a task processing method applied to a second agent according to an exemplary embodiment.

[0020] Figure 4 This is a schematic diagram illustrating a task processing system according to an exemplary embodiment.

[0021] Figure 5 This is a block diagram illustrating a task processing apparatus applied to a first agent end according to an exemplary embodiment.

[0022] Figure 6 This is a block diagram illustrating a task processing apparatus applied to a second agent end according to an exemplary embodiment.

[0023] Figure 7 This is a block diagram illustrating a terminal-side electronic device according to an exemplary embodiment.

[0024] Figure 8 This is a block diagram illustrating a server-side electronic device according to an exemplary embodiment. Detailed Implementation

[0025] The specific embodiments of this disclosure will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit this disclosure.

[0026] In the following description, the words "first" and "second" are used only to distinguish the purpose of the description and should not be interpreted as indicating or implying relative importance or order.

[0027] As cloud computing platforms continue to expand, the management complexity of cloud resources such as virtual machines, containers, bare metal, networks, and storage continues to increase. Existing cloud control systems mainly rely on API calls, script orchestration, or predefined automated processes to achieve resource management, but the following problems still exist: First, the operation method is complex. The cloud control system has a high barrier to entry, meaning that users must master the underlying API interfaces, parameter definitions, and calling order, making it difficult to complete complex operations through high-level semantics.

[0028] Secondly, automation capabilities rely on predefined processes, and cloud control systems lack dynamic adaptability. Existing cloud control systems are usually based on fixed orchestration logic, making it difficult to cope with complex tasks across resources and systems.

[0029] Third, there is a lack of a unified task abstraction and execution control mechanism. In existing cloud control systems, user intent cannot be directly mapped into a controllable execution process, and the system lacks unified constraints and state management for the execution process.

[0030] Fourth, there are security risks in directly taking over control with AI technology. In existing solutions, if the output of large AI models is directly converted into control commands, problems such as illusions, unauthorized operations, and irreversibility may occur, making it difficult to meet the reliability requirements of the production environment.

[0031] To address the above issues, this disclosure first provides a task processing method applied to the first agent end. Figure 1 This is a flowchart illustrating a task processing method applied to a first agent according to an exemplary embodiment. See also... Figure 1 As shown, the method includes the following steps: In step S11, a task list is generated based on the task request input by the user.

[0032] The task list includes multiple subtasks, each with a different task constraint. This task constraint is used to characterize the task conditions required to execute the subtask.

[0033] In step S12, a consumption request is sent to the second agent so that the second agent can match the target subtask from multiple subtasks in the task list according to its own pre-configured task processing type, verify the task constraints of the target subtask, determine the verification result, and process the target subtask according to the verification result.

[0034] It is worth noting that both the first agent and the second agent in this disclosure are intelligent agents. An intelligent agent is a proxy that can perceive the environment and take actions to achieve a specific goal. An intelligent agent can be at least one of software, hardware, or a system. The first agent is a central scheduling agent (main agent), which is responsible for receiving user requests, task decomposition, and task distribution and scheduling. The second agent is an edge execution agent (edge ​​sub-agent), which is pre-configured with its own skills or task processing types and is responsible for task selection, constraint verification, and initiating resource implementation and execution.

[0035] In one implementation, the task list is a set of all subtasks formed by splitting the total task, which are uniformly hosted in a preprocessing queue. Task constraints are generated by user input, system permissions, and resource environment. Task constraints are used to characterize the objective conditions that must be met for the execution of subtasks. The consumption request is a task consumption notification issued by the first agent to trigger the second agent to pull the matching subtasks.

[0036] In this invention, the main agent (first agent) is responsible for intent parsing and task orchestration, and then sends RSC requests to other different types of sub-agents (second agents). AI agents communicate with each other through RSC (Remote Skill Call), that is, issuing sub-tasks with task constraints. RSC is a custom calling protocol of this invention, which relies on constraint verification to achieve controlled cross-agent skill calls. RSC is not a traditional API call or remote procedure call, but a resource control execution mechanism based on a set of constraints.

[0037] In one embodiment, the first agent first parses the user task request, splitting it into a task list consisting of several subtasks. The splitting process relies on operation type and resource attributes, with each subtask inheriting the overall task constraints and forming its own unique task constraints. Then, the first agent sends a consumption request. The second agent, based on its locally preset task processing type, filters and matches the target subtask from the task list, and verifies the task constraints carried by the subtask according to its own skill constraint rules. Only if the verification passes can the resource operation be executed; otherwise, the task is terminated. This solution achieves layered isolation between the scheduling layer and the execution layer. Task routing relies on the task processing type preset by the agent for matching, and a constraint verification process is added before task execution.

[0038] For example, if a user inputs the request "Create a cloud computing configuration virtual machine and mount a 100G cloud disk in region A", firstly, the first agent obtains a task list through intent recognition parsing. This task list includes two sub-tasks: creating a virtual machine and mounting a cloud disk. The constraints for the virtual machine sub-task include region, instance size, and VPC (Virtual Private Cloud) information, while the constraints for the disk mounting sub-task include instance ID (Identity / Identification) and disk capacity. Then, the first agent issues a consumption request, causing the second agent (for computing) to match the sub-task "Create Virtual Machine" and the second agent (for storage) to match the sub-task "Mount Disk". Each second agent uses its built-in skill rules to verify the corresponding task constraints and completes resource creation after the verification results indicate that the parameters are compliant.

[0039] Based on the above technical solution, diverse task requests from users can be received and broken down into multiple subtasks with corresponding configuration constraints. Users can initiate operations without needing to be familiar with the underlying APIs, lowering the barrier to entry for cloud computing. Furthermore, relying on the second agent's own task processing type to automatically match subtasks breaks free from fixed orchestration constraints, enabling flexible handling of cross-resource composite tasks. Then, by implementing pre-execution constraint checks to intercept non-compliant instructions, the system avoids the risks of unauthorized access and misoperation that can result from direct instruction issuance by the agent, thus improving the intelligence and security of cloud resource computing and management.

[0040] Optionally, the task constraints include at least one of resource type constraints, permission and identity constraints, prerequisite dependency constraints, and rollback requirement constraints.

[0041] It is worth noting that resource type constraints are used to verify whether the cloud resource platform has idle resources, that is, the hard requirements for the type, specifications, availability zone, and resource reserves of cloud resources for task execution; permission and identity constraints are the scope of operation permissions defined based on the identity of the initiating user; prerequisite dependency constraints are used to verify whether the dependent resources have been created and are ready before the task is executed; and rollback requirement constraints are the business rules that require triggering a rollback to restore the original environment when the task fails abnormally.

[0042] In one implementation, the task constraints bound to the subtask are extracted from user requirements and real-time resource environment, and any one or more of the four types of constraints can be configured as needed. During the verification phase, the second agent uses its own preset skill constraint rules to check the above task constraints item by item. If any one of the verifications fails, execution is rejected, thus achieving multi-dimensional execution access control.

[0043] Using the example of creating a virtual machine and mounting a disk as described above, the task constraints include the current user's computing creation permissions (permission constraints), available cloud computing configuration resources in region A (resource constraints), the target VPC and subnet already existing (prerequisite constraints), and automatic destruction of generated resources if creation fails (rollback constraints). Mounting is performed after all constraints are verified to be correct.

[0044] By combining and verifying multiple dimensions of constraints, illegal execution is intercepted from multiple levels, including resources, permissions, dependencies, and fault fallback. This reduces task anomalies caused by unauthorized operations, insufficient resources, and missing dependencies, thereby improving the reliability of cloud resource management and business security.

[0045] Optionally, after the first agent distributes the subtasks to the second agents, the completion rate of the subtasks can be used to determine whether the task is completed. This includes the following steps: The first step is for the first agent to check the completion status of each subtask in the task list according to a preset cycle, and determine the completion rate of each subtask based on its completion status.

[0046] The second step is to determine the global task is complete and output the task result if the subtask completion rate is greater than or equal to the preset completion rate. The task result is determined based on the completed subtasks. If the subtask completion rate is less than the preset completion rate, the global task is determined to have failed.

[0047] Among them, the preset period is the polling interval time pre-configured by the first agent, which is used to periodically check the progress of the task; the subtask completion rate is the ratio of the number of subtasks that have been verified to be successfully executed to the total number of all subtasks; the preset completion rate is a customizable judgment threshold, preferably 50%, which serves as the judgment benchmark for the success or failure of the global task.

[0048] It's worth noting that after a subtask is handed over to the second agent, the first agent periodically checks the execution status of all subtasks and calculates the completion rate. The completion rate is compared to a preset threshold: if the threshold is met, the overall task is considered successful, and the data from each successful subtask is aggregated to generate the final result; if the threshold is not met, the global task is considered failed, and an exception rollback can be triggered as needed. Distributed task consistency management is achieved through threshold arbitration.

[0049] For example, the overall task is broken down into two subtasks: creating a virtual machine and mounting a cloud disk, with a preset completion rate of 50%. If both subtasks are executed successfully, the completion rate is 100% ≥ 50%, the global task is successful, and the instance ID and disk mounting information are summarized as the result; if only one subtask is successful, the completion rate is still 50%, which is considered a success; if no task is completed, the global task fails.

[0050] In another implementation, see Figure 2 As shown, after receiving the new Task_id (task identifier), the first agent determines that a new user request has been generated (corresponding to...). Figure 2 (NEW in the context), where the task identifier is a string or value used in computer systems and software development to uniquely identify a specific task, process, or job. The first agent first generates the task information of the total task from the client request. This total task includes multiple subtasks, and then inserts the total task into the preprocessing task queue (corresponding to...). Figure 2 (Preprocessing in the process). Then, it triggers task consumption requests (corresponding to) to all its hosted sub-agents (secondary agents). Figure 2 (The second agent consumes the preprocessing task queue). After receiving the request, the second agent consumes the preprocessing task queue, translates the task information into a command format acceptable to the system, and then sends it out. After the system's checking module confirms successful sending, it transfers the corresponding task from the preprocessing queue to the postprocessing task queue (corresponding to...). Figure 2 (Post-processing in the process). The first agent periodically checks the subtasks in the post-processing task queue. If the total number of successful subtasks exceeds 50% of the number of nodes in the distributed cluster (e.g., the total number of all subtasks in the post-processing task queue) (adjustable), then the user's total task request is successful, and the data information corresponding to the total task is written to the resource object for persistent storage.

[0051] By using the above method, even if a small number of subtasks fail, the entire task will not be rendered useless, thus improving the system's scheduling fault tolerance and task closed-loop management capabilities.

[0052] This disclosure also provides a task processing method applied to a second agent. Figure 3 This is a flowchart illustrating a task processing method applied to a second agent according to an exemplary embodiment. See also... Figure 3 As shown, the method includes the following steps: In step S31, upon receiving a consumption request from the first agent, the target subtask is matched from multiple subtasks in the task list according to the task processing type pre-configured by the second agent.

[0053] The task list is generated by the first agent based on the task request input by the user. The task list includes multiple subtasks, each of which corresponds to different task constraints. The task constraints are used to characterize the task conditions required to execute the subtask.

[0054] In step S32, the task constraints of the target subtask are verified, and the verification result is determined.

[0055] In step S33, the target subtask is processed based on the verification result.

[0056] In one implementation, after receiving a consumption request, the second agent first filters the target subtasks from the task list issued by the first agent based on its own preset task processing type. Then, it retrieves the skill constraints preset locally on the second agent and verifies the resource, permission, and pre-dependency constraints attached to the target subtask item by item, outputting the verification result of pass or fail. Finally, if the verification passes, the corresponding resource operation is executed; if the verification fails, the task execution is terminated and the exception is reported upwards, thereby realizing a layered chain of task-oriented distribution, pre-verification, and controlled execution.

[0057] By employing the above methods, sub-tasks are accurately matched based on the business type, eliminating cross-type error handling issues; pre-execution constraint checks intercept non-compliant instructions in advance, avoiding illegal resource operations, thus improving the operational security and resource scheduling accuracy of the cloud resource platform.

[0058] Optionally, the second agent is pre-configured with a skill library, which includes multiple skills. Each skill corresponds to a different skill identifier and skill constraints, whereby the skill constraints are used to characterize the conditions required to execute the skill.

[0059] It is worth noting that the skill library is a collection of skills pre-deployed locally on the second agent end, which is used to store all the business capabilities that this agent can execute; the skill identifier is the unique code of each skill; the skill constraint is the execution access rule inherent to the skill itself, which is used to specify the parameters, permissions, applicable scenarios and other judgment criteria for the tasks that the skill can accept.

[0060] A Skill is an execution unit that encapsulates a single cloud resource operation. This execution unit has built-in input / output specifications and skill constraints. It is not a simple function call unit, but an execution capability unit with constraint descriptions. Each Skill includes at least a parameter structure template, such as adding input_schema (input constraints, see the skill details definition below) to the metadata (metadata used to store basic skill identification information) in Skill.md (skill description document); an output structure definition, such as adding output_schema (output constraints, see the skill details definition below) to the metadata in Skill.md; and a rollback policy description.

[0061] Currently, Skill execution itself is highly uncertain, relying on explicit parameter input from the user. To enhance determinism, this disclosure adds `input_schema` and `output_schema` to the YAML formatter (a YAML format file used to uniformly describe skill metadata and ability specifications) based on the current Skill specification definition. Example code is shown below: input_schema: type: object # Fixed: The input is always an object description: "Input parameters required for this skill" # Optional, skill description properties: field_name_1: type: string# Supports: string, number, integer, boolean, array, object Description: "A description of field 1 to help the LLM understand the meaning of this parameter." # Optional constraints (select as needed): # enum: [value1, value2]# Limit the enumeration value # minLength: 2 # Minimum string length # maxLength: 100 # Maximum string length # pattern: "^[az]+$" # Regular expression matching # default: "default value" # default value # examples: ["Example 1", "Example 2"]# Example values ​​(for LLM reference) output_schema: type: object # Fixed: The output is always an object (even if it's a single return value, it's wrapped in an object). Description: "The return result structure of this skill" properties: status: type: string Description: "Task execution status" enum: [success, partial, failed] message: type: string Description: "Status description information, which can be freely organized by LLM." data: type: object # Core business data is encapsulated in a data object Description: "Core returned data" properties: # The actual business fields should be written here, for example: # result_id: #type: string # result_list: #type: array #items: #type: object #properties: ... In one embodiment, step S32 above includes the following implementation: Step 1: Read the skill identifier of the target subtask.

[0062] Step 2: Match the skill identifiers of multiple skills in the skill library with the skill identifiers of the target subtask to determine the first verification result.

[0063] Step 3: If the first verification result indicates failure, then determine the verification result as unsuccessful.

[0064] If the first verification result indicates success, firstly, the target skill that successfully matches the target subtask is determined. Then, the task constraints of the target subtask are verified based on the skill constraints of the target skill to determine the second verification result.

[0065] If the second verification result indicates that the verification is passed, then the verification result is determined to be passed; if the second verification result indicates that the verification is failed, then the verification result is determined to be failed.

[0066] In one implementation, S32 is divided into two levels of verification: First, the skill identifier carried by the target subtask is extracted and compared with the local skill identifier for a first-level verification. If the identifier is not found, the verification fails directly. After the identifier is successfully matched, the corresponding target skill is retrieved, and the task constraints attached to the subtask are verified item by item using the skill constraints that the skill carries. A second-level verification result is formed, and the final verification status is determined by combining the two levels of conclusions.

[0067] For example, the virtual machine creation subtask carries a first identifier, `vm_create`, and the computation agent skill library contains this first identifier and its associated skill constraints (region format, instance specification range, and user permissions). If the identifier matches successfully, the skill constraints are used to verify the subtask constraints. If the verification result shows that the region format, instance specification, and user permissions all comply with the rules, the second-level verification passes, and the task is allowed to execute. If the instance specification exceeds the skill limit, the second-level verification fails, and the task is rejected.

[0068] By employing a two-level, step-by-step verification process, tasks without corresponding processing capabilities are quickly eliminated through identification, reducing invalid verification overhead. Then, task conditions are verified based on skill constraints, achieving layered control over execution access. This further avoids task mismatch, parameter violations, and illegal operations beyond the scope of skill capabilities, thereby improving verification efficiency and system security.

[0069] Optionally, step S33 above includes the following implementation: If the verification result is successful, the cloud resource platform is invoked to execute the target subtask, and the task status of the target subtask in the task list is updated to complete.

[0070] Among them, the cloud resource platform is the underlying management platform that carries physical resources such as virtual machines, storage, and networks, and is responsible for implementing actual resource creation and modification operations.

[0071] If the verification result is unsuccessful, a task failure message is sent to the first agent.

[0072] For example, if both levels of verification for the virtual machine creation subtask pass, the compute agent calls the cloud platform interface to create the instance and marks the subtask as completed; if the parameter format does not conform to the skill constraints, the verification fails, and the storage agent immediately reports the mount task failure to the first agent and does not initiate disk mount.

[0073] Using the above methods, resource calls are directly blocked if the verification fails, avoiding invalid requests from occupying platform resources; the execution results are updated and reported in real time, which facilitates the first agent to summarize the status, calculate the completion rate, and improve the closed-loop management of the entire task chain.

[0074] Optionally, the skill constraints of each skill in the skill library include at least one of the following: task type constraints, permission and identity constraints, prerequisite dependency constraints, rollback capability constraints, and data structure constraints.

[0075] This disclosure also provides a task processing system, which includes at least one first agent, multiple second agents, and a cloud resource platform.

[0076] In this system, both the first agent and the second agent are intelligent agents in the field of artificial intelligence, and the cloud resource platform is used to execute the target sub-tasks assigned by the second agent.

[0077] In one implementation, see Figure 4 As shown, the task processing system includes a user interaction module, an intent parsing and task modeling module, an RSC module, a Skill management module, an execution control module, a result feedback module, and a cloud resource interface module.

[0078] 1. The user interaction module is used to receive task requests from users or external systems, where task requests can be natural language, semi-structured text, or API requests.

[0079] 2. The intent parsing and task modeling module is used to convert user input into a unified task object. This includes at least: task objective, resource object, execution constraints, and context information, and then converts the task object into an RSC call.

[0080] 3. The RSC module uses RSC technology to enable interaction between different modules, allowing the caller to obtain the callee's Skill list, interface, and boundary conditions through the RSC interface. This module is used to build constraint sets, verify constraints, and trigger Skill execution.

[0081] 4. The Skill management module is used to maintain the skill set. Each skill corresponds to a minimum controllable execution unit. Skills include not only specific actions, but also input parameter definitions, preconditions, output result definitions, rollback policies, and permission requirements.

[0082] 5. The execution control module is used to uniformly control the skill execution process. It is responsible for pre-execution verification (permissions, parameters, etc.), execution process control (sequence, retry, termination), and post-execution status recording.

[0083] 6. The results feedback module is used to confirm the skill execution results and determine whether the task has been truly completed. Specifically, it is responsible for status confirmation, exception handling, and results feedback.

[0084] 7. Cloud Resource Interface Module: Used to interact with the data warehouse and complete specific resource operations. Interaction methods include API, SDK, and command-line interface.

[0085] In this invention, AI agents are assigned roles, with the main agent (first agent) responsible for intent parsing and task orchestration, and then issuing RSC requests to other sub-agents (second agents). AI agents communicate with each other via RSC to invoke skills. The central agent is only responsible for intent parsing and task assignment, while the edge agents are responsible for handing over the specific operations to the Skill management module and execution control module (i.e., calling the cloud resource platform) to complete the task execution. In other words, the AI ​​agents belong to the planning layer, and the cloud resource platform belongs to the execution layer; the two are connected through structured task objects and skill definitions.

[0086] See Figure 4 As shown, during system operation, after a user submits a task request, it first enters the intent parsing and task modeling module, forming a unified task object. Then, different subtasks are distributed to different sub-AI agents via RSC calls. Upon receiving a subtask, the sub-AI agent's execution control module matches the corresponding skill from the Skill management module, performs constraint checks such as permission verification and boundary checks, and then executes the corresponding skill. Finally, the execution result is handed over to the result inspection module. The result feedback module decides whether to continue subsequent operations or trigger a rollback. The inspection module, acting as a bypass AI agent, performs global environment consistency checks, error correction, and alarms. The entire process forms a closed loop from task input to execution feedback.

[0087] In one embodiment, the role definition of the RSC interface on the agent side includes a skill consumer and a skill provider. The skill consumer, such as an agent or collaborative document, is responsible for initiating the call request, while the skill provider is used to host and execute the specific skill logic and return the result.

[0088] RSC communication protocol mandates the use of HTTPS (Hypertext Transfer Protocol Secure) and prohibits plaintext HTTP (Hypertext Transfer Protocol). Character encoding is uniformly UTF-8, and the request / response body defaults to application / json, while streaming responses use text / event-stream. RSC interfaces use the HTTP protocol for transmission and employ the standard Bearer Token authentication mechanism.

[0089] The following examples will be described in further detail.

[0090] Implementation Example 1 (Virtual Machine Instance Creation and Initialization): Users input their cloud server creation requests via natural language, such as creating a GPU instance for training tasks and mounting a specified data disk. The first agent in the intelligent agent first recognizes the user's input intent, generates a task object including fields such as instance specifications, availability zone, image, network, storage, and security group, and initiates RSC calls to multiple second agents within the agent. These second agents then receive their corresponding subtasks. Each second agent selects the appropriate skill based on its assigned subtask and performs constraint checks. Once the checks pass, it initiates an RSC call to the resource control system. The resource control system performs quota checks, image availability checks, network connectivity checks, and permission checks before executing instance creation, cloud disk mounting, network configuration, and initialization script distribution. After task execution, the system returns the instance ID, IP address, mounting result, and initialization status. The intelligent agent then provides the results back to the user in natural language.

[0091] Implementation Example 2 (Network Changes for Bare Metal Servers): The user inputs "Migrate a bare metal server from the current business network to a new VPC subnet while ensuring uninterrupted service." The first agent in the intelligent agent parses this requirement as a change task with strong constraints and explicitly records task constraints such as the source network, target network, migration window, rollback policy, and interruption tolerance in the task object. This requires the task processing system (including the second agent and cloud resource platform) to verify the available addresses, routing table, NAT (Network Address Translation) / gateway reachability, and dependencies between the target network and existing services before task execution. After confirming that the task constraints have passed verification, the task processing system executes the network switching task according to predefined steps, such as updating the virtual switch or gateway forwarding table, refreshing the ARP (Address Resolution Protocol) / neighbor table, synchronizing routing policies, and performing connectivity probes and service health checks after each step. If any step fails, a rollback process is automatically triggered, restoring the server to its original network configuration and recording the reason for failure and recovery status.

[0092] In another embodiment, the specific implementation of the task processing system during task execution can be chosen in several ways, for example: When a Skill call is triggered during task modeling, you can choose to actively orchestrate the call to the RSC interface. This can be achieved by triggering an LLM (Large Language Model) inference request and by injecting the Skill description into the system context for inference by the LLM system.

[0093] In terms of task routing, RSC endpoints can be proactively defined, or a skill-based subscription and publish model can be implemented through a centralized Skill router.

[0094] Regarding task status transitions: A synchronous waiting approach can be used, returning the result only after the task succeeds. An asynchronous calling approach can be used to report the processing result.

[0095] See Figure 5 As shown, this disclosure also provides a task processing device 210, which is applied to a first agent end and includes: The generation module 211 is configured to generate a task list based on a task request input by a user. The task list includes multiple sub-tasks, each of which corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the sub-task. The sending module 212 is configured to send a consumption request to the second agent, so that the second agent, according to its own pre-configured task processing type, matches the target subtask from multiple subtasks in the task list, verifies the task constraints of the target subtask, determines the verification result, and processes the target subtask according to the verification result.

[0096] Optionally, the task constraints include at least one of resource type constraints, permission and identity constraints, prerequisite dependency constraints, and rollback requirement constraints.

[0097] Optionally, the sending module 212 is also configured to: The completion status of each subtask in the task list is checked according to a preset cycle, and the completion rate of each subtask is determined. If the completion rate of the subtask is greater than or equal to the preset completion rate, the global task is determined to be completed and the task result is output. The task result is determined based on the completed subtasks. If the completion rate of the subtask is less than the preset completion rate, the global task is determined to have failed.

[0098] See Figure 6 As shown, this disclosure also provides a task processing device 220, which is applied to a second agent end and includes: The receiving module 221 is configured to, upon receiving a consumption request sent by the first agent, match a target subtask from multiple subtasks in a task list according to a task processing type pre-configured by the second agent. The task list is generated by the first agent based on a task request input by the user. The task list includes multiple subtasks, each of which corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the subtask. Verification module 222 is configured to verify the task constraints of the target subtask and determine the verification result; Processing module 223 is configured to process the target subtask based on the verification result.

[0099] Optionally, the second agent is pre-configured with a skill library, which includes multiple skills. Each skill corresponds to a different skill identifier and skill constraint, and the skill constraint is used to characterize the conditions required to execute the skill.

[0100] The verification module 222 is further configured to: The skill identifier of the target subtask is obtained by reading it; The skill identifiers of multiple skills in the skill library are matched with the skill identifiers of the target sub-task to determine the first verification result; If the first verification result indicates failure, the verification result is determined to be unsuccessful. If the first verification result indicates success, a target skill that successfully matches the target subtask is determined, and the task constraints of the target subtask are verified according to the skill constraints of the target skill to determine a second verification result. If the second verification result indicates success, the verification result is determined to be successful; if the second verification result indicates failure, the verification result is determined to be unsuccessful.

[0101] Optionally, the processing module 223 is further configured to: If the verification result is successful, the cloud resource platform is invoked to execute the target subtask, and the task status of the target subtask in the task list is updated to complete. If the verification result is unsuccessful, a task failure message is sent to the first agent.

[0102] Optionally, the skill constraints of each skill in the skill library include at least one of the following: task type constraints, permission and identity constraints, prerequisite dependency constraints, rollback capability constraints, and data structure constraints.

[0103] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0104] This disclosure also provides a task processing system, which includes at least one first agent, multiple second agents, and a cloud resource platform.

[0105] The cloud resource platform is used to execute the target sub-tasks assigned by the second agent.

[0106] This disclosure also provides a computer-readable storage medium having computer program instructions stored thereon, which, when executed by a processor, implement the task processing method provided in this disclosure.

[0107] This disclosure also provides an electronic device, the electronic device comprising: processor; Memory used to store processor-executable instructions; The processor is configured as follows: Perform the task processing methods provided in this disclosure.

[0108] Figure 7 This is a block diagram illustrating an electronic device 700 according to an exemplary embodiment. Figure 7As shown, the electronic device 700 may include a processor 701 and a memory 702. The electronic device 700 may also include one or more of a multimedia component 703, an input / output (I / O) interface 704, and a communication component 705.

[0109] The processor 701 controls the overall operation of the electronic device 700 to complete all or part of the steps in the task processing method described above. The memory 702 stores various types of data to support the operation of the electronic device 700. This data may include, for example, instructions for any application or method operating on the electronic device 700, and application-related data. The memory 702 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The I / O interface 704 provides an interface between the processor 701 and other interface modules, such as a keyboard, mouse, buttons, etc. These buttons can be virtual or physical buttons. Communication component 705 is used for wired or wireless communication between the electronic device 700 and other devices. Wireless communication includes, for example, Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or one or more combinations thereof; therefore, the corresponding communication component 705 may include: a Wi-Fi module, a Bluetooth module, or an NFC module.

[0110] In an exemplary embodiment, the electronic device 700 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the task processing method described above.

[0111] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the task processing method described above. For example, the computer-readable storage medium may be the memory 702 including program instructions described above, which may be executed by the processor 701 of the electronic device 700 to complete the task processing method described above.

[0112] In another exemplary embodiment, a computer program product is also provided, which includes a computer program executable by a processor, which, when executed by the processor, implements the steps of the task processing method described above.

[0113] Figure 8 This is a block diagram illustrating an electronic device 800 according to an exemplary embodiment. For example, the electronic device 800 may be provided as a server. (Refer to...) Figure 8 The electronic device 800 includes a processor 822, which may be one or more, and a memory 832 for storing computer programs executable by the processor 822. The computer programs stored in the memory 832 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processor 822 may be configured to execute the computer program to perform the task processing methods described above.

[0114] Additionally, the electronic device 800 may also include a power supply component 826 and a communication component 850. The power supply component 826 can be configured to perform power management of the electronic device 800, and the communication component 850 can be configured to enable communication of the electronic device 800, such as wired or wireless communication. Furthermore, the electronic device 800 may also include an input / output (I / O) interface 858. The electronic device 800 can operate on an operating system, such as Windows Server, stored in memory 832.TM Mac OSX TM Unix TM Linux TM etc.

[0115] In another exemplary embodiment, a computer-readable storage medium including program instructions is also provided, which, when executed by a processor, implement the steps of the task processing method described above. For example, the computer-readable storage medium may be the memory 832 including the program instructions described above, which may be executed by the processor 822 of the electronic device 800 to complete the task processing method described above.

[0116] In another exemplary embodiment, a computer program product is also provided, which includes a computer program executable by a processor, which, when executed by the processor, implements the steps of the task processing method described above.

[0117] The preferred embodiments of this disclosure have been described in detail above with reference to the accompanying drawings. However, this disclosure is not limited to the specific details of the above embodiments. Within the scope of the technical concept of this disclosure, various simple modifications can be made to the technical solutions of this disclosure, and these simple modifications all fall within the protection scope of this disclosure.

[0118] It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. In order to avoid unnecessary repetition, this disclosure will not describe the various possible combinations separately.

[0119] Furthermore, various different embodiments of this disclosure can be combined in any way, as long as they do not violate the spirit of this disclosure, they should also be regarded as the content disclosed in this disclosure.

Claims

1. A task processing method, characterized in that, Applied to the first agent, the method includes: Based on the task request input by the user, a task list is generated. The task list includes multiple sub-tasks, and each sub-task corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the sub-task. A consumption request is sent to the second agent so that the second agent can match the target subtask from multiple subtasks in the task list according to its own pre-configured task processing type, verify the task constraints of the target subtask, determine the verification result, and process the target subtask according to the verification result.

2. The task processing method according to claim 1, characterized in that, The task constraints include at least one of the following: resource type constraints, permission and identity constraints, prerequisite dependency constraints, and rollback requirement constraints.

3. The task processing method according to claim 1, characterized in that, The method further includes: The completion status of each subtask in the task list is checked according to a preset cycle, and the completion rate of each subtask is determined based on its completion status. If the completion rate of the subtask is greater than or equal to the preset completion rate, the global task is determined to be completed and the task result is output. The task result is determined based on the completed subtasks. If the completion rate of the subtask is less than the preset completion rate, the global task is determined to have failed.

4. A task processing method, characterized in that, Applied to the second agent, the method includes: Upon receiving a consumption request from the first agent, the target subtask is matched from multiple subtasks in the task list according to the task processing type pre-configured by the second agent. The task list is generated by the first agent based on the task request input by the user. The task list includes multiple subtasks, each of which corresponds to a different task constraint. The task constraint is used to characterize the task conditions required to execute the subtask. The task constraints of the target subtask are verified, and the verification result is determined. The target subtask is processed based on the verification results.

5. The task processing method according to claim 4, characterized in that, The second agent is pre-configured with a skill library, which includes multiple skills. Each skill corresponds to a different skill identifier and skill constraint. The skill constraint is used to characterize the conditions required to execute the skill. The step of verifying the task constraints of the target subtask and determining the verification result includes: The skill identifier of the target subtask is obtained by reading it; The skill identifiers of multiple skills in the skill library are matched with the skill identifiers of the target sub-task to determine the first verification result; If the first verification result indicates failure, the verification result is determined to be unsuccessful. If the first verification result indicates success, a target skill that successfully matches the target subtask is determined, and the task constraints of the target subtask are verified according to the skill constraints of the target skill to determine a second verification result. If the second verification result indicates success, the verification result is determined to be successful; if the second verification result indicates failure, the verification result is determined to be unsuccessful.

6. The task processing method according to claim 5, characterized in that, The step of processing the target subtask based on the verification result includes: If the verification result is successful, the cloud resource platform is invoked to execute the target subtask, and the task status of the target subtask in the task list is updated to complete. If the verification result is unsuccessful, a task failure message is sent to the first agent.

7. The task processing method according to claim 5, characterized in that, The skill constraints for each skill in the skill library include at least one of the following: task type constraints, permission and identity constraints, prerequisite dependency constraints, rollback capability constraints, and data structure constraints.

8. A task processing system, characterized in that, The task processing system includes at least one first agent, multiple second agents, and a cloud resource platform. The first agent is used to execute the task processing method according to any one of claims 1-3; The second agent is used to execute the task processing method according to any one of claims 4-7; The cloud resource platform is used to execute the target sub-tasks assigned by the second agent.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When executed by the processor, the program instructions implement the steps of the task processing method according to any one of claims 1-7.

10. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store processor-executable instructions; The processor is configured as follows: Perform the steps of the task processing method according to any one of claims 1-7.

Images