Intelligent Software Engineering Implementation Methodology Based on AI Skills Orchestration Workflow

Abstract

This invention discloses an intelligent software engineering implementation method based on AI skill orchestration workflow, belonging to the field of software engineering automation technology. It involves constructing multiple standardized AI skill components, combining these components into workflow templates according to preset orchestration rules, and storing them in a local template library. Project requirement information is obtained, and workflow templates are matched based on these requirements. AI model instances are dynamically allocated to the standardized AI skill components within the workflow templates to generate software engineering workflow instances. Human-machine collaborative review data at key nodes of the software engineering workflow instances is obtained. This invention dynamically allocates mapping scenarios, mapping AI model instances as initial points and component execution processes as growth chains. It pauses, resumes, and switches initial points for parallel tasks based on the growth chain height, reducing imbalances in parallel branch execution and improving model resource utilization and task continuity.

Description

Technical Field

[0001] This invention relates to the field of software engineering automation technology, and more specifically to an intelligent software engineering implementation method based on AI skill-based workflow orchestration. Background Technology

[0002] Intelligent software engineering implementation of AI skills orchestration workflow refers to encapsulating artificial intelligence capabilities such as code generation, code annotation, code review, and test generation into standardized components, and combining them into workflow templates according to orchestration rules such as serial, parallel, or conditional branching. After receiving project requirements, the templates are automatically matched, AI model instances are assigned, and executed to assist in completing tasks such as code output, quality inspection, and collaborative review in the software development process.

[0003] Currently, some software engineering implementation methods often rely on manual breakdown of requirements, selection of tools, calling of models, and connection of various development stages. There is a lack of unified encapsulation and process orchestration between different AI capabilities, and the allocation of AI model instances often adopts a fixed calling method, which makes it difficult to dynamically switch according to the execution progress and resource status, which can easily lead to problems such as imbalance of parallel tasks and low resource utilization. Summary of the Invention

[0004] The purpose of this invention is to provide an intelligent software engineering implementation method based on AI skill-based workflow orchestration, in order to address the shortcomings in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: an intelligent software engineering implementation method based on AI skill orchestration workflow, comprising the following steps: Build multiple standardized AI skill components, combine these components into workflow templates using preset orchestration rules, and store them in a local template library; Obtain project requirements information, match workflow templates based on project requirements information, and dynamically allocate AI model instances to the standardized AI skill components in the workflow templates to generate software engineering workflow instances; Obtain human-machine collaborative review data at key nodes of software engineering workflow instances, inject the human-machine collaborative review data into AI skill standardization components, and adjust the AI ​​skill standardization components in real time.

[0006] In a preferred embodiment, the step of constructing multiple standardized AI skill components and combining these components into a workflow template according to preset orchestration rules includes: AI skills such as code generation, code commenting, and code review are standardized and encapsulated to obtain multiple standardized AI skill components. Based on the input and output parameters of each AI skill standardization component, determine the data transmission relationship between different AI skill standardization components, and based on the execution conditions of each AI skill standardization component, determine the execution dependency relationship between different AI skill standardization components; Based on execution dependencies, multiple standardized AI skill components are combined according to preset orchestration rules to generate a workflow template. The preset orchestration rules include serial orchestration rules, parallel orchestration rules, and conditional branch orchestration rules.

[0007] In a preferred embodiment, the step of dynamically allocating AI model instances to the AI ​​skill standardization components in the workflow template to generate software engineering workflow instances includes: A dynamic allocation mapping scenario is built based on a cloud server, and multiple AI model instances are used as initial points and mapped side by side in the dynamic allocation mapping scenario; In a dynamic allocation mapping scenario, multiple growth chains are configured, and standardized AI skill components are assigned to different growth chains. Different growth chains are then connected to different initial points. A growth chain includes a root node, an active body, and a confluence point. The height of the growth chain is assigned based on the state generated by the AI ​​model instance; Connection between the growth chain and the initial point under the parallel orchestration rule of height switching of the growth chain.

[0008] In a preferred embodiment, the step of assigning AI skill standardization components to different growth chains and connecting the different growth chains to different starting points includes: The standardized AI skills components are mapped to the root node of the growth chain for storage. The root node is the starting point of the growth chain and is used to carry the call address of the currently executed standardized AI skills component. Based on the call address, the API interface in the AI ​​skill standardization component is obtained, a call request is sent to the initial point, and a connection is established.

[0009] In a preferred embodiment, the step of assigning a height to the growth chain based on the state generated by the AI ​​model instance includes: Obtain the output data volume generated by the current AI skill standardization component. The output data volume includes the number of generated lines of code, the number of generated comment characters, and the number of generated test cases. The output data volume is converted into the growth height of the growth chain according to the preset mapping rules, and the growth height of the growth chain is updated in real time.

[0010] In a preferred embodiment, the step of connecting the growth chain and the initial point under the growth chain height switching parallel orchestration rule includes: A height sensor is configured at the convergence point of each growth chain. The height sensor obtains the current growth height of each growth chain under the parallel orchestration rule in real time. By comparing the growth height with a preset difference threshold, when the growth height between each growth chain exceeds the preset difference threshold, the growth chain with the lowest growth height is switched to an idle initial point.

[0011] In a preferred embodiment, the step of switching the growth chain with the lowest growth height to an idle initial point when the growth height between each growth chain exceeds a preset difference threshold includes: When the growth height of the highest growth chain exceeds the growth height of the lowest growth chain, the following occurs: The root node of the lowest growth chain grows a root chain that connects to the free initial point, and the copy of the lowest growth chain is transferred to the free initial point through the root chain. When the replicant is within the difference threshold, the lowest growth chain is switched to the initial point where the replicant is located.

[0012] In a preferred embodiment, the steps of acquiring human-machine collaborative review data at key nodes of the software engineering workflow instance, injecting the human-machine collaborative review data into the AI ​​skill standardization component, and adjusting the AI ​​skill standardization component in real time include: Key nodes are preset in the workflow template. When the software engineering workflow instance reaches the key node, automatic execution is paused and the intermediate products generated by the current AI skill standardization component are collected and pushed to the review terminal. Structured audit feedback data is obtained through the audit terminal. The structured audit feedback data includes audit decision results and remediation suggestions. The structured review comments data are parsed into adjustment parameters, which can be used to adjust the input parameters of the current AI skill standardization components or revert to a specified historical node for re-execution when the review decision result is rejection; Record structured review comments and corresponding adjustment operations, and store them in the review log database.

[0013] The technical effects and advantages provided by the present invention in the above technical solution are as follows: 1. This invention dynamically allocates mapping scenarios, maps AI model instances as initial points, maps component execution processes as growth chains, and pauses, resumes, and switches initial points for parallel tasks based on the height of the growth chain. This can reduce the imbalance in parallel branch execution, improve model resource utilization, and enhance task continuity. 2. This invention encapsulates AI skills such as code generation, code annotation, and code review into standardized AI skill components, and determines data transmission relationships and execution dependencies based on input and output parameters and execution conditions. This enables the formation of a well-structured and reusable workflow template, thereby improving the automation and standardization of software engineering processes. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a flowchart of the method of the present invention.

[0016] Figure 2 This is a logic block diagram for the software engineering implementation of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] For examples, please refer to Figure 1 and Figure 2 As shown in this embodiment, the intelligent software engineering implementation method based on AI skill orchestration workflow includes the following steps: S1. Construct multiple standardized AI skill components, combine these components into workflow templates using preset orchestration rules, and store them in a local template library. S2. Obtain project requirements information, match workflow templates based on project requirements information, and dynamically allocate AI model instances to the standardized AI skill components in the workflow templates to generate software engineering workflow instances. S3. Obtain human-machine collaborative review data at key nodes of the software engineering workflow instance, and inject the human-machine collaborative review data into the AI ​​skill standardization component, and adjust the AI ​​skill standardization component in real time.

[0019] As described in steps S1-S3 above, complex software engineering activities (such as code generation, code commenting, and code review) are deconstructed into standardized AI skill components with appropriate granularity, and workflow templates are formed through rule-based orchestration. During the execution phase, a "dynamic allocation mapping scenario" is built on a cloud server. Utilizing the topological connections and highly dynamic adjustment mechanism of the "growth chain" and "initial point," intelligent and balanced scheduling of AI model instances is achieved. Combined with a human-machine collaborative review mechanism at key nodes, the high quality and closed-loop correction of the generated software engineering outputs are ensured.

[0020] In one embodiment, step S1, which involves constructing multiple standardized AI skill components and combining these components into a workflow template using preset orchestration rules, includes: S11. Standardize and encapsulate AI skills such as code generation, code commenting, and code review to obtain multiple standardized AI skill components; S12. Based on the input and output parameters of each AI skill standardization component, determine the data transmission relationship between different AI skill standardization components, and based on the execution conditions of each AI skill standardization component, determine the execution dependency relationship between different AI skill standardization components. S13. Based on the execution dependencies, combine multiple standardized AI skill components according to preset orchestration rules to generate a workflow template. The preset orchestration rules include serial orchestration rules, parallel orchestration rules, and conditional branch orchestration rules.

[0021] As described in steps S11-S13 above, AI skills such as code generation, code commenting, and code review are standardized and encapsulated to obtain multiple standardized AI skill components. Standardization encapsulation includes: defining a unified input parameter format and output parameter format for each standardized AI skill component; declaring the input parameter format as the component input parameter definition using JSON Schema; and defining the output parameter format as JSON... The schema is declared as the component output parameter definition; and an API call interface is configured for each AI skill; based on the input parameter definition and output parameter definition in each AI skill standardized component, the field names and data types of the output parameters of the first component and the input parameters of the second component are matched. When the output parameters of the first component and the input parameters of the second component match, a data transfer relationship is determined between the first and second components; based on the execution conditions of each AI skill standardized component, when the execution conditions of the second component declare that it can only start after the first component has been executed, an execution dependency relationship is determined between the first and second components; when the execution dependency is a hard dependency, a serial orchestration rule is assigned; when the execution dependency is a no-dependency dependency, a parallel orchestration rule is assigned; when the component output parameter contains a condition judgment field, a conditional branch orchestration rule is assigned; the combined components and orchestration rules are serialized into a structured template file and stored in the local template library.

[0022] In one embodiment, step S2, which involves dynamically allocating AI model instances to the AI ​​skill standardization components in the workflow template to generate a software engineering workflow instance, includes: S21. Based on the cloud server, a dynamic allocation mapping scenario is constructed, and multiple AI model instances are mapped side by side in the dynamic allocation mapping scenario as initial points (a model service node identifier that can be connected to the AI ​​skill standardization component). S22. In the dynamic allocation mapping scenario, configure multiple growth chains, assign AI skill standardization components to different growth chains, and connect different growth chains to different initial points. The growth chain includes a root node, an activity body, and a confluence point (the growth chain is a logical state object used to represent the execution process of an AI skill standardization component. The root node, activity body, and confluence point are three functional parts in this logical object, and the three are associated in the order of "input call, execution record, output monitoring"). S23. Assign the height of the growth chain based on the state generated by the AI ​​model instance; S24. Connection between the growth chain and the initial point under the parallel orchestration rule based on the height switching of the growth chain.

[0023] As described in steps S21-S24 above, the dynamically allocated mapping scenario represents the dynamic binding relationship between AI skill standardization components and AI model instances. Multiple callable AI model instances in the cloud server are mapped side-by-side to multiple initial points, each initial point corresponding to one AI model instance. Each initial point is associated with at least the model identifier, calling address, current running state, and available resource information of the AI ​​model instance. The growth chain carries the execution process and state changes of the AI ​​skill standardization components, with each growth chain corresponding to at least one AI skill standardization component. The height of the growth chain is assigned based on the state generated by the AI ​​model instance. The height of the growth chain represents the state generated by the AI ​​skill standardization component on the corresponding AI model instance. The state generated by the AI ​​model instance includes the amount of output data, execution time, and response state generated during the execution of the AI ​​skill standardization component. For example, for code generation AI skill standardization components, the number of generated lines of code can be converted into the height of the growth chain; for code commenting AI skill standardization components, the number of generated comment characters can be converted into the height of the growth chain; for code review AI skill standardization components, the number of reviewed files or generated issues can be converted into the height of the growth chain. As the AI ​​model instance continues to generate results, the height of the corresponding growth chain is updated in real time. Based on the height of the growth chain, the connection between the growth chain and the initial point is switched under the parallel orchestration rules. For example, when multiple AI skill standardization components execute simultaneously according to the parallel orchestration rules, the current height of each parallel growth chain is continuously obtained, and the height difference between different growth chains is compared. If the height of a certain growth chain is significantly lower than that of other growth chains, it indicates that the AI ​​skill standardization component corresponding to that growth chain is executing slowly, or that the AI ​​model instance corresponding to its connected initial point is experiencing high load, large response latency, or execution abnormalities. In this case, the growth chain is disconnected from the current initial point and reconnected to an idle or low-load initial point, allowing the AI ​​skill standardization component in that growth chain to switch to a new AI model instance to continue execution, achieving global resource distribution and automatically avoiding busy AI model instances.

[0024] In one embodiment, step S21, which involves assigning AI skill standardization components to different growth chains and connecting the different growth chains to different starting points, includes: S211. Map the AI ​​skill standardization components to the root node of the growth chain for storage. The root node is the starting point of the growth chain and is used to carry the call address of the AI ​​skill standardization component to be executed. S212. Based on the call address, obtain the API interface in the AI ​​skill standardization component, send a call request to the initial point, and establish a connection.

[0025] As described in steps S211-S212 above, the AI ​​skill standardization component to be executed is mapped to the root node of the growth chain for storage. The growth chain is a dynamically allocated component execution state object in the mapping scenario, used to record the execution process of the AI ​​skill standardization component; the root node is the starting point of the growth chain, used as the entry point for calling the AI ​​skill standardization component. By mapping the AI ​​skill standardization component to the root node, the originally independent AI skill standardization component can be transformed into a schedulable execution object. The root node not only stores the component's entry point but also stores the basic information required for component execution, so that when connecting the initial point later, it can be clearly stated which AI skill standardization component to call, which input parameters to use, and which workflow instance to belong to. A call request is sent to the initial point via the API interface. Upon receiving the call request, the initial point invokes the corresponding AI model instance to execute the AI ​​skill standardization component based on the component type and call address in the request. When the initial point confirms that it can handle the call request, a connection is established between the growth chain and the initial point. This connection indicates that the AI ​​skill standardization component currently carried by the growth chain is executed by the AI ​​model instance corresponding to the initial point.

[0026] In one embodiment, step S23, which assigns a height to the growth chain based on the state generated by the AI ​​model instance, includes: S231. Obtain the output data volume generated by the current AI skill standardization component. The output data volume includes the number of generated lines of code, the number of generated comment characters, and the number of generated test cases. S232. Convert the output data volume into the growth height of the growth chain according to the preset mapping rules (which refers to the calculation rules for converting the output data volume of different types of AI skill standardized components into a unified numerical indicator "growth height of the growth chain"), and update the growth height of the growth chain in real time.

[0027] As described in steps S231-S232 above, since the output forms of different AI skill standardization components are different, corresponding mapping rules are set for different component types. For example, it can be set that every 10 lines of code generated corresponds to 1 height unit, every 100 comment characters generated corresponds to 1 height unit, and every test case generated corresponds to 1 height unit. The corresponding growth height is calculated based on the currently added output data volume, and this growth height is superimposed on the existing growth height of the growth chain. The growth height is the cumulative height value formed by the growth chain in the current execution cycle, used to represent the cumulative execution progress of the corresponding AI skill standardization component. After each acquisition of new output data volume, the system updates the growth height of the growth chain in real time and stores the updated growth height in the activity state data of the growth chain. The updated growth height can be used by subsequent steps to determine the difference in execution progress between multiple parallel growth chains, thereby triggering connection switching, pausing execution, or resuming execution.

[0028] In one embodiment, step S24, which involves connecting the growth chain to the initial point under the growth chain height-switching parallel orchestration rule, includes: S241. Configure a height sensor (a status monitoring module deployed at the junction of each growth chain to read the current growth height of the growth chain in real time) at the junction of each growth chain. The height sensor obtains the current growth height of all growth chains under the parallel orchestration rules in real time. S242. By comparing the growth height with a preset difference threshold (the maximum height difference allowed between the highest and lowest growth heights in a parallel growth chain group), when the growth height between each growth chain exceeds the preset difference threshold, the growth chain with the lowest growth height is switched to an idle initial point.

[0029] As described in steps S241-S242 above, the convergence point is the monitoring end of the growth chain, used to receive the output results returned by the AI ​​model instance and record the growth height of the growth chain. Since the growth height of each growth chain is different, there will be a height difference between each growth chain. The height difference is the difference between the growth chain with the highest growth height and the growth chain with the lowest growth height in the parallel growth chain group, that is: growth height difference = highest growth height - lowest growth height. By using the height difference, it is determined whether there is an imbalance in the execution progress among the parallel AI skill standardization components. When an imbalance occurs (when it exceeds the preset difference threshold), the initial point of the growth chain is changed. The purpose of this is to classify and judge the difference in execution progress between parallel growth chains, and trigger the pause of the growth chain with the highest growth height or switch the growth chain with the lowest growth height respectively, thereby avoiding the imbalance of parallel branch execution, the accumulation of intermediate products and the blocking of inefficient model instances, and improving the execution stability, resource utilization and continuous execution capability of software engineering workflow instances.

[0030] In one embodiment, step S242, which involves switching the growth chain with the lowest growth height to an idle initial point when the growth height between each growth chain exceeds a preset difference threshold, includes: S2421. When the growth height of the highest growth chain and the growth height of the lowest growth chain exceed a preset difference threshold; S2422. The root node of the lowest growth chain grows a root chain (temporary scheduling connection channel) that connects to an idle initial point, and transmits the replicas of the lowest growth chain to the idle initial point through the root chain. S2423. When the replicant is within the difference threshold, switch the lowest growth chain to the initial point where the replicant is located.

[0031] As described in steps S2421-S2423 above, the root chain is a temporary scheduling connection channel derived from the root node of the lowest growth chain. It is used to generate a derived root chain to connect to an idle initial point and perform pre-execution when the growth height of the highest growth chain exceeds a preset difference threshold with the growth height of the lowest growth chain. The root chain records the source growth chain identifier, the current AI skill standardization component identifier, the component call address, the context identifier, the breakpoint position, and the target idle initial point identifier. When a root chain needs to be generated, an idle initial point is selected from the dynamic allocation mapping scenario, and a temporary connection request is sent to the API interface corresponding to the idle initial point based on the AI ​​skill standardization component call address and context identifier recorded in the root chain to establish a temporary scheduling connection channel between the root chain and the idle initial point. The replica is a temporary execution instance created on an idle initial point based on the state snapshot of the lowest growth chain. When generating the replica, the current execution state of the lowest growth chain is snapshotted to obtain a state snapshot data packet. The state snapshot data packet includes the component call address, input parameters, execution context, unfinished task data, breakpoint position, and current growth height. The state snapshot data packet is sent to the idle initiation point via the temporary scheduling connection channel corresponding to the root chain. After parsing the state snapshot data packet, the idle initiation point creates a temporary execution instance that is consistent with the state of the minimum growth chain and uses the temporary execution instance as a copy of the minimum growth chain. The copy continues to execute the current AI skill standardization component from the breakpoint position on the AI ​​model instance corresponding to the idle initiation point.

[0032] For example, when executing the code generation component, code annotation component, and test case generation component in parallel, if the highest growth chain height corresponding to the code generation component is 80, the lowest growth chain height corresponding to the test case generation component is 30, and the preset difference threshold is 20, then the height difference is 50, exceeding the preset difference threshold. At this point, the root node of the lowest growth chain grows a root chain and connects to an idle initial point. The state snapshot data packet of the lowest growth chain is sent to the idle initial point via a temporary scheduling connection channel. After parsing the state snapshot data packet, the idle initial point creates a temporary execution instance with the same state as the lowest growth chain and uses this temporary execution instance as a copy of the lowest growth chain. The copy continues to generate test cases from the breakpoint of the lowest growth chain. When the copy reaches a height of 65, the height difference between it and the highest growth chain is 15, within the preset difference threshold. The lowest growth chain is then switched to the initial point where the copy is located, and the AI ​​model instance corresponding to the idle initial point continues execution. After the switch is complete, the root chain is released.

[0033] In one embodiment, step S3, which involves acquiring human-machine collaborative review data at key nodes of a software engineering workflow instance and injecting this data into an AI skills standardization component to adjust the component in real time, includes: S31. Preset key nodes in the workflow template. When the software engineering workflow instance reaches the key node, pause automatic execution and collect the intermediate products generated by the current AI skill standardization component (data generated by the current AI skill standardization component during execution but not yet in the final delivery stage, including source code files, code comments, code review reports, test cases, defect fix patches, etc.) and push them to the review terminal. S32. Obtain structured audit opinion data through the audit terminal. The structured audit opinion data includes audit decision results and remediation suggestion instructions. S33. Parse the structured review opinion data into adjustment parameters, adjust the input parameters of the current AI skill standardization component, or revert to the specified historical node and re-execute when the review decision result is rejection; S34. Record the structured review comments and corresponding adjustment operations, and store them in the review log database.

[0034] As described in steps S31-S34 above, key nodes refer to process nodes that require manual judgment, confirmation, or approval. These can be set after code generation, code review, test case generation, defect fixing, or before release. Key nodes can be bound to an AI skill standardization component or set at the convergence point of multiple AI skill standardization components. Reviewers view intermediate products through the review terminal and input review comments. The review terminal converts the reviewers' input comments into structured review comment data. The structured review comment data includes review decision results and remediation suggestion instructions. The review decision results represent the reviewers' conclusion on the current intermediate product, including approval, approval after modification, and rejection. Remediation suggestion instructions represent adjustment requirements for the current or subsequent AI skill standardization components, including modifying input parameters, adjusting code specifications, increasing security levels, supplementing test scope, modifying prompt constraints, specifying rollback nodes, and re-execution conditions. When the review decision is "pass," the pause status of the key node is lifted, allowing the software engineering workflow instance to continue executing subsequent AI skill standardization components. When the review decision is "pass after modification," the input parameters of the current AI skill standardization component are adjusted according to the repair suggestion instructions. This could include adding code style constraints, modifying the target programming language version, supplementing interface fields, increasing test coverage, or adjusting output format requirements. Based on the adjusted input parameters, the current AI skill standardization component is then invoked again, or subsequent AI skill standardization components continue to execute. When the review decision is "reject," the software engineering workflow instance is rolled back to a specified historical node and re-executed based on the rollback node identifier in the structured review opinion data. The specified historical node can be a code generation node, code comment node, code review node, or test generation node. During rollback, the review opinion data and necessary context information are retained, and invalid subsequent intermediate artifacts are removed or marked to prevent the continued propagation of erroneous results. The structured review opinion data and corresponding adjustment operations are recorded and stored in the review log library. The review log library allows for tracing the impact of each manual review on the AI ​​skill standardization component and provides data support for subsequent optimization of workflow templates, adjustment of component parameters, and improvement of review rules.

[0035] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. An intelligent software engineering implementation method based on AI skill-based workflow orchestration, characterized in that, Includes the following steps: Build multiple standardized AI skill components, combine these components into workflow templates using preset orchestration rules, and store them in a local template library; Obtain project requirements information, match workflow templates based on project requirements information, and dynamically allocate AI model instances to the standardized AI skill components in the workflow templates to generate software engineering workflow instances; Obtain human-machine collaborative review data at key nodes of software engineering workflow instances, inject the human-machine collaborative review data into the AI ​​skill standardization component, and adjust the AI ​​skill standardization component in real time.

2. The intelligent software engineering implementation method based on AI skill orchestration workflow according to claim 1, characterized in that, The step of constructing multiple standardized AI skill components and combining these components into a workflow template according to preset orchestration rules includes: The AI ​​skills of code generation, code commenting, and code review are standardized and encapsulated to obtain multiple standardized AI skill components; Based on the input and output parameters of each AI skill standardization component, determine the data transmission relationship between different AI skill standardization components, and based on the execution conditions of each AI skill standardization component, determine the execution dependency relationship between different AI skill standardization components; Based on execution dependencies, multiple standardized AI skill components are combined according to preset orchestration rules to generate a workflow template. The preset orchestration rules include serial orchestration rules, parallel orchestration rules, and conditional branch orchestration rules.

3. The intelligent software engineering implementation method based on AI skill orchestration workflow according to claim 1, characterized in that, The steps of dynamically allocating AI model instances to the standardized AI skill components in the workflow template to generate software engineering workflow instances include: A dynamic allocation mapping scenario is built based on a cloud server, and multiple AI model instances are used as initial points and mapped side by side in the dynamic allocation mapping scenario; In a dynamic allocation mapping scenario, multiple growth chains are configured, and standardized AI skill components are assigned to different growth chains. Different growth chains are then connected to different initial points. A growth chain includes a root node, an active body, and a confluence point. The height of the growth chain is assigned based on the state generated by the AI ​​model instance; Connection between the growth chain and the initial point under the parallel orchestration rule of height switching of the growth chain.

4. The intelligent software engineering implementation method for AI skill-based workflow orchestration according to claim 3, characterized in that, The step of assigning standardized AI skill components to different growth chains and connecting the different growth chains to different starting points includes: The standardized AI skills components are mapped to the root node of the growth chain for storage. The root node is the starting point of the growth chain and is used to carry the call address of the currently executed standardized AI skills component. Based on the call address, the API interface in the AI ​​skill standardization component is obtained, a call request is sent to the initial point, and a connection is established.

5. The intelligent software engineering implementation method for AI skill-based workflow orchestration according to claim 3, characterized in that, The step of assigning the height of the growth chain based on the state generated by the AI ​​model instance includes: Obtain the output data volume generated by the current AI skill standardization component. The output data volume includes the number of generated lines of code, the number of generated comment characters, and the number of generated test cases. The output data volume is converted into the growth height of the growth chain according to the preset mapping rules, and the growth height of the growth chain is updated in real time.

6. The intelligent software engineering implementation method for AI skill-based workflow orchestration according to claim 3, characterized in that, The steps for connecting the growth chain and the initial point under the height-switching parallel orchestration rule of the growth chain include: A height sensor is configured at the convergence point of each growth chain. The height sensor obtains the current growth height of each growth chain under the parallel orchestration rule in real time. By comparing the growth height with a preset difference threshold, when the growth height between each growth chain exceeds the preset difference threshold, the growth chain with the lowest growth height is switched to an idle initial point.

7. The intelligent software engineering implementation method for workflow orchestration based on AI skills according to claim 6, characterized in that, The step of switching the growth chain with the lowest growth height to an idle initial point when the growth height between each growth chain exceeds a preset difference threshold includes: When the growth height of the highest growth chain exceeds the growth height of the lowest growth chain, the following occurs: The root node of the lowest growth chain grows a root chain that connects to the free initial point, and the copy of the lowest growth chain is transferred to the free initial point through the root chain. When the replicant is within the difference threshold, the lowest growth chain is switched to the initial point where the replicant is located.

8. The intelligent software engineering implementation method for workflow orchestration based on AI skills according to claim 1, characterized in that, The steps of obtaining human-machine collaborative review data at key nodes of the software engineering workflow instance, injecting the human-machine collaborative review data into the AI ​​skill standardization component, and adjusting the AI ​​skill standardization component in real time include: Pre-set key nodes in the workflow template. When the software engineering workflow instance reaches a key node, pause automatic execution and collect the intermediate products generated by the current AI skill standardization component, and push them to the review terminal. Structured audit feedback data is obtained through the audit terminal. The structured audit feedback data includes audit decision results and remediation suggestions. The structured review comments data are parsed into adjustment parameters, which can be used to adjust the input parameters of the current AI skill standardization components or revert to a specified historical node for re-execution when the review decision result is rejection; Record structured review comments and corresponding adjustment operations, and store them in the review log database.

Images