An agent skill node-oriented compatibility conflict detection method and system

Abstract

The application discloses a kind of compatibility conflict detection method and system for agent skill node, which comprises the following steps: collecting skill node information set to generate skill contract document, maintaining node relationship based on relationship dependency graph for each round of updated skill contract document, and extracting contract difference description set of each round of update;According to the contract difference description set, determine the candidate detection node from the relationship dependency graph, carry out multidimensional compatibility verification on the candidate detection node based on interface structure, business semantics and data distribution, and obtain the identification result of compatibility conflict;According to the identification result, traverse the relationship dependency graph along the dependency chain to obtain the conflict influence set corresponding to the skill node, calculate the global influence score of each skill node in the conflict influence set based on the skill node attribute to obtain the conflict influence report, and determine the release decision of the current round of update according to the conflict influence report.The application can realize effective detection of compatibility conflict, thereby improving the response efficiency and stability of update.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a compatibility conflict detection method and system for intelligent agent skill nodes. Background Technology

[0002] In an intelligent agent platform, each intelligent agent typically corresponds to one or more skill nodes required by a business scenario. A skill node is an independent computing unit that digitally encapsulates expert experience and deterministic execution instructions. It possesses strong professional capabilities in specific business domains of enterprises / organizations. By combining and calling multiple skill nodes in a multi-agent collaborative mode, complex business task processes can be completed in a targeted manner. In the task process, the version update of any skill node will propagate downstream along the dependency chain. Among these, how to effectively detect compatibility conflicts of skill nodes has become a key technical issue.

[0003] In existing technologies, the number of skill nodes and the number of version updates of intelligent agent platforms continue to grow with the expansion of business scale. In the past, the solution to compatibility conflict issues adopted a manual ledger management method with a long cycle. This method, which manually maintains the version dependency table, not only fails to cover the implicit dependencies of updated skill nodes across teams, but also has serious time delays in high-frequency iteration scenarios with business scale expansion. Secondly, the skill node updates of the intelligent agent platform need to be completed uniformly within a fixed release window. When updating and releasing, the simultaneous changes of thousands of skill nodes place higher demands on the timeliness and accuracy of fault location. The lack of timeliness and accuracy will lead to the inability to respond to the regulatory requirements for compliant updates, seriously blocking the version update and release process. Subsequently, the intelligent agent platform uses version numbers for rule verification and judges compatibility issues based on version number conventions. However, it cannot identify semantic conflicts in interface fields, resulting in a lack of awareness of business semantics. In particular, it is difficult to detect violations of destructive changes, leading to frequent silent events where detection is effective but business fails.

[0004] It is evident that existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a compatibility conflict detection method and system for intelligent agent skill nodes, which can effectively detect compatibility conflicts, thereby improving the update response efficiency and stability.

[0006] To address the aforementioned technical problems, the first aspect of this invention discloses a compatibility conflict detection method for intelligent agent skill nodes, the method comprising: Collect skill node information sets to generate skill contract documents, maintain node relationships in the relation dependency graph of each round of updates based on the skill contract documents, and extract the contract difference description set of each round of updates; Candidate detection nodes are determined from the relationship dependency graph based on the contract difference description set. Multi-dimensional compatibility verification of the candidate detection nodes is performed based on interface structure, business semantics and data distribution to obtain the identification results of compatibility conflicts. Based on the identification results, the dependency graph is traversed along the dependency chain to obtain the conflict impact set corresponding to the skill node. The global impact score of each skill node in the conflict impact set is calculated based on the skill node attributes to obtain the conflict impact report. The release decision for the current round of updates is determined based on the conflict impact report.

[0007] As an optional implementation, in the first aspect of the present invention, the skill node information set includes a list of input parameters, a list of output fields, field names, data types, value ranges, an enumerated value set, a precision threshold, a dependent rule base name, and a version number corresponding to all skill nodes. The skill contract document is a structured document generated by preprocessing the skill node information set. It is stored in the contract registration center of the intelligent agent platform through a key-value structure. The skill contract document is automatically archived and retained when the version of the skill node is updated in each round.

[0008] As an optional implementation, in the first aspect of the present invention, the relationship dependency graph is a directed acyclic graph, where each node represents a skill node, and each directed edge between nodes represents a dependency relationship for skill node invocation. The relationship dependency graph is updated in real time through message passing when skill nodes are registered, version updates are released, and dependency relationships change. The contract difference description set includes destructive changes, extensible changes, and internal changes. Any contract difference description is obtained by identifying and extracting the field differences and / or semantic differences between the skill contract documents before and after the update for the corresponding skill node.

[0009] As an optional implementation, in a first aspect of the invention, determining candidate detection nodes from a relation dependency graph based on a set of contract difference descriptions includes: Extract the associated skill nodes with disruptive changes from the set of contract difference descriptions corresponding to each round of updates; Based on the directed edge relationships in the relation dependency graph, determine the downstream nodes of the call dependency relationships corresponding to the associated skill nodes, and identify the associated skill nodes and downstream nodes as candidate detection nodes. The destructive changes include deleting, modifying, expanding, and shrinking fields, individual values, and value ranges in the skill contract.

[0010] As an optional implementation, in the first aspect of the present invention, multi-dimensional compatibility verification is performed on candidate detection nodes based on interface structure, business semantics, and data distribution to obtain the identification result of compatibility conflicts, including: Interface structure compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the input source mapping of the contract document record in the candidate detection node, performing an intersection operation on the input source mapping and the contract difference description, and determining the candidate detection nodes whose intersection operation result is not an empty set as having an interface structure conflict. Business semantic compatibility verification involves traversing candidate detection nodes in the relational dependency graph, reading the business description text of the fields recorded in the contract document of the candidate detection node, identifying the change in the scope corresponding to the business meaning through keyword difference detection, marking the candidate detection node that is determined to be a change in scope as a business semantic conflict to prompt for composite confirmation, and comparing and matching the rule base version number declared in the contract in the candidate detection node with the list of valid rule base versions, and determining that the candidate detection node that does not match has a compliant version conflict. Data distribution compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the feature distribution constraints declared in the contract documents of the candidate detection nodes, comparing the output distribution description of the associated skill nodes in the candidate detection nodes with the feature distribution constraints of the downstream nodes corresponding to the associated skill nodes, and determining the candidate detection nodes that violate the distribution constraints as data distribution conflicts. Specifically, the feature distribution constraints are the feature distribution assumptions of the input and output features of the skill nodes that depend on the machine learning model during model training. By merging interface structure conflicts, business semantic conflicts, compliance version conflicts, and data distribution conflicts, the identification results of compatibility conflicts corresponding to candidate detection nodes are obtained.

[0011] As an optional implementation, in the first aspect of the present invention, obtaining the conflict impact set corresponding to the skill node by traversing the dependency graph along the dependency chain based on the identification result includes: Based on the candidate detection nodes identified as having compatibility conflicts in the identification results, the starting point corresponding to the breadth-first traversal of the relationship dependency graph is determined, and the upstream nodes of the candidate detection nodes are visited sequentially along the dependency chain layer by layer. The multidimensional compatibility check is performed on each upstream node until a compatible node is included in the upstream branch of any dependency chain composed of the upstream nodes. Then, the traversal of the upstream branch is stopped, and the identification results of compatibility conflicts corresponding to all upstream nodes in the upstream branch are output. The compatible node is an upstream node that indirectly depends on the skill node corresponding to the compatibility conflict along the dependency chain and does not have any compatibility conflict. The identification results of compatibility conflicts corresponding to the candidate detection nodes and the identification results of compatibility conflicts corresponding to the upstream nodes are combined to obtain the conflict impact set of the relationship dependency graph in the update round.

[0012] As an optional implementation, in the first aspect of the invention, a conflict impact report is obtained by calculating the global impact score of each skill node in the conflict impact set based on the skill node attributes, including: The statistical period for all skill nodes is determined based on the business update cycle. The call frequency of each skill node in the conflict impact report is collected based on the statistical period. The average call frequency of all skill nodes in the relationship dependency graph within the statistical period is calculated. The frequency ratio between the call frequency and the average call frequency is calculated to obtain the activity score corresponding to each skill node in the conflict impact report. Based on the business semantics, determine the business level score of each skill node in the conflict impact report on its dependency chain; Based on the identification results of the compatibility conflicts, determine the conflict type score corresponding to each skill node in the conflict impact report; For any skill node in the conflict impact report, an initial impact score is determined based on the business level score and conflict type score. The number of times the skill node affects downstream nodes along its dependency chain is determined. The initial impact score of the skill node is accumulated based on a preset impact step factor and the number of impacts. The impact step factor of the layer-by-layer accumulation process is adjusted according to the initial impact score corresponding to the skill node at the current level. The dependency impact score corresponding to each skill node in the conflict impact report is output. The global impact score of each skill node in the conflict impact set is calculated by weighted summation of the activity score, business level score, conflict type score, and dependency impact score. The skill nodes in the conflict impact set are sorted from high to low according to the global impact score to obtain a skill node list. For each skill node in the skill node list, the conflict type and impact propagation path of the skill node are recorded. The impact propagation path corresponds to other skill nodes affected by the skill node. The release recommendation corresponding to the compatibility conflict in the current round is determined according to the recorded content of the skill node list. A conflict impact report is obtained based on the skill node list and the release recommendation.

[0013] As an optional implementation, in a first aspect of the invention, determining the release decision for the current update based on the conflict impact report includes: Based on the conflict impact set, global impact score, and release recommendation corresponding to the conflict impact report, a judgment is made based on preset release conditions, and a release decision that meets the release conditions is selected to be executed. The process of issuing the decision is specifically as follows: The release decision adopts a phased canary release strategy, which distributes the request traffic for each updated skill node proportionally and records the operation indicators independently; During the canary release process, collect the average values ​​of business output fields and request exception rate of skill nodes before and after the update within the canary window period; Calculate the deviation of the mean value of the business output field from its previous value to obtain the business consistency index. The difference between the request anomaly rate and the rate before the update is calculated to obtain the anomaly rate index; Based on the business consistency index and the anomaly rate index, a weighted calculation is performed using an indicator function to obtain the rollback score in the release decision. When the rollback score is less than the preset rollback threshold, a rollback decision is triggered and a rollback report is generated. When the rollback score is not less than the preset rollback threshold, it is determined whether the gray release duration meets the observation window threshold. If so, the request traffic ratio is increased according to the preset traffic step size to complete the full switch of skill node update. If not, a second judgment is made based on the rollback score when the gray release duration meets the observation window threshold.

[0014] A second aspect of this invention discloses a compatibility conflict detection system for intelligent agent skill nodes, the system comprising: The contract management module is used to collect skill node information sets to generate skill contract documents, maintain node relationships in the dependency graph of each round of updates based on the skill contract documents, and extract the contract difference description set of each round of updates. The multi-dimensional verification module is used to determine candidate detection nodes from the relation dependency graph based on the contract difference description set, and to perform multi-dimensional compatibility verification on the candidate detection nodes based on interface structure, business semantics and data distribution to obtain the identification results of compatibility conflicts. The conflict detection module is used to traverse the dependency graph along the dependency chain based on the identification results to obtain the conflict impact set corresponding to the skill node, calculate the global impact score of each skill node in the conflict impact set based on the skill node attributes to obtain the conflict impact report, and determine the release decision for the current round of updates based on the conflict impact report.

[0015] As an optional implementation, in the second aspect of the present invention, the skill node information set includes a list of input parameters, a list of output fields, field names, data types, value ranges, an enumerated value set, a precision threshold, a dependent rule base name, and a version number corresponding to all skill nodes. The skill contract document is a structured document generated by preprocessing the skill node information set. It is stored in the contract registration center of the intelligent agent platform through a key-value structure. The skill contract document is automatically archived and retained when the version of the skill node is updated in each round.

[0016] As an optional implementation, in the second aspect of the present invention, the relationship dependency graph is a directed acyclic graph, where each node represents a skill node, and each directed edge between nodes represents a dependency relationship for skill node invocation. The relationship dependency graph is updated in real time through message passing when skill nodes are registered, version updates are released, and dependency relationships change. The contract difference description set includes destructive changes, extensible changes, and internal changes. Any contract difference description is obtained by identifying and extracting the field differences and / or semantic differences between the skill contract documents before and after the update for the corresponding skill node.

[0017] As an optional implementation, in a second aspect of the invention, determining candidate detection nodes from a relation dependency graph based on a set of contract difference descriptions includes: Extract the associated skill nodes with disruptive changes from the set of contract difference descriptions corresponding to each round of updates; Based on the directed edge relationships in the relation dependency graph, determine the downstream nodes of the call dependency relationships corresponding to the associated skill nodes, and identify the associated skill nodes and downstream nodes as candidate detection nodes. The destructive changes include deleting, modifying, expanding, and shrinking fields, individual values, and value ranges in the skill contract.

[0018] As an optional implementation, in the second aspect of the present invention, multi-dimensional compatibility verification is performed on candidate detection nodes based on interface structure, business semantics, and data distribution to obtain the identification result of compatibility conflicts, including: Interface structure compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the input source mapping of the contract document record in the candidate detection node, performing an intersection operation on the input source mapping and the contract difference description, and determining the candidate detection nodes whose intersection operation result is not an empty set as having an interface structure conflict. Business semantic compatibility verification involves traversing candidate detection nodes in the relational dependency graph, reading the business description text of the fields recorded in the contract document of the candidate detection node, identifying the change in the scope corresponding to the business meaning through keyword difference detection, marking the candidate detection node that is determined to be a change in scope as a business semantic conflict to prompt for composite confirmation, and comparing and matching the rule base version number declared in the contract in the candidate detection node with the list of valid rule base versions, and determining that the candidate detection node that does not match has a compliant version conflict. Data distribution compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the feature distribution constraints declared in the contract documents of the candidate detection nodes, comparing the output distribution description of the associated skill nodes in the candidate detection nodes with the feature distribution constraints of the downstream nodes corresponding to the associated skill nodes, and determining the candidate detection nodes that violate the distribution constraints as data distribution conflicts. Specifically, the feature distribution constraints are the feature distribution assumptions of the input and output features of the skill nodes that depend on the machine learning model during model training. By merging interface structure conflicts, business semantic conflicts, compliance version conflicts, and data distribution conflicts, the identification results of compatibility conflicts corresponding to candidate detection nodes are obtained.

[0019] As an optional implementation, in a second aspect of the invention, obtaining the conflict impact set corresponding to the skill node by traversing the dependency graph along the dependency chain based on the identification result includes: Based on the candidate detection nodes identified as having compatibility conflicts in the identification results, the starting point corresponding to the breadth-first traversal of the relationship dependency graph is determined, and the upstream nodes of the candidate detection nodes are visited sequentially along the dependency chain layer by layer. The multidimensional compatibility check is performed on each upstream node until a compatible node is included in the upstream branch of any dependency chain composed of the upstream nodes. Then, the traversal of the upstream branch is stopped, and the identification results of compatibility conflicts corresponding to all upstream nodes in the upstream branch are output. The compatible node is an upstream node that indirectly depends on the skill node corresponding to the compatibility conflict along the dependency chain and does not have any compatibility conflict. The identification results of compatibility conflicts corresponding to the candidate detection nodes and the identification results of compatibility conflicts corresponding to the upstream nodes are combined to obtain the conflict impact set of the relationship dependency graph in the update round.

[0020] As an optional implementation, in a second aspect of the invention, a conflict impact report is obtained by calculating the global impact score of each skill node in the conflict impact set based on skill node attributes, including: The statistical period for all skill nodes is determined based on the business update cycle. The call frequency of each skill node in the conflict impact report is collected based on the statistical period. The average call frequency of all skill nodes in the relationship dependency graph within the statistical period is calculated. The frequency ratio between the call frequency and the average call frequency is calculated to obtain the activity score corresponding to each skill node in the conflict impact report. Based on the business semantics, determine the business level score of each skill node in the conflict impact report on its dependency chain; Based on the identification results of the compatibility conflicts, determine the conflict type score corresponding to each skill node in the conflict impact report; For any skill node in the conflict impact report, an initial impact score is determined based on the business level score and conflict type score. The number of times the skill node affects downstream nodes along its dependency chain is determined. The initial impact score of the skill node is accumulated based on a preset impact step factor and the number of impacts. The impact step factor of the layer-by-layer accumulation process is adjusted according to the initial impact score corresponding to the skill node at the current level. The dependency impact score corresponding to each skill node in the conflict impact report is output. The global impact score of each skill node in the conflict impact set is calculated by weighted summation of the activity score, business level score, conflict type score, and dependency impact score. The skill nodes in the conflict impact set are sorted from high to low according to the global impact score to obtain a skill node list. For each skill node in the skill node list, the conflict type and impact propagation path of the skill node are recorded. The impact propagation path corresponds to other skill nodes affected by the skill node. The release recommendation corresponding to the compatibility conflict in the current round is determined according to the recorded content of the skill node list. A conflict impact report is obtained based on the skill node list and the release recommendation.

[0021] As an optional implementation, in a second aspect of the invention, determining the release decision for the current update based on the conflict impact report includes: Based on the conflict impact set, global impact score, and release recommendation corresponding to the conflict impact report, a judgment is made based on preset release conditions, and a release decision that meets the release conditions is selected to be executed. The process of issuing the decision is specifically as follows: The release decision adopts a phased canary release strategy, which distributes the request traffic for each updated skill node proportionally and records the operation indicators independently; During the canary release process, collect the average values ​​of business output fields and request exception rate of skill nodes before and after the update within the canary window period; Calculate the deviation of the mean value of the business output field from its previous value to obtain the business consistency index. The difference between the request anomaly rate and the rate before the update is calculated to obtain the anomaly rate index; Based on the business consistency index and the anomaly rate index, a weighted calculation is performed using an indicator function to obtain the rollback score in the release decision. When the rollback score is less than the preset rollback threshold, a rollback decision is triggered and a rollback report is generated. When the rollback score is not less than the preset rollback threshold, it is determined whether the gray release duration meets the observation window threshold. If so, the request traffic ratio is increased according to the preset traffic step size to complete the full switch of skill node update. If not, a second judgment is made based on the rollback score when the gray release duration meets the observation window threshold.

[0022] A third aspect of this invention discloses another compatibility conflict detection system for intelligent agent skill nodes, the system comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the compatibility conflict detection method for intelligent agent skill nodes disclosed in the first aspect of the present invention.

[0023] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the compatibility conflict detection method for intelligent agent skill nodes disclosed in the first aspect of the present invention.

[0024] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention generates skill contract documents by collecting skill node information sets, and maintains node relationships in the dependency graph for each round of updates based on these skill contract documents. Compared to manual ledger methods, it can automatically and completely capture explicit and implicit dependencies across teams and agents, solving the problem of incomplete coverage in manually maintained version dependency tables. Furthermore, it uses a contract difference description set to drive update detection, enabling real-time response to dependency changes in high-frequency iteration scenarios, significantly improving the timeliness of compatibility testing, avoiding version conflicts caused by delayed manual updates, and overcoming the limitations of dependency version number conventions through multi-dimensional compatibility verification, accurately identifying the language of interface fields. This system effectively reduces silent events where detection is effective but business operations fail due to a lack of semantic awareness, thereby enhancing compatibility assurance at the business semantic level. By traversing the dependency graph along the dependency chain to generate a conflict impact set and combining it with skill node attributes to calculate an impact score and generate a conflict impact report, it assists in release decisions. In scenarios with large-scale synchronous changes to skill nodes, it can quickly and accurately locate the scope and degree of impact of affected nodes, avoiding version release process blockages caused by insufficient timeliness and accuracy of fault location, and improving the response efficiency and release stability to regulatory compliance update requirements. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 This is a flowchart illustrating a compatibility conflict detection method for intelligent agent skill nodes disclosed in an embodiment of the present invention.

[0027] Figure 2 This is a schematic diagram of the structure of a compatibility conflict detection system for intelligent agent skill nodes disclosed in an embodiment of the present invention.

[0028] Figure 3 This is a schematic diagram of another compatibility conflict detection system for intelligent agent skill nodes disclosed in an embodiment of the present invention. Detailed Implementation

[0029] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and 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.

[0030] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0031] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0032] This invention discloses a compatibility conflict detection method and system for agent skill nodes. It generates a skill contract document by collecting skill node information sets and maintains node relationships in the dependency graph for each update based on the skill contract document. Compared to manual ledger methods, it can automatically and completely capture explicit and implicit dependencies across teams and agents, solving the problem of incomplete coverage in manually maintained version dependency tables. Furthermore, it uses a contract difference description set to drive update detection, enabling real-time response to dependency changes in high-frequency iteration scenarios, significantly improving the timeliness of compatibility detection and avoiding version conflicts caused by delayed manual updates. It also overcomes the limitations of dependency version number conventions through multi-dimensional compatibility verification. This system can accurately identify semantic conflicts and destructive changes in interface fields, effectively reducing silent events where detection is effective but business operations fail due to a lack of semantic awareness. It enhances compatibility assurance at the business semantic level. By traversing the dependency graph along the dependency chain to generate a conflict impact set and combining it with skill node attributes to calculate an impact score and generate a conflict impact report, it assists in release decisions. In scenarios with large-scale synchronous changes to skill nodes, it can quickly and accurately locate the scope and extent of affected nodes, avoiding version release process blockages caused by insufficient timeliness and accuracy in fault location, and improving the response efficiency and release stability to regulatory compliance update requirements. These are explained in detail below.

[0033] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating a compatibility conflict detection method for intelligent agent skill nodes disclosed in an embodiment of the present invention. Figure 1 The described compatibility conflict detection method for agent skill nodes can be applied to data processing systems / data processing devices / data processing servers (including local processing servers or cloud processing servers). For example... Figure 1 As shown, the compatibility conflict detection method for agent skill nodes may include the following operations: 101. Collect skill node information set to generate skill contract document, maintain node relationship of the relation dependency graph of each round of update based on skill contract document, and extract the contract difference description set of each round of update.

[0034] It should be noted that the agent platform completes complex business tasks by combining and calling multiple skill nodes. Each skill node is a callable unit that encapsulates specific business logic and has a clear input / output interface contract. Taking an internet finance business scenario as an example, a typical credit approval agent workflow typically consists of multiple skill nodes connected in series and parallel, including but not limited to identity verification, credit retrieval (using authorized, anonymized data), income analysis (using authorized, anonymized data), credit decision-making, and compliance review. In actual agent platforms, these multiple nodes are usually developed and iterated independently by different teams. With the expansion of multiple business scales, the number of registered skill nodes on the agent platform typically exceeds one thousand, and the number of version updates exceeds 150 per month. Skills have multiple layers of dependencies, with an average dependency depth of 4 to 6 layers. When a skill node undergoes a version upgrade, changes to its interface contract (such as field deletion, type change, or enumeration value adjustment) will propagate downstream along the dependency chain, causing compatibility issues. Currently, the detection of compatibility conflicts needs to meet core requirements such as real-time performance, accuracy, and business semantic awareness.

[0035] Optionally, the skill node information set includes a list of input parameters, a list of output fields, field names, data types, value ranges, enumeration value sets, precision thresholds, dependent rule base names, and version numbers for all skill nodes.

[0036] Optionally, the skill contract document is a structured document generated by preprocessing the skill node information set, and is stored in the contract registration center of the intelligent agent platform through a key-value structure.

[0037] Specifically, skill contract documents are automatically archived and retained for each version release of a new skill node.

[0038] Optionally, the dependency graph is a directed acyclic graph, where each node represents a skill node, and each directed edge between nodes represents the dependency relationship of skill node invocation. The dependency graph is updated in real time through message passing when skill nodes are registered, version updates are released, and dependency relationships change.

[0039] Optionally, the contract difference description set includes destructive changes, extensible changes, and internal changes. Any contract difference description is obtained by identifying and extracting the field difference degree and / or semantic difference degree of the skill contract documents before and after the update for the corresponding skill node. It can be understood that the field difference degree and semantic difference pair can be obtained by automatically reading the new version contract of the skill and the current production version contract through the platform system when the skill node submits a new version, and then comparing them field by field.

[0040] 102. Based on the contract difference description set, candidate detection nodes are determined from the relation dependency graph. Multi-dimensional compatibility verification is performed on the candidate detection nodes based on interface structure, business semantics, and data distribution to obtain the identification results of compatibility conflicts.

[0041] Optionally, interface structure compatibility checks can be performed by traversing all downstream nodes in the dependency graph that directly depend on the skill and checking whether the input parameter declarations of each downstream node reference fields that have undergone destructive changes.

[0042] Optionally, business semantic compatibility verification can only detect direct reference conflicts at the field level to resolve interface structure compatibility issues, but cannot detect implicit conflicts at the semantic level. This mainly includes new compatibility verification for changes in calculation methods and conflicts between regulatory rule base versions.

[0043] Optionally, the compatibility verification of data distribution specifically involves skill nodes that rely on machine learning models (taking the Internet finance business scenario as an example, the skill node may rely on machine learning models such as credit scoring models or risk rating models). If data distribution compatibility issues occur during model training and subsequent stages, such as when the output field of the upstream skill undergoes enumeration value expansion or numerical range changes, even if the interface structure is compatible, it may cause the input feature distribution of the downstream model to shift, resulting in distorted model output results.

[0044] 103. Based on the identification results, traverse the dependency graph along the dependency chain to obtain the conflict impact set corresponding to the skill node. Calculate the global impact score of each skill node in the conflict impact set based on the skill node attributes to obtain the conflict impact report. Determine the release decision for the current update based on the conflict impact report.

[0045] Optionally, the conflict impact set is a collection of all skill nodes on the dependency graph that may be affected by compatibility conflicts caused by skill node updates. It is understood that the impact of the conflict is not limited to the first-level downstream nodes that directly depend on the skill, but will also propagate deeper along the dependency chain, such as layer by layer upstream along the caller direction.

[0046] Optionally, skill node attributes can include call frequency, corresponding business level, corresponding detected conflict type, and other skill node attributes associated with the skill node. These attributes can be stored in the skill contract document for easy access.

[0047] Optionally, the global impact score can be the activity score of the skill node, the business level score, the conflict type score, and the dependency impact score. It can be understood that by calculating the global impact score for each affected skill node in the conflict impact set, the potential harm of the conflict to the actual business can be quantified, and the operation and maintenance personnel can be assisted in determining the processing priority.

[0048] Optionally, the conflict impact report can list all affected nodes in descending order of global impact score, while also indicating the conflict type, propagation path, and corresponding handling methods for the release recommendations (such as forcibly blocking the release, allowing it after manual confirmation, allowing it automatically and monitoring it, etc.).

[0049] It should be noted that the release decision can be selected based on the actual situation reflected in the conflict impact report. For example, if there is no compatibility conflict, the release can be automatically approved and the new version of the skill can directly enter the gray release process. If there are nodes with some low impact (such as scores below the decision response threshold), the release can be automatically approved and monitoring alarms can be enabled. If there are nodes with some high impact, the release can be paused, a conflict impact report can be generated, and the operation and maintenance team can be asked to manually confirm it. If the impact is significantly higher than the decision response threshold, the release can be forcibly blocked, and the skill development team can be required to fix the compatibility conflict issues and resubmit. It is understood that the prerequisite for the above release decisions is that the valid compatibility conflict detection results serve as the supporting basis.

[0050] As can be seen, the above-described embodiments of the invention generate skill contract documents by collecting skill node information sets, and maintain node relationships in the dependency graph of each round of updates based on the skill contract documents. Compared with manual ledger methods, this can automatically and completely capture explicit and implicit dependencies across teams and agents, solving the problem of incomplete coverage of manually maintained version dependency tables. Furthermore, by using a contract difference description set to drive update detection, it can respond in real time to dependency changes in high-frequency iteration scenarios, significantly improving the timeliness of compatibility detection and avoiding version conflicts caused by delays in manual updates. Through multi-dimensional compatibility verification, it overcomes the limitations of dependency version number conventions and can accurately identify interface words. This system effectively reduces silent events where detection is effective but business operations fail due to a lack of semantic awareness, thereby enhancing compatibility assurance at the business semantic level. By traversing the dependency graph along the dependency chain to generate a conflict impact set and combining it with skill node attributes to calculate an impact score and generate a conflict impact report, it assists in release decisions. In scenarios with large-scale synchronous changes to skill nodes, it can quickly and accurately locate the scope and degree of impact of affected nodes, avoiding version release process blockages caused by insufficient timeliness and accuracy of fault location, and improving the response efficiency and release stability to regulatory compliance update requirements.

[0051] As an optional embodiment, the step of determining candidate detection nodes from the relation dependency graph based on the contract difference description set in the above steps includes: Extract the associated skill nodes with disruptive changes from the set of contract difference descriptions corresponding to each round of updates; Based on the directed edge relationships in the relation dependency graph, determine the downstream nodes of the call dependency relationships corresponding to the associated skill nodes, and identify the associated skill nodes and downstream nodes as candidate detection nodes. The destructive changes include deleting, modifying, expanding, and shrinking fields, individual values, and value ranges in the skill contract.

[0052] Optionally, contract difference types can be categorized by their destructive nature into destructive changes, extensible changes, and internal changes. Destructive changes can be interpreted as deleting existing output fields, modifying the data type of existing fields, reducing the enumeration value set, increasing numerical precision requirements, or deleting the default value of existing input parameters to make them mandatory. Extensible changes can be interpreted as adding output fields, expanding the enumeration value set, or adding optional input parameters. Internal changes can be interpreted as modifying only the internal calculation logic of the skill, with the input and output contracts remaining completely unchanged. As an example, in an internet finance lending scenario, when an event-fetching skill node is upgraded from v1.8 to v2.0, the "historical overdue days" (integer) field in the output field is deleted, and a new output field "overdue level" (such as an enumeration type with values ​​A / B / C / D / E) is added. This change is identified as a destructive change and needs to enter the subsequent verification process.

[0053] As can be seen, through the above optional embodiments, by extracting destructive changes from the contract difference description set and specifically defining the operation type of destructive changes, it is possible to proactively capture illegal change behaviors that cause silent business failures. This effectively compensates for the lack of business semantic awareness, significantly reduces the occurrence rate of silent events, enhances the ability to accurately identify destructive changes, and provides high-quality input for subsequent business semantic conflict verification. After identifying the skill nodes associated with destructive changes, the downstream calling nodes are accurately determined as candidate detection nodes based on the directed edge relationships of the dependency graph, rather than redundantly scanning all skill nodes. By focusing verification computing resources on the dependency links actually affected by destructive changes, the detection scope is narrowed to ensure response timeliness in large-scale update scenarios. This greatly improves the response speed of compatibility testing in the data preparation stage, provides high-quality input for high-accuracy fault location in the subsequent process, and ensures the process within the version release window and the ability to respond quickly to regulatory compliance requirements.

[0054] As an optional embodiment, the above steps, including performing multi-dimensional compatibility verification on candidate detection nodes based on interface structure, business semantics, and data distribution to obtain the identification result of compatibility conflicts, include: Interface structure compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the input source mapping of the contract document record in the candidate detection node, performing an intersection operation on the input source mapping and the contract difference description, and determining the candidate detection nodes whose intersection operation result is not an empty set as having an interface structure conflict. Business semantic compatibility verification involves traversing candidate detection nodes in the relational dependency graph, reading the business description text of the fields recorded in the contract document of the candidate detection node, identifying the change in the scope corresponding to the business meaning through keyword difference detection, marking the candidate detection node that is determined to be a change in scope as a business semantic conflict to prompt for composite confirmation, and comparing and matching the rule base version number declared in the contract in the candidate detection node with the list of valid rule base versions, and determining that the candidate detection node that does not match has a compliant version conflict. Data distribution compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the feature distribution constraints declared in the contract documents of the candidate detection nodes, comparing the output distribution description of the associated skill nodes in the candidate detection nodes with the feature distribution constraints of the downstream nodes corresponding to the associated skill nodes, and determining the candidate detection nodes that violate the distribution constraints as data distribution conflicts. Specifically, the feature distribution constraints are the feature distribution assumptions of the input and output features of the skill nodes that depend on the machine learning model during model training. By merging interface structure conflicts, business semantic conflicts, compliance version conflicts, and data distribution conflicts, the identification results of compatibility conflicts corresponding to candidate detection nodes are obtained.

[0055] Specifically, the interface structure compatibility check reads the "input source mapping" section of the downstream skill's contract document. This section records the data source of each input parameter of the downstream skill (from a certain output field corresponding to a certain upstream skill). The intersection operation is performed between this mapping and the contract difference description set of the upstream skill. If the intersection is not empty, it is determined that there is an interface structure conflict.

[0056] Furthermore, business semantic compatibility verification is performed on the basis of interface structure compatibility verification. This is because interface structure verification can only detect direct reference conflicts at the field level and cannot detect implicit conflicts at the semantic level. Specifically, business semantic verification targets the following two types of conflicts: When the name and type of an output field of an upstream skill remain unchanged, but its calculation method (i.e., business meaning) changes, the downstream skill can still parse the field normally, but the calculation result will produce a systematic deviation. By comparing the "field business meaning description" text in the skill contract document, the keyword difference detection method is used to identify the change in calculation method and mark it as a semantic warning to prompt manual confirmation. Regulatory rule library version conflict: Compliance skills rely on external regulatory rule libraries (such as credit policy rule libraries). During verification, the rule library version number declared in the skill contract is checked to see if it is within the scope of the valid rule library version list maintained by the platform. If the rule library version that the skill relies on has expired or has been marked as invalid by the regulatory agency, a compliance version conflict is determined to exist.

[0057] Furthermore, data distribution compatibility verification targets skill nodes that rely on machine learning models (such as credit scoring models or risk rating models). These models are trained based on specific feature distribution assumptions. When the output fields of upstream skills undergo enumeration value expansion or numerical range changes, even if the interface structure is compatible, it may cause the input feature distribution of downstream models to shift, resulting in distorted model output results. This can be determined by comparing the feature distribution constraints declared in the skill contract (including the expected mean, standard deviation, maximum and minimum value range of each input feature) with the output distribution description of the corresponding output fields in the new version of the upstream skill contract, thus identifying whether there is a violation of distribution constraints that causes compatibility conflicts.

[0058] As can be seen, through the above optional embodiments, by traversing the dependency graph to read the input source mapping recorded in the contract document and performing an intersection operation with the contract difference description to determine interface structure conflicts, the identification of dependency relationships is transformed from manual maintenance of ledgers to automated graph analysis based on contract documents. This can fully cover explicit and implicit dependency links across teams and agents, accurately identify candidate nodes affected by disruptive changes through intersection operations, solve the problems of incomplete coverage by manual ledgers and time lag in high-frequency iteration scenarios, significantly improve the accuracy and response speed of fault location during large-scale skill node updates, and identify changes in business caliber by performing keyword difference detection on the field business description text and prompting manual review, thus solving the problem that relying solely on version numbers cannot perceive semantic conflicts at the interface field level. The problem is that by comparing the rule base version number declared in the contract with the list of valid rule base versions, it proactively identifies compliance version conflicts caused by the expiration of the rule base. This dual verification mechanism works together from the two dimensions of semantic understanding and rule compliance to make up for the perception defects of business semantics. It can effectively curb silent events and ensure the actual consistency between business logic execution and regulatory compliance requirements. By targeting skill nodes that rely on machine learning models, it verifies whether the feature distribution assumptions of input features and output features between their upstream and downstream are matched, and determines that nodes that violate distribution constraints are data distribution conflicts. It breaks through the limitations of traditional interface structures and version verification, and can detect deep compatibility problems caused by the shift in input and output data distribution due to model iteration, thereby improving the operational stability in complex intelligent agent collaborative scenarios.

[0059] As an optional embodiment, the step above, obtaining the conflict impact set corresponding to the skill node by traversing the dependency graph along the dependency chain based on the identification result, includes: Based on the candidate detection nodes identified as having compatibility conflicts in the identification results, the starting point corresponding to the breadth-first traversal of the relationship dependency graph is determined, and the upstream nodes of the candidate detection nodes are visited sequentially along the dependency chain layer by layer. The multidimensional compatibility check is performed on each upstream node until a compatible node is included in the upstream branch of any dependency chain composed of the upstream nodes. Then, the traversal of the upstream branch is stopped, and the identification results of compatibility conflicts corresponding to all upstream nodes in the upstream branch are output. The compatible node is an upstream node that indirectly depends on the skill node corresponding to the compatibility conflict along the dependency chain and does not have any compatibility conflict. The identification results of compatibility conflicts corresponding to the candidate detection nodes and the identification results of compatibility conflicts corresponding to the upstream nodes are combined to obtain the conflict impact set of the relationship dependency graph in the update round.

[0060] Specifically, when the identification results determine that a skill node has a compatibility conflict, the actual impact of the conflict is not limited to the first-level downstream nodes that directly depend on the skill, but also propagates deeper along the dependency chain. The system performs a breadth-first traversal on the dependency graph, starting from the skill node that has changed and expanding upwards layer by layer towards the upstream caller. Therefore, it further collects all potentially affected skill nodes to form a conflict impact set. During the traversal, a compatibility check is performed on each visited node. If a node does not have any conflict with the changed skill (for example, although the node indirectly depends on the changed skill, its input source mapping does not reference any affected fields), then the skill node is marked as compatible, and its upstream nodes are no longer traversed, thereby pruning and reducing unnecessary computation.

[0061] As can be seen, through the above optional embodiments, by starting with the compatibility conflict node, performing a breadth-first traversal along the dependency graph and performing multi-dimensional compatibility checks layer by layer upstream, and using the appearance of a compatible node as the traversal stopping condition, the true impact range of the conflict caused by the version update propagating upstream along the dependency chain can be accurately located. This automatically excludes upstream branches that, although having indirect dependencies, are not actually affected by the conflict. Compared to existing technologies that require manual investigation or full scanning of all skill nodes due to a lack of effective impact analysis methods, this significantly reduces the scope of fault investigation and significantly shortens the conflict impact assessment time within the version release window. To address the shortcomings of existing technologies in the timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed, this technology effectively controls the depth and breadth of dependency link analysis by setting a stopping condition (i.e., terminating the traversal of a branch once a compatible node unaffected by any conflict is found in the upstream branch). This prevents meaningless full-path backtracking along long links, saves computing resources, and removes upstream noise nodes that are not actually affected from the generated conflict impact set. This makes the subsequent impact reports more focused and accurate, facilitating operations and development personnel to make targeted remediation decisions quickly, thereby responding to the urgent need to improve the efficiency of regulatory compliance update response.

[0062] As an optional embodiment, the step above, calculating the global impact score of each skill node in the conflict impact set based on skill node attributes to obtain a conflict impact report, includes: The statistical period for all skill nodes is determined based on the business update cycle. The call frequency of each skill node in the conflict impact report is collected based on the statistical period. The average call frequency of all skill nodes in the relationship dependency graph within the statistical period is calculated. The frequency ratio between the call frequency and the average call frequency is calculated to obtain the activity score corresponding to each skill node in the conflict impact report. Based on the business semantics, determine the business level score of each skill node in the conflict impact report on its dependency chain; Based on the identification results of the compatibility conflicts, determine the conflict type score corresponding to each skill node in the conflict impact report; For any skill node in the conflict impact report, an initial impact score is determined based on the business level score and conflict type score. The number of times the skill node affects downstream nodes along its dependency chain is determined. The initial impact score of the skill node is accumulated based on a preset impact step factor and the number of impacts. The impact step factor of the layer-by-layer accumulation process is adjusted according to the initial impact score corresponding to the skill node at the current level. The dependency impact score corresponding to each skill node in the conflict impact report is output. The global impact score of each skill node in the conflict impact set is calculated by weighted summation of the activity score, business level score, conflict type score, and dependency impact score. The skill nodes in the conflict impact set are sorted from high to low according to the global impact score to obtain a skill node list. For each skill node in the skill node list, the conflict type and impact propagation path of the skill node are recorded. The impact propagation path corresponds to other skill nodes affected by the skill node. The release recommendation corresponding to the compatibility conflict in the current round is determined according to the recorded content of the skill node list. A conflict impact report is obtained based on the skill node list and the release recommendation.

[0063] Optionally, the business level score can be set with a business level coefficient of 1, 0.6, and 0.3 based on the core level, important level, and ordinary level, or it can be specifically set according to other business level classifications; Optionally, the conflict type score can be a conflict type coefficient with a value of 1, 0.5, or 0.8 set based on the conflict type of disruptive change, semantic warning, and compliant version conflict, or it can be specifically set according to other conflict type classifications.

[0064] As can be seen, through the above optional embodiments, by comprehensively considering multiple dimensions such as call frequency activity, business semantic level, conflict type severity, and dependency propagation impact, each skill node in the conflict impact concentration is weighted, scored, and ranked. Compared with the shortcomings of existing technologies that cannot assess the degree of dynamic impact, this quantitative scoring achieves accurate measurement and visual ranking of conflict impact. This allows operations and maintenance personnel to quickly focus on the key conflict node with the highest global impact score within a limited release window when thousands of skill nodes are simultaneously changing, greatly shortening the fault assessment and decision response time. It solves the problems of blind decision-making and delayed fault location in large-scale update scenarios. By statistically analyzing the number of times downstream nodes are affected, and dynamically adjusting the impact step size factor based on the initial impact score of the affected level nodes, it simulates the non-linear amplification effect of severe conflicts or core node conflicts on downstream nodes in actual business, avoiding the conflict impact caused by the lack of propagation simulation. Whether underestimated or overestimated, the resulting global impact score more accurately reflects the potential harm of conflicts to the overall business chain, providing a more scientific quantitative basis for subsequent release decisions (such as blocking releases, partial rollbacks, or forced upgrades). This effectively reduces silent business failures caused by inaccurate assessments. By including global impact ranking in the final output skill node list, it also clearly marks the specific conflict type and impact propagation path of each node, avoiding the information silo problem of not being able to determine the scope of impact due to version number verification alone. At the same time, it reduces the blind spots of implicit dependency chains that cannot be traced by manual ledgers. By providing a clear impact topology view and root cause analysis clues, it significantly reduces the cross-team communication costs and troubleshooting complexity caused by skill nodes being developed and iterated by multiple teams in the intelligent agent platform. This enables development and operations personnel to quickly formulate targeted remediation strategies and release decisions, realizing a perceptible decision-making process from passive perception to proactive discovery of compatibility conflicts.

[0065] As an optional embodiment, the step above, determining the release decision for the current update based on the conflict impact report, includes: Based on the conflict impact set, global impact score, and release recommendation corresponding to the conflict impact report, a judgment is made based on preset release conditions, and a release decision that meets the release conditions is selected to be executed. The process of issuing the decision is specifically as follows: The release decision adopts a phased canary release strategy, which distributes the request traffic for each updated skill node proportionally and records the operation indicators independently; During the canary release process, collect the average values ​​of business output fields and request exception rate of skill nodes before and after the update within the canary window period; Calculate the deviation of the mean value of the business output field from its previous value to obtain the business consistency index. The difference between the request anomaly rate and the rate before the update is calculated to obtain the anomaly rate index; Based on the business consistency index and the anomaly rate index, a weighted calculation is performed using an indicator function to obtain the rollback score in the release decision. When the rollback score is less than the preset rollback threshold, a rollback decision is triggered and a rollback report is generated. When the rollback score is not less than the preset rollback threshold, it is determined whether the gray release duration meets the observation window threshold. If so, the request traffic ratio is increased according to the preset traffic step size to complete the full switch of skill node update. If not, a second judgment is made based on the rollback score when the gray release duration meets the observation window threshold.

[0066] Optionally, request exceptions can be parsing failures, timeouts, null outputs, etc., and this application does not specify any particular exceptions.

[0067] Specifically, the indicator function determines whether the business consistency metric exceeds the consistency threshold and the anomaly rate threshold by using the consistency threshold and anomaly rate threshold, respectively. The business consistency threshold is initially set to 2% of the output field deviation, and the anomaly rate threshold is initially set to 0.5%. These values ​​can be adjusted according to release requirements. Essentially, when the business consistency metric exceeds the consistency threshold, the indicator function outputs 1; otherwise, it outputs 0. The indicator function for the anomaly rate metric is not elaborated upon further. In the weighted calculation of the indicator function, the initial weights for both the business consistency metric and the anomaly rate metric are 0.5, which can also be adjusted according to release requirements.

[0068] As can be seen, through the above optional embodiments, by adopting a proportional canary release strategy, the operational metrics of updated skill nodes are independently recorded within a limited traffic window, and the deviation of the average value of business output fields and the difference in request anomaly rate are calculated. This addresses the release blocking problem caused by insufficient timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed. Through small-scale verification under real business traffic, potential functional or performance anomalies can be captured in time before a full switch, avoiding global business interruption caused by large-scale synchronous changes. This significantly improves the security and controllability of version releases while meeting regulatory compliance update timeliness requirements. By monitoring the request anomaly rate and introducing the deviation of the average value of business output fields as a business consistency indicator, even if the interface structure is compatible but the business logic undergoes unexpected changes (such as...), the system can effectively address the release blocking problem caused by insufficient timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed. In the silent failure scenario (calculation result deviation), abnormal fluctuations in business consistency indicators can be accurately captured and converted into rollback scores, thereby triggering protective rollback decisions. This effectively prevents the risk of effective detection but business failure, ensuring the business correctness of intelligent agent collaborative tasks. By setting clear rollback score thresholds and gray-scale observation window durations, when the rollback score is lower than the threshold, a rollback is immediately triggered and a report is generated. Otherwise, the traffic ratio is dynamically increased until a full switch is achieved based on the observation window duration and secondary judgment. This closed-loop decision-making mechanism transforms release risk control from passive manual experience judgment to proactive data-driven decision-making. It can automatically, quickly, and accurately make decisions on continuing release or emergency rollback in scenarios of high business scale expansion and high frequency version iteration, significantly reducing operation and maintenance manpower costs and decision delays, and ensuring the continuous reliability of intelligent agent platform version updates.

[0069] In summary, the technical solutions disclosed in the embodiments of the present invention have the following advantages: 1. Significantly improved real-time performance: Contract discrepancy extraction and conflict impact analysis are completed within seconds after the skill version is submitted, reducing the original manual review cycle from 3 to 5 working days to seconds. In particular, it supports financial institutions to complete the hot update of compliance skills within 24 hours after regulatory policy changes.

[0070] 2. Comprehensive conflict detection coverage: Through joint verification of three dimensions—interface structure, business semantics, and data distribution—it covers implicit conflict types that existing semantic version number rules cannot identify, eliminating the inspection loopholes where development teams miss the review process due to version number marking.

[0071] 3. Precise Scope of Impact: Based on dependency graph topology propagation analysis, it accurately identifies all affected nodes in multi-layer dependency chains, avoiding the waste of resources in full regression testing, and compressing the test scope to the actual set of affected nodes, effectively reducing unnecessary test execution by about 70%.

[0072] 4. Minimize the impact of production failures: The gray-scale verification mechanism controls the initial impact of the new version to 1% of the traffic. Through real-time health monitoring and automatic rollback, the impact of silent failures on business operations is compressed from the full volume to a very small proportion during the gray-scale probe phase, effectively ensuring business continuity.

[0073] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a compatibility conflict detection system for intelligent agent skill nodes disclosed in an embodiment of the present invention. Figure 2 The described compatibility conflict detection system for agent skill nodes can be applied to data processing systems / data processing devices / data processing servers (including local processing servers or cloud processing servers). For example... Figure 2 As shown, the compatibility conflict detection system for agent skill nodes may include: The contract management module 201 is used to collect skill node information sets to generate skill contract documents, maintain node relationships in the relation dependency graph of each round of updates based on the skill contract documents, and extract the contract difference description set of each round of updates. The multidimensional verification module 202 is used to determine candidate detection nodes from the relation dependency graph based on the contract difference description set, and to perform multidimensional compatibility verification on the candidate detection nodes based on interface structure, business semantics and data distribution to obtain the identification result of compatibility conflict. The conflict detection module 203 is used to traverse the dependency graph along the dependency chain according to the identification results to obtain the conflict impact set corresponding to the skill node, calculate the global impact score of each skill node in the conflict impact set based on the skill node attributes to obtain the conflict impact report, and determine the release decision of the current round of update based on the conflict impact report.

[0074] As can be seen, the above-described embodiments of the invention generate skill contract documents by collecting skill node information sets, and maintain node relationships in the dependency graph of each round of updates based on the skill contract documents. Compared with manual ledger methods, this can automatically and completely capture explicit and implicit dependencies across teams and agents, solving the problem of incomplete coverage of manually maintained version dependency tables. Furthermore, by using a contract difference description set to drive update detection, it can respond in real time to dependency changes in high-frequency iteration scenarios, significantly improving the timeliness of compatibility detection and avoiding version conflicts caused by delays in manual updates. Through multi-dimensional compatibility verification, it overcomes the limitations of dependency version number conventions and can accurately identify interface words. This system effectively reduces silent events where detection is effective but business operations fail due to a lack of semantic awareness, thereby enhancing compatibility assurance at the business semantic level. By traversing the dependency graph along the dependency chain to generate a conflict impact set and combining it with skill node attributes to calculate an impact score and generate a conflict impact report, it assists in release decisions. In scenarios with large-scale synchronous changes to skill nodes, it can quickly and accurately locate the scope and degree of impact of affected nodes, avoiding version release process blockages caused by insufficient timeliness and accuracy of fault location, and improving the response efficiency and release stability to regulatory compliance update requirements.

[0075] As an optional embodiment, candidate detection nodes are determined from the relation dependency graph based on a set of contract difference descriptions, including: Extract the associated skill nodes with disruptive changes from the set of contract difference descriptions corresponding to each round of updates; Based on the directed edge relationships in the relation dependency graph, determine the downstream nodes of the call dependency relationships corresponding to the associated skill nodes, and identify the associated skill nodes and downstream nodes as candidate detection nodes. The destructive changes include deleting, modifying, expanding, and shrinking fields, individual values, and value ranges in the skill contract.

[0076] As can be seen, through the above optional embodiments, by extracting destructive changes from the contract difference description set and specifically defining the operation type of destructive changes, it is possible to proactively capture illegal change behaviors that cause silent business failures. This effectively compensates for the lack of business semantic awareness, significantly reduces the occurrence rate of silent events, enhances the ability to accurately identify destructive changes, and provides high-quality input for subsequent business semantic conflict verification. After identifying the skill nodes associated with destructive changes, the downstream calling nodes are accurately determined as candidate detection nodes based on the directed edge relationships of the dependency graph, rather than redundantly scanning all skill nodes. By focusing verification computing resources on the dependency links actually affected by destructive changes, the detection scope is narrowed to ensure response timeliness in large-scale update scenarios. This greatly improves the response speed of compatibility testing in the data preparation stage, provides high-quality input for high-accuracy fault location in the subsequent process, and ensures the process within the version release window and the ability to respond quickly to regulatory compliance requirements.

[0077] As an optional implementation, multi-dimensional compatibility verification is performed on candidate detection nodes based on interface structure, business semantics, and data distribution to obtain the identification results of compatibility conflicts, including: Interface structure compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the input source mapping of the contract document record in the candidate detection node, performing an intersection operation on the input source mapping and the contract difference description, and determining the candidate detection nodes whose intersection operation result is not an empty set as having an interface structure conflict. Business semantic compatibility verification involves traversing candidate detection nodes in the relational dependency graph, reading the business description text of the fields recorded in the contract document of the candidate detection node, identifying the change in the scope corresponding to the business meaning through keyword difference detection, marking the candidate detection node that is determined to be a change in scope as a business semantic conflict to prompt for composite confirmation, and comparing and matching the rule base version number declared in the contract in the candidate detection node with the list of valid rule base versions, and determining that the candidate detection node that does not match has a compliant version conflict. Data distribution compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the feature distribution constraints declared in the contract documents of the candidate detection nodes, comparing the output distribution description of the associated skill nodes in the candidate detection nodes with the feature distribution constraints of the downstream nodes corresponding to the associated skill nodes, and determining the candidate detection nodes that violate the distribution constraints as data distribution conflicts. Specifically, the feature distribution constraints are the feature distribution assumptions of the input and output features of the skill nodes that depend on the machine learning model during model training. By merging interface structure conflicts, business semantic conflicts, compliance version conflicts, and data distribution conflicts, the identification results of compatibility conflicts corresponding to candidate detection nodes are obtained.

[0078] As can be seen, through the above optional embodiments, by traversing the dependency graph to read the input source mapping recorded in the contract document and performing an intersection operation with the contract difference description to determine interface structure conflicts, the identification of dependency relationships is transformed from manual maintenance of ledgers to automated graph analysis based on contract documents. This can fully cover explicit and implicit dependency links across teams and agents, accurately identify candidate nodes affected by disruptive changes through intersection operations, solve the problems of incomplete coverage by manual ledgers and time lag in high-frequency iteration scenarios, significantly improve the accuracy and response speed of fault location during large-scale skill node updates, and identify changes in business caliber by performing keyword difference detection on the field business description text and prompting manual review, thus solving the problem that relying solely on version numbers cannot perceive semantic conflicts at the interface field level. The problem is that by comparing the rule base version number declared in the contract with the list of valid rule base versions, it proactively identifies compliance version conflicts caused by the expiration of the rule base. This dual verification mechanism works together from the two dimensions of semantic understanding and rule compliance to make up for the perception defects of business semantics. It can effectively curb silent events and ensure the actual consistency between business logic execution and regulatory compliance requirements. By targeting skill nodes that rely on machine learning models, it verifies whether the feature distribution assumptions of input features and output features between their upstream and downstream are matched, and determines that nodes that violate distribution constraints are data distribution conflicts. It breaks through the limitations of traditional interface structures and version verification, and can detect deep compatibility problems caused by the shift in input and output data distribution due to model iteration, thereby improving the operational stability in complex intelligent agent collaborative scenarios.

[0079] As an optional implementation, the conflict impact set corresponding to the skill node is obtained by traversing the dependency graph along the dependency chain based on the identification results, including: Based on the candidate detection nodes identified as having compatibility conflicts in the identification results, the starting point corresponding to the breadth-first traversal of the relationship dependency graph is determined, and the upstream nodes of the candidate detection nodes are visited sequentially along the dependency chain layer by layer. The multidimensional compatibility check is performed on each upstream node until a compatible node is included in the upstream branch of any dependency chain composed of the upstream nodes. Then, the traversal of the upstream branch is stopped, and the identification results of compatibility conflicts corresponding to all upstream nodes in the upstream branch are output. The compatible node is an upstream node that indirectly depends on the skill node corresponding to the compatibility conflict along the dependency chain and does not have any compatibility conflict. The identification results of compatibility conflicts corresponding to the candidate detection nodes and the identification results of compatibility conflicts corresponding to the upstream nodes are combined to obtain the conflict impact set of the relationship dependency graph in the update round.

[0080] As can be seen, through the above optional embodiments, by starting with the compatibility conflict node, performing a breadth-first traversal along the dependency graph and performing multi-dimensional compatibility checks layer by layer upstream, and using the appearance of a compatible node as the traversal stopping condition, the true impact range of the conflict caused by the version update propagating upstream along the dependency chain can be accurately located. This automatically excludes upstream branches that, although having indirect dependencies, are not actually affected by the conflict. Compared to existing technologies that require manual investigation or full scanning of all skill nodes due to a lack of effective impact analysis methods, this significantly reduces the scope of fault investigation and significantly shortens the conflict impact assessment time within the version release window. To address the shortcomings of existing technologies in the timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed, this technology effectively controls the depth and breadth of dependency link analysis by setting a stopping condition (i.e., terminating the traversal of a branch once a compatible node unaffected by any conflict is found in the upstream branch). This prevents meaningless full-path backtracking along long links, saves computing resources, and removes upstream noise nodes that are not actually affected from the generated conflict impact set. This makes the subsequent impact reports more focused and accurate, facilitating operations and development personnel to make targeted remediation decisions quickly, thereby responding to the urgent need to improve the efficiency of regulatory compliance update response.

[0081] As an optional implementation, a conflict impact report is obtained by calculating the global impact score of each skill node in the conflict impact set based on the skill node attributes, including: The statistical period for all skill nodes is determined based on the business update cycle. The call frequency of each skill node in the conflict impact report is collected based on the statistical period. The average call frequency of all skill nodes in the relationship dependency graph within the statistical period is calculated. The frequency ratio between the call frequency and the average call frequency is calculated to obtain the activity score corresponding to each skill node in the conflict impact report. Based on the business semantics, determine the business level score of each skill node in the conflict impact report on its dependency chain; Based on the identification results of the compatibility conflicts, determine the conflict type score corresponding to each skill node in the conflict impact report; For any skill node in the conflict impact report, an initial impact score is determined based on the business level score and conflict type score. The number of times the skill node affects downstream nodes along its dependency chain is determined. The initial impact score of the skill node is accumulated based on a preset impact step factor and the number of impacts. The impact step factor of the layer-by-layer accumulation process is adjusted according to the initial impact score corresponding to the skill node at the current level. The dependency impact score corresponding to each skill node in the conflict impact report is output. The global impact score of each skill node in the conflict impact set is calculated by weighted summation of the activity score, business level score, conflict type score, and dependency impact score. The skill nodes in the conflict impact set are sorted from high to low according to the global impact score to obtain a skill node list. For each skill node in the skill node list, the conflict type and impact propagation path of the skill node are recorded. The impact propagation path corresponds to other skill nodes affected by the skill node. The release recommendation corresponding to the compatibility conflict in the current round is determined according to the recorded content of the skill node list. A conflict impact report is obtained based on the skill node list and the release recommendation.

[0082] As can be seen from the above optional embodiments, by comprehensively considering multiple dimensions such as call frequency activity, business semantic level, conflict type severity, and dependency propagation impact, each skill node in the conflict impact concentration is weighted, scored, and ranked. Compared to the shortcomings of existing technologies that cannot assess the degree of dynamic impact, this quantitative scoring achieves accurate measurement and visual ranking of conflict impact. This allows operations and maintenance personnel to quickly focus on the key conflict node with the highest global impact score within a limited release window when thousands of skill nodes are simultaneously changing, greatly shortening the fault assessment and decision response time. It solves the problems of blind decision-making and delayed fault location in large-scale update scenarios. By statistically analyzing the number of times downstream nodes are affected, and dynamically adjusting the impact step size factor based on the initial impact score of the affected level nodes, it simulates the non-linear amplification effect of severe conflicts or core node conflicts on downstream processes in actual business, avoiding the lack of propagation simulation. The resulting global impact score more accurately reflects the potential harm of the conflict to the overall business chain, thus providing a more scientific quantitative basis for subsequent release decisions (such as blocking releases, partial rollbacks, or forced upgrades). This effectively reduces silent business failures caused by inaccurate assessments. By including global impact ranking in the final output skill node list, the specific conflict type and its impact propagation path for each node are clearly marked, avoiding the information silo problem of not being able to determine the scope of impact due to version number verification alone. At the same time, it reduces the blind spots of implicit dependency chains that cannot be traced by manual ledgers. By providing a clear impact topology view and root cause analysis clues, it significantly reduces the cross-team communication costs and troubleshooting complexity caused by skill nodes being developed and iterated by multiple teams in the intelligent agent platform. This enables development and operation personnel to quickly formulate targeted remediation strategies and release decisions, realizing a perceptible decision-making process from passive perception to proactive discovery of compatibility conflicts.

[0083] As an optional embodiment, the system may further include a grayscale control module for performing the release, the grayscale control module being used to perform the following steps: Based on the conflict impact set, global impact score, and release recommendation corresponding to the conflict impact report, a judgment is made based on preset release conditions, and a release decision that meets the release conditions is selected to be executed. The process of issuing the decision is specifically as follows: The release decision adopts a phased canary release strategy, which distributes the request traffic for each updated skill node proportionally and records the operation indicators independently; During the canary release process, collect the average values ​​of business output fields and request exception rate of skill nodes before and after the update within the canary window period; Calculate the deviation of the mean value of the business output field from its previous value to obtain the business consistency index. The difference between the request anomaly rate and the rate before the update is calculated to obtain the anomaly rate index; Based on the business consistency index and the anomaly rate index, a weighted calculation is performed using an indicator function to obtain the rollback score in the release decision. When the rollback score is less than the preset rollback threshold, a rollback decision is triggered and a rollback report is generated. When the rollback score is not less than the preset rollback threshold, it is determined whether the gray release duration meets the observation window threshold. If so, the request traffic ratio is increased according to the preset traffic step size to complete the full switch of skill node update. If not, a second judgment is made based on the rollback score when the gray release duration meets the observation window threshold.

[0084] As can be seen, through the above optional embodiments, by adopting a proportional canary release strategy, the operational metrics of updated skill nodes are independently recorded within a limited traffic window, and the deviation of the average value of business output fields and the difference in request anomaly rate are calculated. This addresses the release blocking problem caused by insufficient timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed. Through small-scale verification under real business traffic, potential functional or performance anomalies can be captured in time before a full switch, avoiding global business interruption caused by large-scale synchronous changes. This significantly improves the security and controllability of version releases while meeting regulatory compliance update timeliness requirements. By monitoring the request anomaly rate and introducing the deviation of the average value of business output fields as a business consistency indicator, even if the interface structure is compatible but the business logic undergoes unexpected changes (such as...), the system can effectively address the release blocking problem caused by insufficient timeliness and accuracy of fault location when thousands of skill nodes are simultaneously changed. In silent failure scenarios (such as calculation result deviations), abnormal fluctuations in business consistency indicators can be accurately captured and converted into rollback scores, thereby triggering protective rollback decisions. This effectively prevents the risk of detection being effective but business failures, ensuring the business correctness of intelligent agent collaborative tasks. By setting clear rollback score thresholds and gray-scale observation window durations, when the rollback score falls below the threshold, a rollback is immediately triggered and a report is generated. Otherwise, the traffic ratio is dynamically increased based on the observation window duration and secondary judgments until a full switch is achieved. This closed-loop decision-making mechanism transforms release risk control from passive manual experience judgment to proactive data-driven decision-making. It can automatically, quickly, and accurately make decisions to continue release or urgently roll back in scenarios of high business scale expansion and high-frequency version iteration, significantly reducing operation and maintenance manpower costs and decision delays, and ensuring the continuous reliability of intelligent agent platform version updates. As an optional implementation, [further details are needed].

[0085] Example 3 Please see Figure 3 , Figure 3 This is another compatibility conflict detection system for intelligent agent skill nodes disclosed in the embodiments of the present invention. Figure 3 The described compatibility conflict detection system for agent skill nodes is applied in a data processing system / data processing device / data processing server (wherein, the server includes a local processing server or a cloud processing server). Figure 3 As shown, the compatibility conflict detection system for agent skill nodes may include: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the compatibility conflict detection method for intelligent agent skill nodes described in Embodiment 1.

[0086] Example 4 This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps of the compatibility conflict detection method for intelligent agent skill nodes described in Embodiment 1.

[0087] Example 5 This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps of the compatibility conflict detection method for intelligent agent skill nodes described in Embodiment 1.

[0088] The foregoing has described specific embodiments of this specification; other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0090] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0091] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0096] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0097] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0098] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0099] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0100] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0101] Finally, it should be noted that the compatibility conflict detection method and system for intelligent agent skill nodes disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A compatibility conflict detection method for intelligent agent skill nodes, characterized in that, The method includes: Collect skill node information sets to generate skill contract documents, maintain node relationships in the relation dependency graph of each round of updates based on the skill contract documents, and extract the contract difference description set of each round of updates; Candidate detection nodes are determined from the relationship dependency graph based on the contract difference description set. Multi-dimensional compatibility verification of the candidate detection nodes is performed based on interface structure, business semantics and data distribution to obtain the identification results of compatibility conflicts. Based on the identification results, the dependency graph is traversed along the dependency chain to obtain the conflict impact set corresponding to the skill node. The global impact score of each skill node in the conflict impact set is calculated based on the skill node attributes to obtain the conflict impact report. The release decision for the current round of updates is determined based on the conflict impact report.

2. The compatibility conflict detection method for intelligent agent skill nodes according to claim 1, characterized in that, The skill node information set includes a list of input parameters, a list of output fields, field names, data types, value ranges, enumeration value sets, precision thresholds, dependent rule base names, and version numbers for all skill nodes. The skill contract document is a structured document generated by preprocessing the skill node information set. It is stored in the contract registration center of the intelligent agent platform through a key-value structure. The skill contract document is automatically archived and retained when the version of the skill node is updated in each round.

3. The compatibility conflict detection method for intelligent agent skill nodes according to claim 1, characterized in that, The relationship dependency graph is a directed acyclic graph, where each node represents a skill node, and each directed edge between nodes represents the dependency relationship between skill node calls. The relationship dependency graph is updated in real time through message passing when skill nodes are registered, version updates are released, and dependency relationships change. The contract difference description set includes destructive changes, extensible changes, and internal changes. Any contract difference description is obtained by identifying and extracting the field differences and / or semantic differences between the skill contract documents before and after the update for the corresponding skill node.

4. The compatibility conflict detection method for intelligent agent skill nodes according to claim 3, characterized in that, Candidate detection nodes are determined from the dependency graph based on the contract difference description set, including: Extract the associated skill nodes with disruptive changes from the set of contract difference descriptions corresponding to each round of updates; Based on the directed edge relationships in the relation dependency graph, determine the downstream nodes of the call dependency relationships corresponding to the associated skill nodes, and identify the associated skill nodes and downstream nodes as candidate detection nodes. The destructive changes include deleting, modifying, expanding, and shrinking fields, individual values, and value ranges in the skill contract.

5. The compatibility conflict detection method for intelligent agent skill nodes according to claim 1, characterized in that, Based on interface structure, business semantics, and data distribution, multi-dimensional compatibility verification is performed on candidate detection nodes to obtain the identification results of compatibility conflicts, including: Interface structure compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the input source mapping of the contract document record in the candidate detection node, performing an intersection operation on the input source mapping and the contract difference description, and determining the candidate detection nodes whose intersection operation result is not an empty set as having an interface structure conflict. Business semantic compatibility verification involves traversing candidate detection nodes in the relational dependency graph, reading the business description text of the fields recorded in the contract document of the candidate detection node, identifying the change in the scope corresponding to the business meaning through keyword difference detection, marking the candidate detection node that is determined to be a change in scope as a business semantic conflict to prompt for composite confirmation, and comparing and matching the rule base version number declared in the contract in the candidate detection node with the list of valid rule base versions, and determining that the candidate detection node that does not match has a compliant version conflict. Data distribution compatibility verification involves traversing the candidate detection nodes in the relational dependency graph, reading the feature distribution constraints declared in the contract documents of the candidate detection nodes, comparing the output distribution description of the associated skill nodes in the candidate detection nodes with the feature distribution constraints of the downstream nodes corresponding to the associated skill nodes, and determining the candidate detection nodes that violate the distribution constraints as data distribution conflicts. Specifically, the feature distribution constraints are the feature distribution assumptions of the input and output features of the skill nodes that depend on the machine learning model during model training. By merging interface structure conflicts, business semantic conflicts, compliance version conflicts, and data distribution conflicts, the identification results of compatibility conflicts corresponding to candidate detection nodes are obtained.

6. The compatibility conflict detection method for intelligent agent skill nodes according to claim 5, characterized in that, Based on the identification results, the conflict impact set corresponding to the skill node is obtained by traversing the dependency graph along the dependency chain, including: Based on the candidate detection nodes identified as having compatibility conflicts in the identification results, the starting point corresponding to the breadth-first traversal of the relationship dependency graph is determined, and the upstream nodes of the candidate detection nodes are visited sequentially along the dependency chain layer by layer. The multidimensional compatibility check is performed on each upstream node until a compatible node is included in the upstream branch of any dependency chain composed of the upstream nodes. Then, the traversal of the upstream branch is stopped, and the identification results of compatibility conflicts corresponding to all upstream nodes in the upstream branch are output. The compatible node is an upstream node that indirectly depends on the skill node corresponding to the compatibility conflict along the dependency chain and does not have any compatibility conflict. The identification results of compatibility conflicts corresponding to the candidate detection nodes and the identification results of compatibility conflicts corresponding to the upstream nodes are combined to obtain the conflict impact set of the relationship dependency graph in the update round.

7. The compatibility conflict detection method for intelligent agent skill nodes according to claim 1, characterized in that, The conflict impact report is obtained by calculating the global impact score of each skill node in the conflict impact set based on the skill node attributes, including: The statistical period for all skill nodes is determined based on the business update cycle. The call frequency of each skill node in the conflict impact report is collected based on the statistical period. The average call frequency of all skill nodes in the relationship dependency graph within the statistical period is calculated. The frequency ratio between the call frequency and the average call frequency is calculated to obtain the activity score corresponding to each skill node in the conflict impact report. Based on the business semantics, determine the business level score of each skill node in the conflict impact report on its dependency chain; Based on the identification results of the compatibility conflicts, determine the conflict type score corresponding to each skill node in the conflict impact report; For any skill node in the conflict impact report, an initial impact score is determined based on the business level score and conflict type score. The number of times the skill node affects downstream nodes along its dependency chain is determined. The initial impact score of the skill node is accumulated based on a preset impact step factor and the number of impacts. The impact step factor of the layer-by-layer accumulation process is adjusted according to the initial impact score corresponding to the skill node at the current level. The dependency impact score corresponding to each skill node in the conflict impact report is output. The global impact score for each skill node in the conflict impact set is calculated by weighting and summing the activity score, business level score, conflict type score, and dependency impact score. The skill nodes in the conflict impact set are sorted from high to low according to the global impact score to obtain a skill node list. For each skill node in the skill node list, the conflict type and impact propagation path of the skill node are recorded. The impact propagation path corresponds to other skill nodes affected by the skill node. The release recommendation corresponding to the compatibility conflict in the current round is determined according to the recorded content of the skill node list. A conflict impact report is obtained based on the skill node list and the release recommendation.

8. The compatibility conflict detection method for intelligent agent skill nodes according to claim 1, characterized in that, The decision to release the current update is determined based on the conflict impact report, including: Based on the conflict impact set, global impact score, and release recommendation corresponding to the conflict impact report, a judgment is made based on preset release conditions, and a release decision that meets the release conditions is selected to be executed. The process of issuing the decision is specifically as follows: The release decision adopts a phased canary release strategy, which distributes the request traffic for each updated skill node proportionally and records the operation indicators independently; During the canary release process, collect the average values ​​of business output fields and request exception rate of skill nodes before and after the update within the canary window period; Calculate the deviation of the mean value of the business output field from its previous value to obtain the business consistency index. The difference between the request anomaly rate and the rate before the update is calculated to obtain the anomaly rate index; Based on the business consistency index and the anomaly rate index, a weighted calculation is performed using an indicator function to obtain the rollback score in the release decision. When the rollback score is less than the preset rollback threshold, a rollback decision is triggered and a rollback report is generated. When the rollback score is not less than the preset rollback threshold, it is determined whether the gray release duration meets the observation window threshold. If so, the request traffic ratio is increased according to the preset traffic step size to complete the full switch of skill node update. If not, a second judgment is made based on the rollback score when the gray release duration meets the observation window threshold.

9. A compatibility conflict detection system for intelligent agent skill nodes, characterized in that, The system includes: The contract management module is used to collect skill node information sets to generate skill contract documents, maintain node relationships in the dependency graph of each round of updates based on the skill contract documents, and extract the contract difference description set of each round of updates. The multi-dimensional verification module is used to determine candidate detection nodes from the relation dependency graph based on the contract difference description set, and to perform multi-dimensional compatibility verification on the candidate detection nodes based on interface structure, business semantics and data distribution to obtain the identification results of compatibility conflicts. The conflict detection module is used to traverse the dependency graph along the dependency chain based on the identification results to obtain the conflict impact set corresponding to the skill node, calculate the global impact score of each skill node in the conflict impact set based on the skill node attributes to obtain the conflict impact report, and determine the release decision for the current round of updates based on the conflict impact report.

10. A compatibility conflict detection system for intelligent agent skill nodes, characterized in that, The system includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the compatibility conflict detection method for intelligent agent skill nodes as described in any one of claims 1-8.

Images