Transparent and trustworthy method for skill achievement based on dlf

Abstract

This invention discloses a transparent and trustworthy method for achieving skill work results based on DLF (Device Flow Framework), comprising the following steps: constructing a vertical model dataset classification system and skill metadata templates, and predefining OFD (Obstacle Course Function) electronic credential templates; recording operational behaviors in real time during skill execution; mapping the recorded information to the corresponding OFD templates to generate a trustworthy OFD electronic credential for a single skill result; generating a DLF trustworthy credential set and a DLF visual call overview diagram for multi-step skill chains; and binding the DLF trustworthy credentials with the vertical model results to achieve bidirectional full-link traceability. This invention constructs a trustworthy electronic credential mechanism for the entire process of skill call. It ensures that each skill call obtains a unique, tamper-proof, and trustworthy identity, fundamentally solving the problems of unclear sources and untraceable processes in skill work results.

Description

Technical Field

[0001] This invention belongs to the field of document technology, specifically a transparent and reliable method for achieving Skill work results based on DLF. Background Technology

[0002] A Skill is a standardized package of capabilities, typically existing as a folder containing specific files. It achieves machine parsing through a fixed structure (such as a SKILL.md file) and can be linked to external scripts and resources, combining the reasoning capabilities of large models with the deterministic execution of programs. In specific application areas, when a Skill is invoked, the content of the skill comes from multiple sources. Therefore, the source of the work output generated through a Skill may be unclear, making the process of creating the output untraceable. Summary of the Invention

[0003] The purpose of this invention is to provide a transparent and reliable method for realizing Skill work results based on DLF, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: a transparent and reliable method for realizing Skill work results based on DLF, comprising the following steps: Step S1: Construct a vertical model dataset classification system and Skill metadata templates, and predefine OFD electronic voucher templates according to different skill categories; Step S2: Record the operation behavior, calling context, and intermediate result snapshots in real time during the execution of the Skill; Step S3: Map the recorded information to the corresponding OFD template, and after encryption, digital signature and chain hash association, generate a trusted electronic certificate of OFD for a single Skill achievement; Step S4: For a multi-step Skill chain, organize the skill credentials into a DLF trusted credential set according to the arrangement relationship, and apply overall signature and consistency verification to the credential set; Step S5: Generate a multi-dimensional navigation file containing a timeline, logical dependencies, and role behavior chains, supporting transparent display and auditing; Step S6: Generate a DLF visualization overview map to provide a panoramic view of the skill chain execution; Step S7: Bind the DLF trusted credentials to the vertical model results to achieve two-way full-link traceability; Step S8: Construct a standardized DLF voucher output encapsulation format and API interface.

[0005] Preferably, in step S1, a classification system for the vertical model dataset and a skill metadata template are constructed. Based on the data source and business attributes, the datasets involved in the vertical model are divided into five categories: industry professional knowledge data, specific business data, multi-source heterogeneous perception data, model inference process data, and skill arrangement and dependency data. Each major category is further subdivided into subcategories and minor categories to form a complete five-level classification system.

[0006] Preferably, based on a five-level classification system, the data items involved in each skill in the DLF credential set are tagged and stored according to the hierarchical path of "multi-source heterogeneous perception data, medium category, and small category". During tracing, it supports multi-granularity retrieval and filtering by classification level, including: filtering all skill credentials that use this type of data by primary category; locating the skill call of a specific data source by sub-category; when the final result of the vertical model has deviations or anomalies, quickly locking the key skill credentials under a certain type of data path that affects the result. Through the collaborative work of the classification retrieval mechanism and the index table of the DLF credential set, a two-dimensional tracing capability of "classification tree + credential chain" is formed.

[0007] Preferably, in step S1, a Skill metadata template is constructed. For each Skill in the vertical model, a structured metadata model for each Skill is predefined, including basic skill information, interface specifications, dependency declarations, execution constraints, and version associations. A corresponding OFD electronic voucher template is also constructed for each Skill category to distinguish the skill's output type. The output types include skill call type, model inference type, and data operation type. An OFD electronic voucher template is pre-constructed for each Skill category, and a standard template name is defined. The template for skill call type is named SkillCall_Template.ofd, the template for model inference type is named ModelInference_Template.ofd, and the template for data operation type is named DataOperation_Template.ofd.

[0008] Preferably, when a Skill is triggered, the system monitors and records the complete context of the Skill call in real time, including the call time, the caller's identity, the global call chain identifier, and the triggering method. At the same time, each input data item is categorized according to a five-level classification system, the original input content and preprocessing results are recorded, intermediate calculation results, decision paths, knowledge retrieval basis, and abnormal information during the execution process are captured, as well as the final output results, confidence and probability distribution, and interpretability evidence. Resource consumption and execution status are also recorded synchronously. All of the above information is persistently stored in the server background and an operation snapshot information package is generated, recording metadata information, operation behavior, and intermediate result snapshots in real time.

[0009] Preferably, in steps S2 and S3, the recorded operation snapshots, input / output content, and call context information are mapped to the corresponding type templates constructed in step S1. The structured data is converted into the OFD standard XML core data layer, the visualization content is filled into the information layer display area, and the original logs and attachment files are saved to the attachment layer. Differentiated encryption strategies are adopted for data fields with different sensitivity levels. Sensitive fields are encrypted at the field level using the SM4 national cryptographic algorithm. Then, the entire credential content is digitally signed with an SM2 digital certificate and anchored with a timestamp. At the same time, a content hash value is generated for each credential and a chain hash association is established with the previous credential, ultimately generating an OFD trusted electronic credential for a single operation. An OFD trusted electronic credential for a single Skill result is generated. The credential naming rules are as follows: the skill call credential is named Cred_SkillCall_{SkillID}{Timestamp}.ofd, the model inference credential is named Cred_ModelInference{ModelID}{SkillID}{Timestamp}.ofd, and the data operation credential is named Cred_DataOp_{Data Type}{SkillID}{Timestamp}.ofd.

[0010] Preferably, when the business scenario involves the serial invocation, conditional branching, or parallel execution of multiple Skills, the invocation order, data flow dependency mapping, triggering conditions, and roles and responsibility chains of multiple users and systems are recorded synchronously. The OFD electronic credentials corresponding to each skill are organized into a DLF dynamic format file according to the arrangement relationship. The DLF dynamic format file includes a credential set index table and an arrangement relationship diagram. Based on the independent encrypted signature of each credential, a credential set-wide security mechanism is applied to each credential, including SM2 overall signing of the entire DLF file package, establishing cross-credential encrypted associations, supporting aggregate verification and consistency checks, ultimately forming a complete DLF trusted electronic credential set. The naming rule for the trusted electronic credential set file is as follows: the skill chain credential set is named DLF_SkillChain_{chainID}{timestamp}.ofd, the credential set index table is named DLF_Index{chainID}.ofd, and the arrangement relationship diagram is named DLF_DAG_{chainID}.ofd.

[0011] Preferably, a multi-dimensional navigation file is generated based on the DLF voucher set, including a timeline view, a logical dependency view, and a role behavior chain view. The execution process of the skill chain can be dynamically displayed according to time sequence, skill call relationship, or logical dependency relationship. It is connected to a DLF electronic voucher parser and reader, providing reading modes including overall overview, voucher-by-voucher browsing, comparison and traceability, and auditing. It supports online real-time signature verification and offline batch signature verification, verifies validity, and generates navigation files. The naming rules for navigation files are as follows: the timeline view navigation file is named DLF_Navigation_TimeLine_{chain ID}.ofd, the logical dependency view navigation file is named DLF_Navigation_LogicDep_{chain ID}.ofd, and the role behavior chain view navigation file is named DLF_Navigation_RoleChain_{chain ID}.ofd. Finally, a DLF visual call overview map is generated, and the overview map file is named DLF_Overview_{chain ID}_{timestamp}.ofd.

[0012] Preferably, the DLF trusted credentials are bound to the vertical model output to achieve full-chain traceability. The generated DLF credential set is bound to the final output of the vertical model, and a unique DLF identifier is embedded in the output. The complete skill chain credential set can be queried through the unique identifier for bidirectional traceability. At the same time, the credential mechanism is integrated into the entire lifecycle of the vertical model, including training, deployment, inference, and feedback, forming a closed-loop trusted credential chain. The DLF trusted credentials are bound to the vertical model output, and a binding identifier is embedded in the vertical model output. The identifier is used to associate with the corresponding OFD credential file. The identifier naming rule is as follows: the credential set binding identifier is named DLF_Bind_{chain ID}{output ID}, and the single credential binding identifier is named DLF_Bind_Single{credential ID}.

[0013] Preferably, the DLF voucher set is output as a standardized ZIP compressed package, which contains a manifest file, a guide file, a voucher folder, an attachment folder, and a signature folder, forming a unified encapsulation format. It provides a standardized API interface for external systems to call, including single voucher generation, voucher set generation, voucher verification, voucher set signature verification, voucher set query, and guide file export.

[0014] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention constructs a trusted electronic credential mechanism for the entire process of Skill invocation. When a Skill is triggered, the system records the complete context of the Skill invocation in real time, including the invocation time, the identity of the invoker, input data, intermediate calculation results, decision path, final output results, and confidence level, and embeds this information in the OFD electronic credential. Each Skill invocation obtains a unique and tamper-proof trusted identity identifier, fundamentally solving the problems of unclear source and untraceable process of Skill work results.

[0015] (2) This invention extends the trusted credential mechanism to scenarios involving chained invocation and orchestration of multiple skills. When business involves the serial invocation, conditional branching, or parallel execution of multiple skills, the system synchronously records the invocation order, data flow dependency mapping, triggering conditions, and the role responsibility chain of multiple people and systems between skills, and organizes the OFD electronic credential corresponding to each skill into a DLF dynamic format file according to the orchestration relationship. This makes the complex multi-skill collaboration process no longer a difficult-to-trace "black box process," but a trusted skill chain that is verifiable and verifiable.

[0016] (3) This invention achieves end-to-end binding and bidirectional traceability between Skill achievements and vertical model outputs through DLF credential sets. A unique DLF identifier is embedded in the final output of the vertical model, supporting forward tracing from the original skill call to the final achievement, and reverse locating from the final achievement to the key skill credentials, input data, and dependency rules affecting that achievement. Simultaneously, the credential mechanism is integrated into the entire lifecycle of the vertical model's training, deployment, inference, and feedback, forming a closed-loop trusted credential chain, effectively ensuring the auditability and accountability of Skill-driven AI services throughout their entire lifecycle.

[0017] (4) Based on the independent encryption and signing of individual Skill OFD credentials, this invention further applies an overall signing, cross-credential encryption, and aggregate verification mechanism to the DLF credential set. Any addition, deletion, tampering, or order adjustment of individual Skill credentials in the credential set will destroy the overall signature and consistency verification of the credential set, providing a higher level of anti-tampering capability than distributed storage, and ensuring the integrity and trustworthiness of the Skill call chain.

[0018] (5) This invention is based on the national standard OFD format, supports national cryptographic algorithms (SM series) from the bottom layer, is compatible with the existing OFD ecosystem, and meets the requirements of localization and high-standard security. At the same time, it provides standardized guide documents and API interfaces, supports online real-time signature verification and offline batch signature verification, facilitates seamless integration with existing business systems, audit systems and third-party evidence storage platforms, and is easy to promote and deploy across the industry. Attached Figure Description

[0019] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, 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. Example

[0021] Dynamic layout files (DLF) are files that exist as a package containing a group of OFD files and their relationships and display relationships. This allows for the aggregation of layout files and complex dynamic operations.

[0022] like Figure 1 As shown, this invention combines DLF generation technology to provide a transparent and reliable method for realizing Skill work results based on DLF, including the following steps: Step S1: Construct a vertical model dataset classification system and Skill metadata templates. Predefine OFD electronic voucher templates according to different skill categories. Different skill categories refer to various specialized business skills or operational skills categorized for OFD electronic voucher processing scenarios. These skills typically serve specific voucher processing tasks in vertical fields (such as finance, taxation, government affairs, and healthcare). Each skill category corresponds to a type of voucher processing capability and requires a predefined matching OFD electronic voucher template. Therefore, based on data sources and business attributes, the datasets involved in the vertical model are divided into five major categories: industry professional knowledge data, specific business data, multi-source heterogeneous perception data, model inference process data, and Skill orchestration and dependency data. Each major category is further subdivided into subcategories and minor categories to form a complete five-level classification system. Here, "skill" refers to specific functional modules pre-packaged or dynamically invoked by AI models (such as large language models), enabling them to execute operational tools, access external knowledge, or complete complex tasks. In the skill grading of the vertical model, based on a five-level classification system, the data items involved in each skill in the DLF credential set are tagged and stored according to the hierarchical path of "multi-source heterogeneous sensing data, medium category, and minor category". Source tracing involves analyzing statements, calling the vertical model, performing intent analysis, and retrieving corresponding skill information. When this is achieved, multi-granularity retrieval and filtering by classification level is supported, including: filtering all skill credentials using this type of data by primary category (e.g., "multi-source heterogeneous sensing data"); locating skill calls from specific data sources by sub-categories (e.g., "sensing data → monitoring information → complaint and reporting data"); and quickly identifying key skill credentials under a specific data path affecting the final result of the vertical model when deviations or anomalies occur. This classification retrieval mechanism works in conjunction with the index table of the DLF credential set to form a two-dimensional source tracing capability of "classification tree + credential chain". Therefore, based on the hierarchical analysis of the vertical model, the corresponding skill call information can be accurately located.

[0023] Therefore, in constructing the skill metadata template, a structured metadata model is predefined for each skill in the vertical model. The metadata model includes basic skill information, interface specifications, dependency declarations, execution constraints, and version associations. A corresponding OFD electronic voucher template is also constructed for each skill category to distinguish the skill's output type. The output types include skill call type, model inference type, and data operation type. An OFD electronic voucher template is pre-constructed for each skill category, and a standard template name is defined. The template for skill call type is named SkillCall_Template.ofd, the template for model inference type is named ModelInference_Template.ofd, and the template for data operation type is named DataOperation_Template.ofd.

[0024] Meanwhile, when a user performs an operation, Skill uses a vertical model to match the corresponding skill based on the user's semantic needs and calls this model to trigger the skill's execution. At the same time, the system monitors and records the complete context of the skill call in real time, including the call time, the caller's identity, the global call chain identifier, and the triggering method. Simultaneously, each input data item is categorized according to a five-level classification system, recording the original input content and preprocessing results, capturing intermediate calculation results, decision paths, knowledge retrieval basis, and abnormal information during the execution process, as well as the final output results, confidence and probability distribution, and interpretability evidence. Resource consumption and execution status are also recorded synchronously. All of the above information is persistently stored on the server backend and an operation snapshot information package is generated, recording metadata information, operation behavior, and intermediate result snapshots in real time.

[0025] Step S2: Record the operation behavior, calling context, and intermediate result snapshots in real time during the execution of the Skill; Step S3: Map the recorded information to the corresponding OFD template, and after encryption, digital signature and chain hash association, generate a trusted electronic certificate of OFD for a single Skill achievement; The specific process is as follows: Recorded operation snapshots, input / output content, and call context information are mapped to the corresponding type template constructed in step S1. Structured data is converted into the OFD standard XML core data layer. Visualized content is filled into the information layer display area, and the original logs and attachments are saved to the attachment layer. Differentiated encryption strategies are adopted for data fields with different sensitivity levels. Sensitive fields are encrypted at the field level using the SM4 national cryptographic algorithm. Then, a digital certificate is used to perform an SM2 digital signature on the entire credential content and anchor a timestamp. Simultaneously, a content hash value is generated for each credential and a chain hash association is established with the previous credential. Finally, an OFD trusted electronic credential for a single operation is generated, i.e., an OFD trusted electronic credential for a single Skill achievement is generated. The credential naming rules are as follows: Skill call credential is named Cred_SkillCall_{SkillID}{Timestamp}.ofd, Model inference credential is named Cred_ModelInference{ModelID}{SkillID}{Timestamp}.ofd, and Data operation credential is named Cred_DataOp_{Data Type}{SkillID}{Timestamp}.ofd.

[0026] Step S4: When a business scenario involves the serial invocation, conditional branching, or parallel execution of multiple skills, synchronously record the invocation order, data flow dependency mapping, triggering conditions, and the roles and responsibility chains of multiple users and systems. For multi-step skill chains, skill credentials are organized into a DLF trusted credential set according to their arrangement relationship. A comprehensive credential set signature and consistency verification are applied. This primarily involves organizing the OFD electronic credentials corresponding to each skill into a dynamic DLF format file, which includes a credential set index table and an arrangement relationship diagram. Based on the independent encrypted signature of each credential, a comprehensive credential set security mechanism is applied to each credential. This includes SM2 comprehensive signing of the entire DLF file package, establishing cross-credential encrypted associations, supporting aggregate verification and consistency verification, ultimately forming a complete DLF trusted electronic credential set. The naming convention for the trusted electronic credential set file is as follows: the skill chain credential set is named DLF_SkillChain_{chainID}{timestamp}.ofd, the credential set index table is named DLF_Index{chainID}.ofd, and the arrangement relationship diagram is named DLF_DAG_{chainID}.ofd.

[0027] Therefore, when business operations involve the sequential invocation, conditional branching, or parallel execution of multiple skills, the system synchronously records the invocation order, data flow dependency mapping, triggering conditions, and the role responsibility chain among multiple people and systems. It also organizes the OFD electronic voucher corresponding to each skill into a DLF dynamic format file according to the arrangement relationship. This transforms complex multi-skill collaboration processes from untraceable "black box processes" into verifiable and trustworthy skill chains.

[0028] Step S5: Generate a multi-dimensional navigation file containing a timeline, logical dependencies, and role behavior chains, supporting transparent display and auditing; Step S6: Generate a DLF visualization overview map to provide a panoramic view of the skill chain execution; Based on the DLF voucher set, multi-dimensional navigation files are generated, including a timeline view, a logical dependency view, and a role behavior chain view. The execution process of the skill chain can be dynamically displayed according to time sequence, skill call relationship, or logical dependency relationship. It connects to the DLF electronic voucher parser and reader, providing reading modes including overall overview, voucher-by-voucher browsing, comparison and traceability, and auditing. It supports online real-time signature verification and offline batch signature verification, verifies validity, and generates navigation files. The naming rules for navigation files are as follows: the timeline view navigation is named DLF_Navigation_TimeLine_{chain ID}.ofd, the logical dependency view navigation is named DLF_Navigation_LogicDep_{chain ID}.ofd, and the role behavior chain view navigation is named DLF_Navigation_RoleChain_{chain ID}.ofd. Finally, a DLF visual call overview diagram is generated, and the overview diagram file is named DLF_Overview_{chain ID}_{timestamp}.ofd.

[0029] Step S7: Bind the DLF trusted credentials with the vertical model output to achieve bidirectional end-to-end traceability. By binding the generated DLF credential set with the final output of the vertical model, a unique DLF identifier is embedded in the output. The complete skill chain credential set is queried through the unique identifier for bidirectional traceability. Simultaneously, the credential mechanism is integrated into the entire lifecycle of the vertical model's training, deployment, inference, and feedback, forming a closed-loop trusted credential chain. Binding the DLF trusted credentials with the vertical model output involves embedding a binding identifier in the vertical model's output. This identifier is used to associate with the corresponding OFD credential file. The identifier naming rules are as follows: the credential set binding identifier is named DLF_Bind_{chain ID}{output ID}, and the single credential binding identifier is named DLF_Bind_Single{credential ID}. This ensures that the result of each call is matched one-to-one with the called skill.

[0030] By embedding a unique DLF identifier into the final output of the vertical model, it is possible to trace forward from the original skill call to the final result, and backward from the final result to locate the key skill credentials, input data, and dependency rules that affect the result. Simultaneously, the credential mechanism is integrated into the entire lifecycle of the vertical model—training, deployment, inference, and feedback—forming a closed-loop trusted credential chain, effectively ensuring the auditability and accountability of Skill-driven AI services throughout their entire lifecycle.

[0031] Step S8: After the DLF credential set is generated, a standardized DLF credential set output encapsulation format and API interface are constructed. The DLF credential set output is a standardized ZIP compressed package containing a manifest file, a guide file, a credential folder, an attachment folder, and a signature folder, forming a unified encapsulation format. A standardized API interface is provided for external systems to call, including single credential generation, credential set generation, credential verification, credential set signature verification, credential set query, and guide file export. Therefore, this invention constructs a trusted electronic credential mechanism for the entire process of Skill invocation. When a Skill is triggered, the system records the complete context of the Skill invocation in real time, including the invocation time, caller identity, input data, intermediate calculation results, decision path, final output result, and confidence level, and embeds this information in the OFD electronic credential. Each Skill invocation obtains a unique, tamper-proof, and trusted identity identifier, fundamentally solving the problems of unclear source and untraceable process of Skill work results.

[0032] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments are to be regarded in all respects as exemplary and not restrictive, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A method for achieving transparent and reliable skill-based work results based on DLF, characterized in that, Includes the following steps: Step S1: Construct a vertical model dataset classification system and Skill metadata templates, and predefine OFD electronic voucher templates according to different skill categories; Step S2: Record the operation behavior, calling context, and intermediate result snapshots in real time during the execution of the Skill; Step S3: Map the recorded information to the corresponding OFD template, and after encryption, digital signature and chain hash association, generate a trusted electronic certificate of OFD for a single Skill achievement; Step S4: For a multi-step Skill chain, organize the skill credentials into a DLF trusted credential set according to the arrangement relationship, and apply overall signature and consistency verification to the credential set; Step S5: Generate a multi-dimensional navigation file containing a timeline, logical dependencies, and role behavior chains, supporting transparent display and auditing; Step S6: Generate a DLF visualization overview map to provide a panoramic view of the skill chain execution; Step S7: Bind the DLF trusted credentials to the vertical model results to achieve two-way full-link traceability; Step S8: Construct a standardized DLF credential set output encapsulation format and API interface.

2. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: In step S1, a classification system for the vertical model dataset and a skill metadata template are constructed. Based on the data source and business attributes, the datasets involved in the vertical model are divided into five categories: industry professional knowledge data, specific business data, multi-source heterogeneous perception data, model inference process data, and skill arrangement and dependency data. Each major category is further subdivided into subcategories and minor categories to form a complete five-level classification system.

3. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: Based on a five-level classification system, the data items involved in each skill in the DLF credential set are tagged and stored according to the hierarchical path of "multi-source heterogeneous perception data, medium category, and minor category". During tracing, it supports multi-granularity retrieval and filtering by classification level, including: filtering all skill credentials that use this type of data by primary category; locating skill calls from specific data sources by sub-category; and quickly locking key skill credentials under a certain data path that affects the final result of the vertical model when there are deviations or anomalies. Through the collaborative work of the classification retrieval mechanism and the index table of the DLF credential set, a two-dimensional tracing capability of "classification tree + credential chain" is formed.

4. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: In step S1, a skill metadata template is constructed. For each skill in the vertical model, a structured metadata model for each skill is predefined, including basic skill information, interface specifications, dependency declarations, execution constraints, and version associations. A corresponding OFD electronic voucher template is also constructed for each skill category to distinguish the skill's output type. The output types include skill call type, model inference type, and data operation type. An OFD electronic voucher template is pre-constructed for each skill category, and a standard template name is defined. The template for skill call type is named SkillCall_Template.ofd, the template for model inference type is named ModelInference_Template.ofd, and the template for data operation type is named DataOperation_Template.ofd.

5. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: When a Skill is triggered, the system monitors and records the complete context of the skill call in real time, including the call time, the caller's identity, the global call chain identifier, and the triggering method. At the same time, each input data item is categorized according to a five-level classification system, the original input content and preprocessing results are recorded, intermediate calculation results, decision paths, knowledge retrieval basis, and abnormal information during the execution process are captured, as well as the final output results, confidence and probability distribution, and interpretability evidence. Resource consumption and execution status are also recorded synchronously. All of the above information is persistently stored in the server background and an operation snapshot information package is generated, recording metadata information, operation behavior, and intermediate result snapshots in real time.

6. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: In steps S2 and S3, the recorded operation snapshots, input / output content, and call context information are mapped to the corresponding type templates constructed in step S1. The structured data is converted into the OFD standard XML core data layer, and the visualization content is filled into the information layer display area. The original logs and attachment files are saved to the attachment layer. Differentiated encryption strategies are adopted for data fields with different sensitivity levels. Sensitive fields are encrypted at the field level using the SM4 national cryptographic algorithm. Then, the entire credential content is digitally signed with an SM2 digital certificate and anchored with a timestamp. At the same time, a content hash value is generated for each credential and a chain hash association is established with the previous credential. Finally, an OFD trusted electronic credential for a single operation, i.e., an OFD trusted electronic credential for a single Skill achievement, is generated. The credential naming rules are as follows: the Skill Call credential is named Cred_SkillCall_{SkillID}{Timestamp}.ofd, the Model Inference credential is named Cred_ModelInference{ModelID}{SkillID}{Timestamp}.ofd, and the Data Operation credential is named Cred_DataOp_{Data Type}{SkillID}{Timestamp}.ofd.

7. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: When business scenarios involve serial invocation, conditional branching, or parallel execution of multiple Skills, the system synchronously records the invocation order, data flow dependency mapping, triggering conditions, and roles and responsibility chains among multiple users and systems. Each Skill's corresponding OFD electronic certificate is organized into a DLF dynamic format file according to its arrangement relationship. This DLF dynamic format file includes a certificate set index table and an arrangement relationship diagram. Based on the independent encrypted signature of each certificate, a certificate set-wide security mechanism is applied to each certificate, including SM2 overall signing of the entire DLF file package, establishing cross-certificate encrypted associations, supporting aggregate verification and consistency checks, ultimately forming a complete DLF trusted electronic certificate set. The naming rules for the trusted electronic certificate set file are as follows: the Skill Chain Certificate Set is named DLF_SkillChain_{chainID}{timestamp}.ofd, the Certificate Set Index Table is named DLF_Index{chainID}.ofd, and the Arrangement Relationship Diagram is named DLF_DAG_{chainID}.ofd.

8. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: Based on the DLF voucher set, multi-dimensional navigation files are generated, including a timeline view, a logical dependency view, and a role behavior chain view. The execution process of the skill chain can be dynamically displayed according to time sequence, skill call relationship, or logical dependency relationship. It connects to the DLF electronic voucher parser and reader, providing reading modes including overall overview, voucher-by-voucher browsing, comparison and traceability, and auditing. It supports online real-time signature verification and offline batch signature verification, verifies validity, and generates navigation files. The naming rules for navigation files are as follows: the timeline view navigation is named DLF_Navigation_TimeLine_{chain ID}.ofd, the logical dependency view navigation is named DLF_Navigation_LogicDep_{chain ID}.ofd, and the role behavior chain view navigation is named DLF_Navigation_RoleChain_{chain ID}.ofd. Finally, a DLF visual call overview diagram is generated, and the overview diagram file is named DLF_Overview_{chain ID}_{timestamp}.ofd.

9. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: By binding DLF trusted credentials with the results of vertical models, end-to-end traceability is achieved. The generated DLF credential set is bound to the final output of the vertical model, and a unique DLF identifier is embedded in the output. The complete skill chain credential set can be queried through the unique identifier for bidirectional traceability. At the same time, the credential mechanism is integrated into the entire lifecycle of the vertical model, including training, deployment, inference, and feedback, forming a closed-loop trusted credential chain. The binding identifier is embedded in the vertical model output and used to associate the corresponding OFD credential file. The naming rules for the identifier are as follows: the credential set binding identifier is named DLF_Bind_{chain ID}{output ID}, and the single credential binding identifier is named DLF_Bind_Single{credential ID}.

10. The method for achieving transparent and reliable Skill work results based on DLF according to claim 1, characterized in that: DLF voucher sets are output as standardized ZIP archives, containing manifest files, navigation files, voucher folders, attachment folders, and signature folders, forming a unified encapsulation format. They provide standardized API interfaces for external systems to call, including single voucher generation, voucher set generation, voucher verification, voucher set signature verification, voucher set query, and navigation file export.

Images