Method and device for automatically filling recruitment application, electronic equipment and readable storage medium

Abstract

The application provides a recruitment application automatic filling method and device, electronic equipment and readable storage medium, which comprises the following steps: in the case of user first dialogue or user intent ambiguity, constructing a job-seeking portrait of the user according to the user language; determining a post result according to the job-seeking portrait, and displaying the post result through the "data block direct-out" mechanism wrapped by Sentinel; in the case of selecting a target post or pasting a post description, calling a resume_get tool to obtain a complete structured resume; in the case of structured resume optimization ending, continuously calling an interview search capability and a test question search capability according to the same target company and the same target post, obtaining related preparation materials, and forming a continuous job-seeking path of "post search-> resume optimization-> interview preparation"; in the case of determining a delivery target, calling an online application automatic process to automatically fill in the recruitment application. The application takes the job-seeking portrait as a starting point, forms an automatic job-seeking process, and can realize the recruitment application automatic filling.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence, and more specifically, to a method, apparatus, electronic device, and readable storage medium for automatically filling out job applications. Background Technology

[0002] Current intelligent assistance systems for job seekers mainly fall into six categories: traditional recruitment website keyword search, browser-extended online application tools, web interaction agents driven by large language models, streaming large model dialogue clients, online resume editing systems, and general AI desktop dialogue clients. However, existing technologies suffer from the following drawbacks: fragmented job search processes, difficulty in reusing user profiles, and insufficient automated collaboration in online applications. Existing solutions can no longer meet the needs of efficient job seeking and urgently require innovative breakthroughs. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method, device, electronic device and readable storage medium for automatically filling in recruitment applications, which can solve the problems of fragmented modules, repeated information entry and easy data loss due to partial patches on resumes in existing job search tools.

[0004] Firstly, this application provides a method for automatically filling out recruitment applications, including: constructing a job-seeking profile of the user based on the user's language in the case of initial user interaction or ambiguous user intent; determining job results based on the job-seeking profile and displaying the job results through the "data block direct output" mechanism wrapped in Sentinel; when a target job is selected or a job description is pasted, calling the resume_get tool to obtain a complete structured resume; after the structured resume optimization is completed, continuing to call the interview experience search capability and written test question search capability based on the same target company and the same target job to obtain relevant preparation materials, forming a continuous job-seeking path of "job search → resume optimization → interview preparation"; and when the target application is determined, calling the online application automation process to automatically fill out the recruitment application.

[0005] In the above implementation process, starting with job seeker profiles, the process of job selection, resume optimization, searching for interview / written test questions, and filling out job applications is linked into a continuous job search path. This forms an automated job search process of job search → resume optimization → interview preparation → filling out job applications, which can solve problems such as fragmented modules, duplicate information entry, and easy loss of data due to partial patches on resumes in existing job search tools.

[0006] In one embodiment, the step of automatically filling out the recruitment application by invoking the online application automation process when the target of the application is determined includes: when the target of the application is determined, triggering the extended content script in BrowserView through the main process to perform page structure annotation and constructing a mapping relationship between element identifiers and element position paths; uniformly encapsulating the mapping relationship and the structured resume through the main process and sending them to the large language model; and executing multiple filling strategies in sequence according to a preset priority strategy based on the fields to be filled returned by the large language model until the recruitment application is successfully filled out.

[0007] In the above implementation process, when filling out a recruitment application, multiple filling strategies are executed sequentially according to a preset priority strategy. Through an automatic fallback mechanism, the robustness and success rate of automated recruitment application filling can be greatly improved, enabling true full-scenario adaptability and thus increasing form filling coverage and success rate.

[0008] In one embodiment, the method further includes: initiating a request through a target bridging server when the AI ​​Gateway subprocess executes a view control tool call; wherein the target bridging server is a bridging server whose environment variables are pre-injected to point to a local HTTP bridging server; calling a first preset method to control BrowserView; and broadcasting a custom IPC event indicating a change in the workspace state to the rendering process through the IPCsyncRenderer channel, thereby driving the rendering process's state management library to complete reactive state synchronization and UI re-rendering.

[0009] In the above implementation, tool calls from the Gateway subprocess are forwarded to the Electron main process, and UI re-rendering is driven synchronously through IPC and cross-process Pinia. This architecture allows the AI ​​Gateway subprocess to reliably drive the BrowserView multi-tab UI while running in an independent process space, without needing to directly hold UI component references, thus simplifying the system architecture.

[0010] In one embodiment, the method further includes: performing intent recognition based on the user language, the job seeker profile, the session context, and the predefined capability directory; after the BrowserView completes the page DOM loading, counting the total number of valid form elements in the currently visible page by injecting a script; injecting a fully simulated Chrome extension runtime environment into the BrowserView by preloading a script; and triggering the execution of page structure annotation by the extension content script in the BrowserView through the main process.

[0011] In the above implementation process, job search scenario capabilities are abstracted into two layers: atomic capabilities and multi-step workflows. A capability catalog records capability identifiers, types, categories, descriptions, and risk attributes. Combined with user job search profiles and session context, natural language requests are deterministically mapped to capabilities such as job postings, resumes, interview experiences, online applications, browsers, and task planning. This routing mechanism eliminates the instability of large models freely reasoning to select atomic capabilities and forces capabilities with destructive flags to follow an access control approval process. Furthermore, by simulating extension runtime APIs such as `chrome.runtime.connect / sendMessage`, `chrome.storage.local`, and `browser.cookies.getAll` in the BrowserView preload script and converting CSS files using the `chrome-extension: / / ` protocol to Blob URLs, recruitment assistance extensions that originally only ran in the Chrome browser can now run unchanged in embedded browsers within desktop applications. This allows for dynamic expansion of multi-recruitment platform coverage along with the extension ecosystem.

[0012] In one embodiment, the "direct output of data blocks" mechanism wrapped by Sentinel to display the job results includes: upon receiving a delta data fragment pushed by the LLM server, forwarding the delta data fragment to the target buffer; scanning the target buffer and marking the content in the target buffer with sentinels; upon detecting a complete matching tag and a complete first target block in the target buffer, completely removing the character range corresponding to the first target block from the output stream; and for the remaining content in the target buffer after removal, gradually appending the detected normal delta content to the regular rendering area of ​​the corresponding message bubble through a streaming Markdown renderer.

[0013] In the above implementation process, the combination of FIFO session queue, real-time sentinel filtering based on safety boundary algorithm (proposing thought tags), and bypass direct rendering of data blocks wrapped in Sentinel can solve three core reliability problems in the streaming output scenario of large language models: truncation of structured content, leakage of internal control tags, and mismatch of concurrent messages, thereby improving the reliability of large language models.

[0014] In one embodiment, after scanning the target buffer and marking the contents of the target buffer with a sentinel, the method further includes: if a complete pairing tag is detected and a complete second target block is detected in the target buffer, extracting a portion of the PAYLOAD content; and writing the entire PAYLOAD into the rendering area of ​​the corresponding message bubble using a second preset method provided by the dialogue storage module.

[0015] In the above implementation, by writing the entire PAYLOAD into the rendering area of ​​the corresponding message bubble, the conventional streaming rendering path based on token incremental appending can be completely bypassed. This can be used to solve the reliability problems of structured data caused by the influence of the context window, tokenizer vocabulary, or output truncation strategy of large language models, such as long URL encoding errors, long table row truncation, and missing escape characters.

[0016] In one embodiment, before proceeding to use interview experience search and written test question search capabilities to obtain relevant preparation materials and form a continuous job search path of "job search → resume optimization → interview preparation" after the structured resume optimization is completed, the method further includes: comparing the responsibilities, job requirements, and keywords in the job details with each module of the structured resume to generate multi-dimensional diagnostic results; determining the rewriting direction based on the multi-dimensional diagnostic results and incrementally rewriting the target modules according to the STAR method; and, upon completion of the incremental rewriting according to the STAR method, calling the resume_update tool to submit the overall saved model of the structured resume.

[0017] In the above implementation process, the online recruitment application auto-fill process adopts a secure rewrite model of overall reading, partial modification, and overall saving. This model first reads the complete AtsResumeDTO JSON, makes minimal modifications to the target fields or modules in memory, and then saves the complete JSON back to the server. This avoids the problem of data loss in other modules due to submitting only partial fields, thus improving data security.

[0018] In one embodiment, before constructing a job profile of the user based on the user's language in the case of the user's first conversation or when the user's intent is ambiguous, the method further includes: constructing and sending a "fake request" through a gateway when the finite state machine enters a warm-up state; listening to the streaming response event of the "fake request" through the system; determining that the warm-up is complete when a chat event is received and the chat state ends; determining the current state of the gateway based on the status code returned by the "fake request"; and processing the user request based on the current state of the gateway.

[0019] In the above implementation, a combination of a three-process isolation architecture (i.e., independent Node 22 child processes + WebSocket JSON-RPC V3 + Ed25519 device signature authentication), LLM provider's proactive TLS link warm-up (one-time session-warmup-{ts} fake request), and stateful finite state machines and lifecycle management (including exponential backoff reconnection and coordination of user requests and gateway states) can reduce the user-perceived AI call time while establishing an observable and recoverable operating mechanism for the gateway process.

[0020] In one embodiment, determining the current state of the gateway based on the status code returned by the "fake request" and processing the user request based on the current state includes: if the "fake request" returns a success HTTP status code, transitioning the finite state machine to the ready state and processing the user request; or, if the "fake request" returns a failure status code or times out, transitioning to the error state; calculating the reconnection waiting time according to the exponential backoff formula, and queuing for reconnection.

[0021] Secondly, this application also provides an automatic job application filling device, comprising: a construction module, used to construct a job profile of the user based on the user's language in the case of the user's first dialogue or when the user's intention is ambiguous; a display module, used to determine the job results based on the job profile and display the job results through the "data block direct output" mechanism wrapped by Sentinel; a first calling module, used to call the resume_get tool to obtain a complete structured resume when a target job is selected or a job description is pasted; an acquisition module, used to continue to call the interview experience search capability and written test question search capability to obtain relevant preparation materials based on the same target company and the same target job after the structured resume optimization is completed, forming a continuous job search path of "job search → resume optimization → interview preparation"; and a second calling module, used to call the online application automation process to automatically fill in the job application when the target application is determined.

[0022] Thirdly, embodiments of this application also provide an electronic device, including: a processor and a memory, wherein the memory stores machine-readable instructions executable by the processor, and when the electronic device is running, the machine-readable instructions are executed by the processor to perform the steps of the method described in the first aspect above, or any possible implementation of the first aspect.

[0023] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the steps of the recruitment application automatic filling method described in the first aspect or any possible implementation of the first aspect.

[0024] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, specific embodiments are described below in conjunction with the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0026] Figure 1 A block diagram illustrating an electronic device provided in an embodiment of this application; Figure 2 A flowchart illustrating the automatic recruitment application filling method provided in this application embodiment; Figure 3 A schematic diagram of the functional modules of the recruitment application auto-filling device provided in this application embodiment. Detailed Implementation

[0027] The technical solutions in the embodiments of this application will now be described with reference to the accompanying drawings.

[0028] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0029] Currently, intelligent assistance systems for job seekers (especially recent college graduates) can be mainly divided into six categories based on their implementation methods: keyword search systems for traditional recruitment websites, browser-extended online application assistance tools and desktop RPA, web interaction agents driven by large language models, streaming large model dialogue clients, online resume editing and template generation systems, and general AI desktop dialogue clients. Most of these solutions are composed of independent modules and share the following common problems: First, the job search process is fragmented, and user profiles cannot be reused in job search, resume editing, interview preparation, and online application submission, making it difficult for the system to form a continuous job search path. Second, the mapping from natural language intent to job search tools is unstable, lacking ability risk labeling and permission control. Third, the automatic filling compatibility of modern component-based front-end forms (such as Ant Design, Lark ATSX, etc.) is insufficient, and traditional scripts have difficulty synchronizing the internal state of the framework. Fourth, the automation of online applications lacks collaboration with AI dialogue and desktop browsers, resulting in a disconnect between the AI ​​decision-making layer and the user-visible UI control layer. Fifth, the reliability of structured data in streaming output is insufficient, and truncation or escaping errors are prone to occur in long tables and long URLs. Sixth, the AI ​​gateway has a significant cold start delay (10-15 seconds), making long-running process lifecycle management difficult. Seventh, multi-step job search tasks lack streaming feedback and observable states, making it impossible for users to confirm or take over at key nodes. Eighth, partial resume patching leads to data loss in multi-turn dialogue scenarios.

[0030] In view of this, this application proposes a method for automatically filling out job applications. Starting with a job seeker profile, it connects job selection, resume optimization, searching for interview / written test questions, and filling out job applications into a continuous job search path. This forms an automated job search process of job search → resume optimization → interview preparation → filling out job applications, which can solve problems such as fragmented modules, duplicate information entry, and easy data loss due to partial patches on resumes in existing job search tools.

[0031] To facilitate understanding of this embodiment, the electronic device that performs the automatic recruitment application filling method disclosed in this application embodiment will first be described in detail.

[0032] like Figure 1 The diagram shown is a block illustration of an electronic device. The electronic device 100 may include a memory 111 and a processor 113. Those skilled in the art will understand that... Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the electronic device 100. For example, the electronic device 100 may also include components that are more... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown.

[0033] The aforementioned memory 111 and processor 113 are electrically connected to each other directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected to each other via one or more communication buses or signal lines. The aforementioned processor 113 is used to execute executable modules stored in the memory.

[0034] The memory 111 can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory 111 stores programs, and the processor 113 executes these programs upon receiving execution instructions. The methods executed by the electronic device 100 as defined in any embodiment of this application can be applied to the processor 113, or implemented by the processor 113.

[0035] The aforementioned processor 113 may be an integrated circuit chip with signal processing capabilities. The processor 113 may be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it may also be a digital signal processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor.

[0036] The electronic device 100 in this embodiment can be used to execute various steps in the various methods provided in the embodiments of this application. The implementation process of the automatic recruitment application filling method is described in detail below through several embodiments.

[0037] Please see Figure 2 This is a flowchart of the automatic recruitment application filling method provided in the embodiments of this application. The following will describe... Figure 2 The specific process shown will be explained in detail.

[0038] Step 201: In the case of the user's first conversation or when the user's intention is ambiguous, construct the user's job-seeking profile based on the user's language.

[0039] Specifically, the profile building process can be triggered by personal skills, collecting fields such as school, major, degree, graduation year, target industry, target position, job type (e.g., internship, full-time, campus recruitment, social recruitment), target city, target company, internship experience, skill set, experience highlights, etc. in rounds to form a structured job profile.

[0040] In one embodiment, the job profile can be persisted in JSON format (e.g., ~ / .nowclaw / profile.json).

[0041] The job seeker profile here can be used as a default parameter in processes such as job recommendations, resume optimization, interview preparation, and online application submission.

[0042] Step 202: Determine job results based on job seeker profiles and display the job results using the "data block output" mechanism of Sentinel.

[0043] In one embodiment, determining job results based on job seeker profiles can be achieved as follows: Job skills are invoked based on the job seeker profile or user input. If the user provides keywords, company, or city, the system uses a job list tool to search based on those criteria. If the user selects a specific job, the job details tool reads the complete target job description; if there are no specific keywords but a profile has been created, matching jobs are obtained according to the target job, city, and company preferences in the profile, following a recommended process. The job results are directly displayed to the user in a clickable Markdown table format using data blocks wrapped in Sentinel.

[0044] The direct data output mechanism here is a comprehensive approach combining FIFO session routing, streaming Sentinel filtering, and data block bypass rendering. It ensures a strict sequential correspondence between concurrent messages and streaming responses through an explicit FIFO queue, removes internal control flags that might be echoed by the model through real-time sentinel detection based on a safety boundary algorithm, and allows skill scripts to bypass the LLM's per-token output limitation and directly push complete structured content to UI rendering through the "data block direct output" mechanism wrapped in Sentinel.

[0045] Step 203: If the target position is selected or the job description is pasted, call the resume_get tool to obtain the complete structured resume.

[0046] It should be understood that selecting a target position or pasting a job description indicates that the user has identified the position they wish to apply for, and can then further refine their resume based on that position.

[0047] A structured resume can include modules such as: basic information, work experience, project experience, professional skills, and self-evaluation.

[0048] Step 204: After the structured resume optimization is completed, continue to use the interview experience search and written test question search capabilities to obtain relevant preparation materials based on the same target company and the same target position, forming a continuous job search path of "job search → resume optimization → interview preparation".

[0049] The preparation materials here can include historical interview questions, historical written test questions, mock interview questions, mock written test questions, etc., and these materials can be selected according to the actual situation.

[0050] Step 205: Once the target audience is determined, the online application automation process is invoked to automatically fill in the recruitment application.

[0051] Among them, the online application automation process is an AI Agent-driven cross-platform online application auto-fill process. The online application automation process is an online application auto-fill method based on Chrome extension API runtime virtualization, with multi-skill intent routing driven by capability catalog as the entry point, bidirectional scheduling between AI Agent and desktop UI as the skeleton, and adaptive triggering of form complexity as the execution strategy.

[0052] Optionally, if a user changes their target job, city, or company at any stage, and the system updates the profile or session context, the result of the application can be written back to the session history.

[0053] In the above implementation process, starting with job seeker profiles, the process of job selection, resume optimization, searching for interview / written test questions, and filling out job applications is linked into a continuous job search path. This forms an automated job search process of job search → resume optimization → interview preparation → filling out job applications, which can solve problems such as fragmented modules, duplicate information entry, and easy loss of data due to partial patches on resumes in existing job search tools.

[0054] In one possible implementation, step 205 includes: when the target audience is determined, triggering the extended content script within BrowserView through the main process to execute page structure annotation and constructing a mapping relationship between element identifiers and element location paths; uniformly encapsulating the mapping relationship and structured resume through the main process and sending it to the large language model; and executing multiple filling strategies in sequence according to a preset priority strategy based on the fields to be filled returned by the large language model, until the recruitment application is successfully filled out.

[0055] Page structure annotation can be achieved by: traversing all valid form elements on the current page and generating a unique HTTP attribute identifier (e.g., data-nc-id) for each. Simultaneously, a mapping dictionary from form element identifiers to element paths (e.g., XPath) is constructed, along with a concise text description of the page form structure. The extracted results are then sent back to the main process cache via an IPC channel for later use.

[0056] Optionally, the structured resume may include modules such as basic information, a list of work experience, and a list of project experience.

[0057] In one embodiment, the mapping relationship and the structured resume are encapsulated and submitted to the server-side LLM inference service. After obtaining the encapsulated mapping relationship and the structured resume, the server calls the large language model to match the semantics of the page form fields with the semantics of the resume data, and returns a set of form field filling instructions.

[0058] Each element here can contain attributes such as field labels, values ​​to be filled, a unique key that identifies the target form element, the element path, and optional execution scripts (e.g., DynamicScript declarative or ActionScript step-by-step).

[0059] The prioritization strategy described above can be a priority chain. This priority chain can be: (1) UI framework component identification strategy; (2) Site-specific adapter strategy (e.g., Beisen, Mocha, ByteDance, Zhilian, Zhizhi, etc.); (3) General dropdown option filling strategy; (4) DynamicScript declarative script execution; (5) ActionScript step-by-step script degradation; (6) The original input is assigned as a fallback.

[0060] The priority of the priority chain mentioned above decreases sequentially from smallest to largest according to the sequence number.

[0061] In one embodiment, the filling results of each step are recorded and reported, and the element currently being filled in is highlighted on the page through a visual breathing light animation.

[0062] Understandably, when the BrowserView panel width is insufficient to meet the minimum desktop width (e.g., 1024px) required for the recruitment website design, the system calls the `ensureDesktopZoom` method. This method internally executes the BrowserView's `setZoomFactor` function, automatically scaling the webpage content. By scaling the webpage, the "available width within the browser" read by JavaScript is never less than a certain minimum value (e.g., 1024px), even if the outer window is dragged very narrow. After the page loads, the `fitPageToView` method dynamically detects the total width of the page elements and the right edge of the child elements, accurately calculates the actual content width, and performs a secondary adaptive scaling.

[0063] In the above implementation process, when filling out a recruitment application, multiple filling strategies are executed sequentially according to a preset priority strategy. Through an automatic fallback mechanism, the robustness and success rate of automated recruitment application filling can be greatly improved, enabling true full-scenario adaptability and thus increasing form filling coverage and success rate.

[0064] In one possible implementation, the method further includes: initiating a request through the target bridging server when the AI ​​Gateway subprocess executes a view control tool call; invoking a first preset method to control the BrowserView; broadcasting a custom IPC event indicating a change in the workspace state to the rendering process through the IPC syncRenderer channel, thereby driving the rendering process's state management library to complete reactive state synchronization and UI re-rendering.

[0065] The target bridging server is a local HTTP bridging server whose environment variable is pre-injected. This environment variable can be NOWCLAW_WORKSPACE_BRIDGE_URL. This bridging server runs in the Electron main process.

[0066] Optionally, the view control tool may include tools for controlling the browser view, controlling the plugin view, etc.

[0067] The first default method here can be a method such as WorkspaceBrowserHost.openPage, and this first default method can be selected according to the actual situation.

[0068] In the above implementation, tool calls from the Gateway subprocess are forwarded to the Electron main process, and UI re-rendering is driven synchronously through IPC and cross-process Pinia. This architecture allows the AI ​​Gateway subprocess to reliably drive the BrowserView multi-tab UI while running in an independent process space, without needing to directly hold UI component references, thus simplifying the system architecture.

[0069] In one possible implementation, the method further includes: performing intent recognition based on user language, job seeker profile, conversation context, and a predefined capability directory; after BrowserView completes page DOM loading, counting the total number of valid form elements in the currently visible page by injecting a script; injecting a fully simulated Chrome extension runtime environment into BrowserView by preloading a script; and triggering the execution of page structure annotations by the main process through the extension content scripts in BrowserView.

[0070] The capability catalog here is uniformly established by the system. The executable capabilities of the system are divided into two categories: atomic capabilities and multi-step workflows. Atomic capabilities represent directly callable atomic capabilities, such as job_list, job_detail, resume_get, resume_update, resume_create, resume_upload, open_browser, browser, show_plugin, interview_search, written_test_search, etc. Multi-step workflows represent workflows containing multiple steps and domain rules, such as user profile creation, job recommendation, resume editing, resume creation, resume upload, and online application assistant. Each capability entry includes metadata such as capability identifier, capability type, Chinese name, user-understandable description, category, whether it is destructive, and whether permission approval is required.

[0071] Understandably, upon receiving a user's language request, intent recognition is performed by combining the job seeker profile, conversation context, and a predefined capability catalog. If the request is a single-step structured task (e.g., "search for ByteDance front-end internships"), the corresponding atomic capability (e.g., job list) is directly invoked; if the request requires multi-step collaboration (e.g., "help me apply for this job"), the corresponding multi-step workflow (e.g., online application) is invoked, and multiple atomic capabilities are orchestrated internally according to a fixed process; for capabilities with destructive flags (e.g., resume deletion), an authorization approval process can be implemented before scheduling.

[0072] In one embodiment, the method for injecting a script to count the total number of valid form elements on the currently visible page can be as follows: "Valid form element" is defined as: input elements whose type attribute does not belong to the set {hidden input field, clickable regular button, submit button, file upload control, radio button, checkbox}, plus all select and textarea elements. The statistical results are triggered in three tiers: no trigger when the number is less than 3; only the FAB floating button is displayed when the number is 3 to 4; and the complete form panel is automatically expanded when the number is greater than or equal to 5.

[0073] Extending the runtime environment can be achieved in the following ways: (1) Forge chrome.runtime.connect and chrome.runtime.sendMessage / onMessage to intercept messages from inter-extension communication libraries such as webext-bridge in bridge relay port mode and forward them locally; (2) Map the get / set / remove interfaces of chrome.storage.local to the memory storage window.__NC_ELECTRON_STORAGE mounted on the global object; (3) Map the browser.cookies.getAll interface to read the global cookie array window.__NC_ELECTRON_COOKIES that the main process has pre-injected via Electron session.defaultSession.cookies.get; (4) To address the limitation that CSS files using the chrome-extension: / / protocol cannot be loaded within BrowserView, the CSS content is converted into a Blob URL and then injected into the page using the URL.createObjectURL(new Blob([cssText], { type: "text / css"})) scheme.

[0074] Using the above method, recruitment assistance extensions that originally could only run in the Chrome browser can now run without any modifications within embedded browsers of desktop applications, and their coverage of multiple recruitment platforms can be dynamically expanded along with the extension ecosystem.

[0075] In the above implementation process, job search scenario capabilities are abstracted into two layers: atomic capabilities and multi-step workflows. A capability catalog records capability identifiers, types, categories, descriptions, and risk attributes. Combined with user job search profiles and session context, natural language requests are deterministically mapped to capabilities such as job postings, resumes, interview experiences, online applications, browsers, and task planning. This routing mechanism eliminates the instability of large models freely reasoning to select atomic capabilities and forces capabilities with destructive flags to follow an access control approval process. Furthermore, by simulating extension runtime APIs such as `chrome.runtime.connect / sendMessage`, `chrome.storage.local`, and `browser.cookies.getAll` in the BrowserView preload script and converting CSS files using the `chrome-extension: / / ` protocol to Blob URLs, recruitment assistance extensions that originally only ran in the Chrome browser can now run unchanged in embedded browsers within desktop applications. This allows for dynamic expansion of multi-recruitment platform coverage along with the extension ecosystem.

[0076] In one possible implementation, the job results are displayed through a "direct data block output" mechanism wrapped in Sentinel, including: upon receiving a delta data fragment pushed by the LLM server, forwarding the delta data fragment to the target buffer; scanning the target buffer and marking the content in the target buffer with sentinels; if a complete matching tag is detected and a complete first target block is detected in the target buffer, completely removing the character range corresponding to the first target block from the output stream; for the remaining content in the target buffer after removal, the ordinary delta content that is detected as a hit is gradually appended to the regular rendering area of ​​the corresponding message bubble through a streaming Markdown renderer.

[0077] Optionally, the method further includes: maintaining a data structure sessionQueryIds:Map on the client side.<string,string[]> The key is the session ID, and the value is a queue of query IDs within the session that have not yet received a complete response. When a user sends a message within the same session, the system generates a unique query ID for each message and adds it to the end of the queue. When the streaming response for the corresponding session arrives, the system retrieves the query ID from the head of the queue according to the FIFO principle and establishes a correspondence with the preceding message bubble; after the response is complete, the query ID is dequeued. This mechanism ensures that when multiple messages are sent rapidly and continuously within the same session, the streaming responses for each message will not be mismatched.

[0078] In one embodiment, before forwarding the delta data fragment to the target buffer, the method further includes: establishing a streaming receive buffer (i.e., the target buffer). The method can then be further implemented by: each time a delta data fragment is received from the LLM server, appending the delta data fragment to the end of the buffer corresponding to the query ID, instead of immediately forwarding it to the UI rendering layer.

[0079] The sentinel markers here can include data block markers and thought process markers.

[0080] Specifically, the sentinel marker can be implemented as follows: if a complete matching tag is detected, either the thought process content is filtered or the data block is rendered directly using Sentinel wrapping. If an open tag is detected but the closing tag has not yet arrived, a safe boundary algorithm is executed: the maximum character length at the end of the buffer that could be a prefix of a certain sentinel marker is calculated, the content at the end of that length is temporarily held, and the remaining safe portion is rendered normally in a streaming manner.

[0081] Normal streaming rendering here can be achieved by gradually appending to the regular rendering area of ​​the corresponding message bubble through a streaming Markdown renderer.

[0082] Specifically, if a complete matching tag is detected and a complete first target block is detected in the target buffer, the character range corresponding to the first target block is completely removed from the output stream.

[0083] For the remaining content after stripping in the target buffer, the normal delta content that is detected as a hit is gradually appended to the regular rendering area of ​​the corresponding message bubble through the streaming Markdown renderer.

[0084] The first target block here can be<think_xxx> ...< / think_xxx> piece.

[0085] In one embodiment, if the streaming response ends normally or is interrupted abnormally, it can be further checked whether there are any unclosed sentinel markers remaining in the target buffer: if there are any, the remaining content is appended to the output in plain text form, and the abnormal event is recorded and reported through the SLS log system.

[0086] In the above implementation process, the combination of FIFO session queue, real-time sentinel filtering based on safety boundary algorithm (proposing thought tags), and bypass direct rendering of data blocks wrapped in Sentinel can solve three core reliability problems in the streaming output scenario of large language models: truncation of structured content, leakage of internal control tags, and mismatch of concurrent messages, thereby improving the reliability of large language models.

[0087] In one possible implementation, after scanning the target buffer and marking the contents of the target buffer with a sentinel, the method further includes: if a complete pairing tag is detected and a complete second target block is detected in the target buffer, extracting a portion of the PAYLOAD content; and writing the entire PAYLOAD into the rendering area of ​​the corresponding message bubble using a second preset method provided by the dialog storage module.

[0088] The second target block can be <<<NOWCLAW_RENDER> >>PAYLOAD<<< / NOWCLAW_RENDER> >>block. The second default method can be the addDataBlock method.

[0089] In the above implementation, by writing the entire PAYLOAD into the rendering area of ​​the corresponding message bubble, the conventional streaming rendering path based on token incremental appending can be completely bypassed. This can be used to solve the reliability problems of structured data caused by the influence of the context window, tokenizer vocabulary, or output truncation strategy of large language models, such as long URL encoding errors, long table row truncation, and missing escape characters.

[0090] In one possible implementation, before step 204, the method further includes: comparing the responsibilities, job requirements, and keywords in the job details with each module of the structured resume to generate multi-dimensional diagnostic results; determining the rewriting direction based on the multi-dimensional diagnostic results and performing incremental rewriting of the target module according to the STAR method; and, after the incremental rewriting according to the STAR method is completed, calling the resume_update tool to submit the overall saved model of the structured resume.

[0091] Optionally, the multi-dimensional diagnostic results can be 5-dimensional diagnostic results, including: completeness, quantification level, expression quality, keyword coverage, and format standardization.

[0092] The STAR method for incremental rewriting of the target module can be achieved as follows: retain a complete copy of AtsResumeDTO in memory, and make minimal changes to the target fields (such as strengthening project scope, responsibilities, and quantitative results indicators); after the rewriting is completed, call the resume_update tool to save the model as a complete structured resume and submit it, thus fundamentally avoiding damage to the unmodified modules.

[0093] In the above implementation process, the online recruitment application auto-fill process adopts a secure rewrite model of overall reading, partial modification, and overall saving. This model first reads the complete AtsResumeDTO JSON, makes minimal modifications to the target fields or modules in memory, and then saves the complete JSON back to the server. This avoids the problem of data loss in other modules due to submitting only partial fields, thus improving data security.

[0094] In one possible implementation, prior to step 201, the method further includes: constructing and sending a "fake request" through the gateway when the finite state machine enters the warm-up state; listening for streaming response events of the "fake request" through the system; determining that the warm-up is complete when a chat event is received and the chat state ends; determining the current state of the gateway based on the status code returned by the "fake request"; and processing the user request based on the current state of the gateway.

[0095] The "fake request" can be constructed in the following ways: (1) using a one-time session key session-warmup-{ts}, completely isolated from the user's actual session; (2) the request content is the minimum valid hint (e.g., a single character "ping"); (3) the real purpose is to trigger the LLM provider's TLS handshake, connection pool initialization, API route resolution, and CDN edge node cache warm-up. After the warm-up is completed, the original 10-15 second initial call delay is reduced to the normal network latency level (usually less than 2 seconds).

[0096] Optionally, the status code may include a success status code, a failure status code, etc., and the status code can be selected according to the actual situation.

[0097] In one embodiment, before constructing and sending a “fake request” through the gateway, the method further includes: initialization of the three-process isolation architecture and device authentication, and initialization of the gateway lifecycle finite state machine.

[0098] The three-process isolation architecture and device authentication initialization can be implemented as follows: When the Electron main process starts, the OpenClaw AIGateway child process is started as an independent system Node 22 executable file through the process startup module. Before starting, all proxy class environment variables (such as HTTP_PROXY) are cleared. The main process and the Gateway child process communicate via the WebSocket JSON-RPC V3 protocol. During the protocol handshake phase, a challenge-response process is executed: the main process loads the pre-generated Ed25519 device key pair from ~ / .openclaw / identity / device.json (permission 0600), signs the handshake payload containing fields such as device identifier, client identifier, role, authorization scope, one-time random number, and platform, and establishes a trusted connection only after the Gateway verifies the signature.

[0099] The gateway lifecycle finite state machine initialization can be implemented in the following way: After the Gateway child process is created, the lifecycle module in the main process initializes a ten-state finite state machine: Idle and not started, process starting, warm-up request execution, ready to serve, processing user requests, error pending recovery, WebSocket reconnection, closing, normal closure, abnormal crash requiring main process decision, etc.

[0100] In the above implementation, a combination of a three-process isolation architecture (i.e., independent Node 22 child processes + WebSocket JSON-RPC V3 + Ed25519 device signature authentication), LLM provider's proactive TLS link warm-up (one-time session-warmup-{ts} fake request), and stateful finite state machines and lifecycle management (including exponential backoff reconnection and coordination of user requests and gateway states) can reduce the user-perceived AI call time while establishing an observable and recoverable operating mechanism for the gateway process.

[0101] In one possible implementation, the current state of the gateway is determined based on the status code returned by the "fake request," and the user request is processed according to the current state, including: if the "fake request" returns a success HTTP status code, the finite state machine is transitioned to the ready state and the user request is processed; or, if the "fake request" returns a failure status code or times out, the process is transitioned to the error state; the reconnection waiting time is calculated according to the exponential backoff formula, and the process is queued for reconnection.

[0102] The backoff formula can be: delay = min(3000 × 2^retryCount, 15000). That is, the first retry waits for 3 seconds, the second for 6 seconds, the third for 12 seconds, and a maximum of 15 seconds; after more than 3 retries, it is marked as a crash; process-level restart is based on delay = min(3000 × 2^restartCount, 30000) (maximum 3 times).

[0103] In one embodiment, if the "fake request" returns a success HTTP status code, the user request is forwarded directly. If it is labeled as "warming up" or "reconnecting," it is added to the waiting queue; if it is labeled as "crashed" or "stopped," the service is indicated as unavailable and a restart is triggered.

[0104] In this embodiment, the normal shutdown process of the gateway can be: preparation → stop → stopped; in the busy state, it waits for the request to be processed before entering the stopped state.

[0105] Based on the same application concept, this application also provides an automatic recruitment application filling device corresponding to the automatic recruitment application filling method. Since the principle of the device in this application is similar to that of the aforementioned automatic recruitment application filling method, the implementation of the device in this application can refer to the description in the above-mentioned method embodiments, and the repeated parts will not be repeated.

[0106] Please see Figure 3 This is a functional module diagram of the recruitment application auto-filling device provided in this embodiment. Each module in the recruitment application auto-filling device in this embodiment is used to execute the steps in the above method embodiments. The recruitment application auto-filling device includes a construction module, a display module, a first calling module, an acquisition module, and a second calling module; wherein, The building module is used to construct a job profile of a user based on the user's language in the case of the user's first conversation or when the user's intention is ambiguous.

[0107] The display module is used to determine job results based on the job seeker profile and to display the job results through the "data block output" mechanism wrapped in Sentinel.

[0108] The first module is used to call the resume_get tool to obtain a complete structured resume when a target position is selected or a job description is pasted.

[0109] The acquisition module is used to continue using the interview experience search and written test question search capabilities to obtain relevant preparation materials after the structured resume optimization is completed, based on the same target company and the same target position, forming a continuous job search path of "job search → resume optimization → interview preparation".

[0110] The second module is used to automatically fill out the recruitment application by calling the online application automation process when the target of the application is determined.

[0111] In one possible implementation, the second calling module is specifically used for: upon determining the target of the application, triggering the extended content script within BrowserView through the main process to execute page structure annotation and constructing a mapping relationship between element identifiers and element location paths; uniformly encapsulating the mapping relationship and the structured resume through the main process and sending them to the large language model; and executing multiple filling strategies sequentially according to a preset priority strategy based on the fields to be filled returned by the large language model, until the job application is successfully filled out.

[0112] In one possible implementation, the automatic job application filling device further includes a driver module, used to initiate a request through a target bridging server when the AIGateway subprocess executes a view control tool call; wherein the target bridging server is a bridging server whose environment variables are pre-injected to point to a local HTTP bridging server; calling a first preset method to control BrowserView; broadcasting a custom IPC event indicating a change in the workspace state to the rendering process through the IPC syncRenderer channel, driving the rendering process's state management library to complete reactive state synchronization and UI re-rendering.

[0113] In one possible implementation, the automatic job application filling device further includes an intent recognition module, used to recognize intent based on the user's language, the job seeker profile, the conversation context, and a predefined capability directory; after the BrowserView completes page DOM loading, it injects a script to count the total number of valid form elements in the currently visible page; it injects a fully simulated Chrome extension runtime environment into the BrowserView by preloading a script; and it triggers the execution of page structure annotations by the main process through the extension content scripts in the BrowserView.

[0114] In one possible implementation, the presentation module is specifically configured to: upon receiving a delta data fragment pushed by the LLM server, forward the delta data fragment to the target buffer; scan the target buffer and mark the content in the target buffer with sentinels; if a complete matching tag is detected and a complete first target block is detected in the target buffer, completely remove the character range corresponding to the first target block from the output stream; for the remaining content in the target buffer after removal, gradually append the detected normal delta content to the regular rendering area of ​​the corresponding message bubble through the streaming Markdown renderer.

[0115] In one possible implementation, the job application autofill device further includes a writing module, used to extract part of the PAYLOAD content when a complete matching tag is detected and a complete second target block is detected in the target buffer; and to write the entire PAYLOAD into the rendering area of ​​the corresponding message bubble using a second preset method provided by the dialogue storage module.

[0116] In one possible implementation, the automatic job application filling device further includes a rewriting module, which compares the responsibilities, job requirements, and keywords in the job details with each module of the structured resume to generate multi-dimensional diagnostic results; determines the rewriting direction based on the multi-dimensional diagnostic results, and performs incremental rewriting of the target module according to the STAR method; and, upon completion of the incremental rewriting according to the STAR method, calls the resume_update tool to submit the overall saved model of the structured resume.

[0117] In one possible implementation, the automatic job application filling device further includes a status determination module, which is used to construct and send a "fake request" through the gateway when the finite state machine enters the preheating state; listen to the streaming response event of the "fake request" through the system; determine that the preheating is complete when a chat event is received and the chat state ends; determine the current state of the gateway based on the status code returned by the "fake request"; and process the user request based on the current state of the gateway.

[0118] In one possible implementation, the state determination module is specifically used to: if the "fake request" returns an HTTP status code of success, transition the finite state machine to the ready state and process the user request; or, if the "fake request" returns a non-success status code or times out, transition to the error state; calculate the reconnection waiting time according to the exponential backoff formula, and queue the request for reconnection.

[0119] Furthermore, this application embodiment also provides a computer-readable storage medium storing a computer program, which, when run by a processor, executes the steps of the automatic recruitment application filling method described in the above method embodiment.

[0120] The computer program product of the automatic recruitment application filling method provided in this application embodiment includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the steps of the automatic recruitment application filling method described in the above method embodiment. For details, please refer to the above method embodiment, which will not be repeated here.

[0121] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0122] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0123] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element. The above description is merely a preferred embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0124] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for automatically filling out job applications, characterized in that, include: In cases of initial user conversation or ambiguous user intent, construct a job-seeking profile of the user based on the user's language. The job postings are determined based on the job seeker profile, and the job postings are displayed using the "data block output" mechanism wrapped in Sentinel. If a target position is selected or a job description is pasted, the resume_get tool will be called to obtain a complete structured resume; After the structured resume optimization is completed, the interview experience search and written test question search capabilities are used to obtain relevant preparation materials based on the same target company and the same target position, forming a continuous job search path of "job search → resume optimization → interview preparation". Once the target audience is identified, the online application automation process is invoked to automatically fill out the job application.

2. The method according to claim 1, characterized in that, The process of automatically filling out recruitment applications by invoking the online application automation workflow, once the target audience is determined, includes: Once the delivery target is determined, the main process triggers the extended content script within BrowserView to execute page structure annotations and build a mapping relationship between element identifiers and element location paths; The mapping relationship and the structured resume are uniformly encapsulated by the main process and sent to the large language model; Based on the fields to be filled returned by the large language model, multiple filling strategies are executed sequentially according to a preset priority strategy until the recruitment application is successfully filled out.

3. The method according to claim 2, characterized in that, The method further includes: When the AI ​​Gateway subprocess executes a view control tool call, it initiates a request through the target bridging server; wherein, the target bridging server is a local HTTP bridging server whose environment variables are pre-injected. Call the first preset method to control the BrowserView; Custom IPC events that indicate changes in the workspace state are broadcast to the rendering process via the IPC syncRenderer channel, driving the rendering process's state management library to complete reactive state synchronization and UI re-rendering.

4. The method according to claim 2, characterized in that, The method further includes: Intent recognition is performed based on the user's language, the job seeker profile, the conversation context, and the predefined capability catalog. After BrowserView finishes loading the page DOM, a script is injected to count the total number of valid form elements in the currently visible page. A fully simulated Chrome extension runtime environment is injected into BrowserView via a preloaded script; The main process triggers the execution of extended content scripts within BrowserView to annotate page structure.

5. The method according to claim 1, characterized in that, The "data block direct output" mechanism wrapped in Sentinel is used to display the job results, including: Upon receiving a delta data fragment pushed by the LLM server, the delta data fragment is forwarded to the target buffer. Scan the target buffer and mark the contents of the target buffer with a sentinel. If a complete pairing tag is detected, and a complete first target block is detected in the target buffer, the character range corresponding to the first target block is completely removed from the output stream; For the remaining content after stripping in the target buffer, the normal delta content that is detected as a hit is gradually appended to the regular rendering area of ​​the corresponding message bubble through the streaming Markdown renderer.

6. The method according to claim 5, characterized in that, After scanning the target buffer and marking the contents of the target buffer with sentinels, the method further includes: If a complete paired tag is detected, and a complete second target block is detected in the target buffer, extract the PAYLOAD portion of the content; The entire PAYLOAD is written into the rendering area of ​​the corresponding message bubble using the second preset method provided by the dialogue storage module.

7. The method according to claim 1, characterized in that, Before proceeding with the structured resume optimization process, and further utilizing interview experience and written test question search capabilities to obtain relevant preparation materials based on the same target company and the same target position, thus forming a continuous job search path of "job search → resume optimization → interview preparation," the method also includes: The job description, responsibilities, requirements, and keywords are compared with the modules of the structured resume to generate multi-dimensional diagnostic results. Based on the multi-dimensional diagnostic results, the rewriting direction is determined, and the target module is incrementally rewritten using the STAR method. Once the incremental rewriting of the STAR rules is complete, the resume_update tool is invoked to submit the overall saved model of the structured resume.

8. The method according to claim 1, characterized in that, Before constructing a job-seeking profile of a user based on their language during the user's initial conversation or when the user's intent is ambiguous, the method further includes: When the finite state machine enters the warm-up state, a "fake request" is constructed and sent through the gateway; The system monitors the streaming response events of the "fake requests"; If a chat event is received and the chat state ends, the preheating is considered complete. The current state of the gateway is determined based on the status code returned by the "fake request", and the user request is processed according to the current state of the gateway.

9. The method according to claim 8, characterized in that, The step of determining the current gateway status based on the status code returned by the "fake request" and processing the user request based on the current status includes: If the "fake request" returns a success HTTP status code, the finite state machine is transitioned to the ready state, and the user request is processed; or, If the "fake request" returns a non-success status code or times out, it will be transferred to an error status. The reconnection waiting time is calculated using the exponential backoff formula, and the system is then queued for reconnection.

10. A device for automatically filling out job applications, characterized in that, include: The module is designed to build a job profile of a user based on their language during the initial conversation or when the user's intent is ambiguous. The display module is used to determine job results based on the job seeker profile and display the job results through the "data block direct output" mechanism wrapped in Sentinel; The first calling module is used to call the resume_get tool to obtain a complete structured resume when a target position is selected or a job description is pasted. The acquisition module is used to continue to use the interview experience search and written test question search capabilities to obtain relevant preparation materials after the structured resume optimization is completed, based on the same target company and the same target position, forming a continuous job search path of "job search → resume optimization → interview preparation". The second module is used to automatically fill out the recruitment application by calling the online application automation process when the target of the application is determined.

11. An electronic device, characterized in that, include: The processor and memory, wherein the memory stores machine-readable instructions executable by the processor, wherein when the electronic device is running, the machine-readable instructions are executed by the processor to perform the steps of the method as described in any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, performs the steps of the method as described in any one of claims 1 to 9.

Images