A private skills asset collaborative network system and method based on AI Agent and blockchain technology

Abstract

The application discloses a private skill asset collaboration network system and method based on AI Agent and blockchain technology. The system comprises: an identity authentication layer, which realizes the trusted proof of the professional qualifications of nodes without exposing the identity through zero-knowledge proof; a demand routing layer, which realizes the directional broadcast of demand ciphertext to the supply node holding the corresponding qualifications based on attribute encryption; an asset delivery layer, which generates one-time output assets based on the private skill module of the supply node local AI Agent, and anchors the asset ownership to the blockchain through NFT; a verification layer, which selects multiple verification nodes through the on-chain verifiable random function, checks the format compliance and logical self-consistency of the output assets, and aggregates the multi-node signature proof on the chain; and a settlement layer, which automatically triggers the token settlement and NFT ownership transfer based on the verification result and the demand side acceptance. The application realizes the safe monetization of private skills without exposing the underlying knowledge, solves the core problems in the prior art, such as the realization of knowledge only after being disclosed, the difficulty in qualification verification in anonymous networks, and the strong subjectivity of the verification mechanism, and has the significant advantages of decentralization, privacy protection and anti-manipulation.

Description

Technical Field

[0001] This invention belongs to the field of blockchain and artificial intelligence interdisciplinary technology, specifically relating to a decentralized collaborative network infrastructure that allows personal AI agents to securely conduct one-time asset transactions by exporting the holder's private professional skills. Background Technology

[0002] The rapid proliferation of personal AI agent systems based on large language models has enabled individual users to encode their professional experience, judgment logic, and private data into reusable skill modules at low cost. These private skills are characterized by high personalization (the skill is embedded with the holder's accumulated experience and private data, and cannot be directly purchased and copied on the market), usability gains (the skill continuously optimizes with increased use and is non-consumable), and isolation and closure (currently, there is a lack of standardized collaboration mechanisms between individual agents, and skills cannot be securely transferred between individuals).

[0003] Existing digital knowledge collaboration and trading platforms (such as traditional crowdsourcing platforms, AI tool marketplaces, knowledge payment platforms, and open-source communities) all face structural contradictions: Traditional crowdsourcing platforms (such as Upwork) trade human service time, but knowledge cannot be converted into assets and cannot be scaled; AI tool marketplaces (such as GPT Store) trade publicly trained general model capabilities, but private expert experience cannot be incorporated, resulting in severe homogenization; knowledge payment platforms (such as Dedao and Coursera) trade content formats, and knowledge providers lose their exclusive value once the course content is made public; open-source communities (such as GitHub) expose the knowledge of contributors completely, lacking a sustainable closed-loop economic incentive mechanism.

[0004] Knowledge providers must make their knowledge public to monetize it, but doing so means losing their exclusive rights. The shortcomings of existing technologies include: (1) Insufficient intellectual property protection: Existing digital knowledge transactions lack technical means to prevent knowledge from being copied and disseminated, and rely solely on legal contracts for constraints; (2) Lack of trust mechanism: Professional qualifications in anonymous networks cannot be verified at low cost, making it difficult to establish a professional knowledge service market; (3) Weak privacy protection: The data involves highly sensitive professional knowledge in the fields of medicine, law, and finance, which cannot be circulated on existing platforms due to data privacy issues; (4) The verification mechanism is highly subjective: the existing rating system (Upwork rating, Taobao credit, etc.) relies on the subjective evaluation of buyers, which has systemic bias and score fraud issues; (5) Centralized power concentration: Existing platforms, acting as intermediaries, accumulate large amounts of user data and knowledge assets, thus structurally exploiting knowledge providers. Summary of the Invention

[0005] This invention aims to solve the following technical problems: (1) How to achieve secure monetization of the skill output without exposing the underlying knowledge of the skill (training data, judgment logic, private database); (2) How to verify the professional qualifications of nodes in an anonymous distributed network in a low-cost and reliable manner; (3) How to minimize the exposure and securely transmit highly sensitive professional data without the intervention of a centralized platform; (4) How to design a professional asset quality verification mechanism free from subjective bias and conflict of interest; (5) How to build a decentralized collaborative infrastructure to ensure that no single entity can censor or confiscate the skills assets of participants.

[0006] The technical solution of the present invention: This invention proposes a distributed private skill asset collaboration network system based on AI Agent, whose core technical solution includes the following five-layer blockchain integration architecture: The authentication layer is used to cryptographically prove to other nodes in the blockchain network that the node holds a valid professional qualification of a specific category without exposing the node holder's identifiable identity information; The demand routing layer is used to broadcast the ciphertext of the demand from demand-side nodes to supply-side nodes in the network that only hold the corresponding qualification certificates, based on attribute encryption or predicate encryption. The asset delivery layer is used by the local AI Agent of the supply-side node to generate one-time output assets for the demand based on its private skill module, and to anchor the decentralized storage identifier of the asset to the blockchain with non-fungible tokens (NFTs) representing its ownership. The verification layer is used to select multiple verification nodes through an on-chain verifiable random source. Each verification node checks the format compliance and logical consistency of the output asset and writes the verification results into the blockchain in the form of a multi-node signature aggregation proof. The settlement layer is used to automatically trigger on-chain settlement of tokens and transfer of ownership of the NFT based on the aggregated proof and the on-chain acceptance confirmation of the demand side.

[0007] Furthermore, the identity verification layer is specifically configured as follows: verifying the real-world professional qualification information of the node through an off-chain multi-signature oracle cluster, generating verifiable qualification certificates based on zk-SNARK or zk-STARK zero-knowledge proof protocols, and writing the certificates into the blockchain identity registration smart contract in accordance with the ERC-725 standard. The certificate status is maintained through a Merkle state tree.

[0008] Furthermore, the demand routing layer is specifically configured as follows: receiving the demand digest after local de-identification processing by the demand party node, encrypting the demand digest using the Ciphertext Policy Attribute Base Encryption Algorithm (CP-ABE), generating a ciphertext demand packet that can only be decrypted by nodes holding the corresponding qualification category attribute key, writing the SHA-256 hash value of the demand digest onto the chain for notarization, and broadcasting the ciphertext demand packet to the distributed P2P network via the Kademlia DHT protocol.

[0009] Furthermore, the asset delivery layer is specifically configured as follows: the AI ​​Agent of the supply-side node generates a one-time structured output asset for a specific need based on a local private skill module. The content of the private skill module does not enter the network. Only the final output asset is uploaded to the InterPlanetary File System (IPFS) or Arweave permanent storage network after end-to-end encryption to obtain a Content Addressing Identifier (CID). The NFT is then minted using the ERC-721 standard. The metadata of the NFT includes the CID, the demand chain transaction hash (Tx Hash), the supply-side node public key hash, and the generation timestamp block height.

[0010] Furthermore, the verification layer is specifically configured as follows: N verification nodes are randomly selected from the nodes holding corresponding qualification ZKP certificates by an on-chain verifiable random function (VRF). The AI ​​Agent of the verification nodes performs logical consistency and format compliance checks on the output assets based on the on-chain verification rule set, and generates a multi-dimensional scoring vector. The verification results of each node are aggregated into a Merkle proof by an M-of-N multi-signature contract and written into the block. The verification rule set is deployed on on-chain contracts for different professional qualification categories, and the verification content includes: the integrity of the argument-based logical chain, compliance with industry standard formats, consistency of cited data and declaration hashes, and filtering of prohibited statements.

[0011] Furthermore, the settlement layer is specifically configured as follows: the demand side's funds are held in custody with ERC-20 functional tokens. After receiving a verification report that meets the M-of-N multi-signature threshold and the demand side confirms acceptance on the chain, the contract automatically releases the main token reward to the supply side node, releases a fixed proportion of protocol token fees to each verification node, and injects protocol fees into the DAO treasury. The key parameters of the protocol are updated by the DAO governance contract through voting by token holders.

[0012] Furthermore, the settlement layer also includes a reputation staking incentive mechanism based on token economics: before participating in verification, a certain number of ERC-20 reputation tokens must be staked; if the assets passed by the verification node are found to have significant defects in subsequent dispute arbitration, the corresponding staked tokens will be destroyed proportionally; the accumulated reputation points on the node chain positively affect its weight coefficient in the subsequent VRF selection.

[0013] Furthermore, the settlement layer also includes an on-chain dispute arbitration sub-process: when the requesting party sends a dispute request transaction within the protection period block height, the VRF will select M expert nodes in the same field to participate in the arbitration, and the arbitration nodes must add staked tokens; the arbitration result is determined by the majority rule principle, and after the result is written into the block, the contract will redistribute the escrow tokens according to the arbitration output, and the losing party will bear the arbitration node token fee.

[0014] This invention also provides a method for collaborating on private skill assets based on AI Agent and blockchain technology, characterized by the following steps: Step S1: The supply-side node publishes cryptographic credentials to the blockchain network, proving that it holds a specific category of professional qualifications, through a zero-knowledge proof protocol; Step S2: The demand-side node performs attribute-based encryption on the demand content that has been locally de-identified, generates a ciphertext demand packet, and broadcasts it to the peer-to-peer network. At the same time, the hash value of the demand digest is stored on the blockchain for evidence. Step S3: After receiving and decrypting the encrypted demand package, the supply-side node calls the private skill module of its local AI Agent to generate a one-time output asset for the demand, stores the asset in a decentralized storage network and obtains a content identifier, and at the same time mints a non-fungible token (NFT) representing the ownership of the asset. Step S4: Select multiple verification nodes through an on-chain verifiable random source. Each verification node performs formal verification on the output asset, generates verification results, and aggregates them into an auditable on-chain proof by multi-node signatures. Step S5: The requesting node reviews the verification results and decides whether to send on-chain acceptance confirmation; Step S6: After receiving the acceptance confirmation and the verification result meets the preset conditions, the smart contract automatically completes the token settlement and the transfer of ownership of the NFT.

[0015] Furthermore, in step S3, no component of the private skill module is transmitted over the network or stored on the blockchain. The components include the large language model fine-tuning weights, the private vector database, the judgment logic prompts, and the private dataset. Only the one-time structured output assets and their content addressing identifiers generated based on the private skill module are transmitted over the network.

[0016] The beneficial effects of this invention are: (1) For the first time, the blockchain NFT standard is combined with AI Agent skill assets to realize a new on-chain digital intellectual property protection paradigm of "only circulating one-time output asset NFTs and not circulating the skills themselves", which eliminates the risk of private knowledge being copied and spread from the cryptographic level; (2) Through zk-SNARK / zk-STARK zero-knowledge proof and off-chain oracle multi-signature anchor, a cryptographically reliable proof of professional qualifications is achieved without exposing node identity information, and the certificate is guaranteed to be tamper-proof through distributed ledger; (3) A three-layer data minimization exposure protocol was constructed, consisting of an on-chain ZKP credential layer, an off-chain CP-ABE encrypted routing layer, and a local privacy computing layer, enabling private knowledge in highly sensitive professional fields to participate in on-chain circulation securely for the first time; (4) It provides a professional verification mechanism with no conflict of interest, which is based on VRF randomness, Merkle multi-signature notarization and token staking penalties, and solves the fundamental defects of traditional rating systems that are highly subjective and can be manipulated. (5) A decentralized collaborative infrastructure based on DAO governance has been implemented. Protocol parameters are collectively governed by token holders. No single node can audit or confiscate the on-chain assets of participants. The system itself does not accumulate any user data or skill assets. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the overall system architecture of the present invention, showing the identity verification layer, demand routing layer, asset delivery layer, verification layer, settlement layer and their data flow; Figure 2 This is a flowchart of the on-chain ZKP qualification certificate registration process in the identity verification layer, including steps such as oracle multi-signature verification, zk-SNARK proof generation, ERC-725 contract writing, and Merkle state tree update. Figure 3 This is a flowchart of the demand encrypted broadcasting and on-chain hash notarization based on CP-ABE in the demand routing layer, including the Kademlia DHT routing path; Figure 4 A complete delivery flowchart for asset generation, IPFS content-addressed storage, and ERC-721 NFT casting and on-chain in the asset delivery layer; Figure 5 This is a flowchart of the multi-node verification process in the verification layer, which is based on VRF random selection and includes the multi-signature Merkle proof aggregation chain writing step. Figure 6 This is the settlement state machine diagram of the EVM smart contract in the settlement layer, including token allocation, dispute arbitration sub-processes, and DAO governance parameter update paths. Detailed Implementation

[0018] Terminology Definition Private skill modules: These refer to a set of reusable functional modules embedded in a node holder's personal AI Agent (based on a large language model or other artificial intelligence model) through configuration, parameter fine-tuning, hint engineering, or customized code. These modules can be formalized as a function f (requirement description) -> structured output asset. Their internal implementation details (including but not limited to weight fine-tuning, private vector database indexes, core rule sets, and judgment logic templates) remain on the holder's local device or trusted execution environment and do not enter any form of network transmission or on-chain storage.

[0019] Local anonymization: This refers to the process by which the requesting node, before publishing a request, uses a named entity recognition model to remove direct identifiers such as personal names, contact information, and precise geographical locations from the request text; or converts free text requests into structured keyword-attribute pairs; or generalizes sensitive values ​​(e.g., replacing specific ages with age ranges). The processed request summary should not contain information that can be reversibly traced back to the identity of the original requesting entity.

[0020] Verification Rule Set: A predefined set of on-chain smart contract rules for different professional qualification categories (such as medical, legal, and financial). Taking medical professional consultation as an example, the "Argument-Based Logic Chain Integrity Check" requires that each medical conclusion in the output report must cite at least one excerpt from an authoritative guideline or literature pre-registered on-chain, and compare the hash value of that excerpt with the hash value stored on-chain. "Format Compliance Check" requires the report to conform to the template structure of a specific specialty. "Prohibited Statement Filtering" checks whether the report contains language that violates regulations, such as "absolute cure" or "guaranteed effectiveness."

[0021] Example 1: refer to Figure 1 The system in this embodiment includes five functional layers.

[0022] 1. Authentication Layer (Layer 0) – On-chain ZKP credential registration Supply-side nodes first need to register their professional qualifications through an identity verification layer. See details for the process. Figure 2 : Step 201: The qualification holder submits their real-world professional license information (medical practitioner's license, lawyer's license, CFA certificate, etc.) to the system-approved off-chain oracle node cluster. The oracle node uses a multi-signature mechanism (M-of-N Multi-signature) to prevent single point of tampering. Steps 202-203: The Oracle node calls the corresponding authoritative database (the National Health Commission's license database, the Ministry of Justice's lawyer database, etc.) to verify authenticity and broadcasts the qualification hash commitment to the participating nodes. After confirmation by the Byzantine Fault Tolerance Consensus (BFT Consensus), it is written into the permissioned chain. Step 204: The qualification holder generates a zero-knowledge proof based on the on-chain commitment using a zk-SNARK circuit to prove the proposition "This node holds a valid professional license of a specific category". The proof process does not reveal the license number, the holder's name, or any identifiable information. Steps 205-207: After the zero-knowledge credential is verified by the on-chain verification node, the credential is written into the blockchain identity registration smart contract using the node's public key as an index and the ERC-725 standard (on-chain identity). The credential's validity period is bound to the block height and the license's validity period, and an immutable audit log is stored on the chain. Steps 208-211: Certificate Revocation Mechanism: When an Oracle node detects that the corresponding license has expired, it sends a revocation transaction to the on-chain registration contract, triggering an on-chain event (Event Log) broadcast. After the Merkle Patricia Trie is updated, the relevant node automatically relinquishes its network access permission for the corresponding qualification category.

[0023] This mechanism replaces traditional centralized identity verification with on-chain verifiable cryptographic credentials, achieving a decentralized root of trust that "proves qualifications without exposing identity".

[0024] 2. Demand Routing Layer (Layer 1) – ABE Ciphertext Broadcast and On-Chain Hash Notification A targeted demand broadcast protocol based on attribute encryption (CP-ABE) combined with a distributed hash table (DHT, using the Kademlia protocol) was designed. When a demand-side node publishes a demand, it goes through the following process (see...). Figure 3 ): Step 301: The demand agent automatically de-identifies the original demand information locally, extracts key feature summaries, and the original data remains locally on the demand side without entering any network layer; Steps 302-303: The demand agent uses the system master public key and the target qualification attribute set to perform ciphertext policy attribute base encryption (CP-ABE) on the demand digest to generate a ciphertext demand packet; at the same time, it calculates the SHA-256 hash value of the demand digest and submits the hash value to the smart contract for on-chain notarization to prevent the demand content from being tampered with afterward. Step 304: After attaching the on-chain transaction hash (Tx Hash), qualification category label (plaintext), ERC-20 token price range and cutoff block height to the encrypted demand packet, it is routed and broadcast to all nodes in the network via the Kademlia DHT network; Steps 305-306: Each node in the network performs preliminary screening based on the qualification category label. Only nodes holding the corresponding qualification attribute key (derived from the ZKP certificate on the Layer 0 chain) can decrypt and view the requirement details. Step 307: The demand broadcast event is synchronously written to the blockchain event log to form an immutable on-chain broadcast record for on-chain traceability during subsequent dispute arbitration.

[0025] 3. Asset Delivery Layer (Layer 2) – IPFS Storage and Asset NFT On-Chain Creation A standardized protocol for asset generation, decentralized storage, and NFT on-chain delivery has been established. After a supplier node accepts an order, it executes the following process (see...). Figure 4 ): Step 401: After decrypting the demand summary, the supplier agent automatically assesses the order availability based on its local skills. Step 402: After reviewing the requirements, the supplier agent holder confirms the order and assists the holder in completing the work based on the local private skill. During the asset generation process, all components of the underlying skill (large language model fine-tuning weights, private vector database, judgment logic Prompt, private dataset) never leave the supplier's local device. Only the final structured output enters the network. Step 403: Generate one-time structured output assets (report PDF, solution document, code package, structured dataset, etc.) Step 404: The final assets are end-to-end encrypted using the demander's public key; Step 405: Upload to the InterPlanetary File System (IPFS) node cluster or the Arweave persistent storage network; Step 406: Obtain the Content Addressing Identifier (CID); Step 407: The system mints a non-fungible token (NFT) representing ownership of this one-time asset using the ERC-721 standard. The NFT metadata fields include: asset CID, demand Tx Hash, supplier node public key hash, and generation timestamp block height. The initial holder of the NFT is the supplier, and it will be transferred to the demander after acceptance and confirmation. Step 408: The asset CID and NFT contract address are submitted to the settlement smart contract simultaneously, triggering the verification node VRF selection process. The selected verification node receives a ciphertext copy of the asset for verification.

[0026] 4. Verification Layer (Layer 3) – VRF Random Selection and Multisignature Merkle Notarization This paper proposes a subjective professional asset quality verification scheme based on the verifiable randomness and multi-signature mechanism of blockchain. Its core innovation lies in the fact that the verifier does not make substantive judgments on the asset conclusions, but only checks the logical consistency and format compliance.

[0027] After the asset CID and NFT contract address are submitted to the settlement smart contract, the verification process is triggered (see...). Figure 5 ): Step 501: Receive Layer2 submission (CID + NFT address) Receive asset data sent from Layer2, with the input parameters being the asset CID identifier + ownership NFT address, and start the full-chain verification process.

[0028] Step 502: Call the ChainlinkVRF verifiable random function to obtain a cryptographically secure random seed for subsequent node random selection. Step 503: Randomly select N verification nodes. Based on VRF random numbers, randomly select N nodes from the pool of verification nodes that hold the corresponding qualified ZKP credentials and are available online. Step 504: Solidify the lottery results: Package the VRF random seed, the list of selected nodes, and the corresponding node signature data onto the blockchain and write them into a block to make the selection process publicly verifiable on the blockchain and avoid human manipulation of node selection; Step 505: Verify the node to download assets → Decrypt Node preprocessing: The selected verification node Agent pulls the original data of the encrypted assets corresponding to the off-chain CID, completes the data decryption, and prepares the data for subsequent compliance verification.

[0029] Step 506: On-chain verification rule contract verification. The node agent calls the pre-deployed on-chain verification rule set contract and verifies each of the four preset verification dimensions: logical chain integrity (checking the logical chain integrity of the argument), format compliance (checking the compliance of industry standard format), hash consistency (checking the hash consistency of the cited data and the declared conclusion), and prohibitive statement filtering (blocking conclusions beyond the scope and unauthorized commitments). Step 507: Generate Validation Vector. The node uses pre-built AI prompt word templates in the contract for fully automated AI reasoning and verification to generate a multi-dimensional structured scoring vector (Validation Vector); the node holder does not need to perform any manual operation throughout the process.

[0030] Step 508: Each verification node signs its own generated verification report with its private key, and submits the report Merkle root + node signature to the on-chain multi-signature contract to start signature aggregation.

[0031] Step 509: On-chain multi-signature notarization: Each verification node signs the Merkle root hash of the structured verification report with its private key and submits it through a multi-signature contract; when the M-of-N signature threshold (M≥(N / 2)+1) is met, the contract writes the aggregated Merkle Proof into the block, forming an immutable on-chain verification notarization with multi-party endorsement; Step 510a: Verification failed: The number of signatures did not meet the statutory threshold. The entire verification process was terminated, the business was rolled back, and the asset verification process was invalidated.

[0032] Step 510: After the threshold is met, the contract aggregates valid signatures, generates an aggregated Merkle Proof, and completes the encapsulation of multi-party endorsed cryptographic proof.

[0033] Step 511: Solidify the aggregated Merkle root hash onto the blockchain and write it into a block to generate an immutable multi-party verification certificate.

[0034] Step 512: Verification complete, trigger Layer 4 settlement; The demand agent reads the score vectors of each verification node from the chain and integrates them into a readable summary, which is automatically integrated into a readable business summary and presented to the demand holder. The holder decides whether to confirm acceptance and triggers the asset settlement logic to the downstream Layer 4.

[0035] 5. Settlement Layer (Layer 4) – ERC-20 Token Economics and DAO Governance It employs an Ethereum Virtual Machine (EVM) compatible smart contract system to achieve fully automated on-chain fund custody and tokenomics-driven incentive settlement (see [link]). Figure 6 ): Step 601: During the demand release phase, the demander will agree on the amount. SKILL's native ERC20 tokens are locked in an Escrow custody contract, ensuring that funds and assets are protected by the contract throughout the entire process. Step 602: Verifying nodes pledge reputation-based tokens to the escrow contract. The pledged amount directly determines their weight in the VRF node candidate pool; subsequent verification failures will trigger token deduction or destruction penalties. Step 603: The escrow contract monitors the multi-signature submission progress of the verification layer in real time through an on-chain event listener. Once the multi-signature rules are met, it automatically triggers a token settlement branch judgment. Step 604 - Failure: The node's multi-signature failed to meet the M-of-N signature threshold. The contract unlocks the escrow tokens and refunds them to the requester. The order is terminated. Step 604 - Success: Multi-signature compliance completed, awaiting manual acceptance and confirmation from the requesting party; if no objection is filed after the dispute protection period expires, the process will automatically proceed to the dispute adjudication stage. Step 605a: If the demand party initiates on-chain arbitration during the protection period, a dispute resolution process will be triggered. The VRF will randomly select M=5 additional expert arbitration nodes in the same field to avoid human manipulation. Step 605b: Selected arbitration nodes must pledge additional tokens to increase the cost of default for negligence or malicious arbitration by the arbitrator. Step 605c: The arbitration adopts a majority voting mechanism to issue a final judgment. The arbitration result is stored on the blockchain for evidence, and is verifiable and tamper-proof throughout the entire chain. Step 605d: Based on the on-chain arbitration conclusion, the escrow funds will be reallocated, and the losing party in the arbitration will bear all service fees for all arbitration nodes in this round. Step 605: No arbitration / After arbitration is implemented, the escrow contract automatically splits the tokens: the supplier settles 85% of the main service fee, all verification nodes equally share a total of 10% of the verification fee, and the remaining 5% is allocated to the DAO treasury as protocol operation and maintenance fees. Step 606: The ownership transfer is automatically completed on-chain, and the corresponding qualified asset NFT is transferred from the supplier's wallet to the demander's address, completing the closed-loop physical delivery of the asset. Step 607: Package and solidify the entire chain of staking, arbitration, token allocation, and asset transfer details onto the blockchain, ensuring permanent blockchain storage and on-chain traceability verification. Step 608: Update the full node's on-chain reputation score based on the results of this round of verification and arbitration; when historical verification content is found to have significant flaws, destroy the corresponding node's staked tokens according to the rules to implement penalties for breach of trust. Step 609: The core system parameters (number of nodes N, profit sharing ratio, staking base, dispute period, etc.) are governed by token-weighted voting executed by the DAO governance contract. The protocol parameters can only be modified when the required number of participants and the approval threshold are met.

[0036] Based on the above system, the present invention provides a method for collaborating on private skill assets, specifically including the following steps: Step S1 (Issuance of Qualification Certificates): The supply-side node issues cryptographic certificates to the blockchain network, proving that it holds specific professional qualifications in a particular category, using a zero-knowledge proof protocol. See the description of the identity verification layer for details.

[0037] Step S2 (Encrypted Broadcast of Demand): The demand-side node encrypts the locally anonymized demand content using attribute-based encryption, generates a ciphertext demand packet, and broadcasts it to the peer-to-peer network. Simultaneously, the hash value of the demand digest is stored on the blockchain for evidence. See the demand routing layer description for details.

[0038] Step S3 (Asset Generation and NFT Minting): After receiving and decrypting the encrypted demand package, the supply-side node invokes its local AI Agent's private skill module to generate a one-time output asset for that demand. The asset is stored in a decentralized storage network and a content identifier is obtained. Simultaneously, an NFT representing ownership of the asset is minted. No component of the private skill module (large language model fine-tuning weights, private vector database, judgment logic prompts, private dataset) enters network transmission or on-chain storage; only the final asset and its CID enter the network.

[0039] Step S4 (Multi-node verification): Multiple verification nodes are selected through an on-chain verifiable random source. Each verification node performs formal verification (format compliance and logical consistency) on the output asset, generates verification results, and aggregates them into an auditable on-chain proof by multi-node signatures.

[0040] Step S5 (Acceptance Confirmation): The requesting node reviews the verification results and decides whether to send on-chain acceptance confirmation. Requesting node holders can make their own decisions based on the verification summary; the system does not output conclusive recommendations.

[0041] Step S6 (Settlement and Ownership Transfer): After receiving the acceptance confirmation and the verification result meets the preset conditions (such as M-of-N multi-signature pass), the smart contract automatically completes the token settlement and NFT ownership transfer, and all settlement records are written to the blockchain.

[0042] Taking the healthcare field as an example: A patient requests a second opinion on a rare disease. The requester (the patient or their agent) publishes an encrypted request in the system, requiring a node holding a "Practicing Physician Qualification - Rare Disease Specialty" to respond. The supplier (a doctor with this qualification) uses their local private skills (including the doctor's accumulated rare disease case database, diagnostic logic chain, and latest literature index) to generate a one-time structured report (PDF) for the patient's condition without revealing any private knowledge. This report is stored on IPFS and minted as an NFT. Three randomly selected doctor agents in the same field verify its format compliance and logical consistency (without evaluating the correctness of the medical conclusions). After successful verification and patient acceptance, the smart contract automatically releases token rewards from the patient's escrow account to the doctor, and NFT ownership is transferred to the patient. The doctor's private skills remain locally and cannot be copied or resold.

Claims

1. A private skills asset collaborative network system based on AI Agent and blockchain technology, characterized in that... include: The identity verification layer is used to cryptographically prove to other nodes in the blockchain network that the node holds a valid professional qualification of a specific category without exposing the node holder's identifiable identity information; The demand routing layer is used to broadcast the ciphertext of the demand from demand-side nodes to supply-side nodes in the network that only hold the corresponding qualification certificates, based on attribute encryption or predicate encryption. The asset delivery layer is used by the local AI Agent of the supply-side node to generate one-time output assets for the demand based on its private skill module, and to anchor the decentralized storage identifier of the asset to the blockchain with non-fungible tokens (NFTs) representing its ownership. The verification layer is used to select multiple verification nodes through an on-chain verifiable random source. Each verification node checks the format compliance and logical consistency of the output asset and writes the verification results into the blockchain in the form of a multi-node signature aggregation proof. The settlement layer is used to automatically trigger on-chain settlement of tokens and transfer of ownership of the NFT based on the aggregated proof and the on-chain acceptance confirmation of the demand side.

2. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 1, characterized in that... The specific configuration of the identity verification layer is as follows: verify the real-world professional qualification information of the node through an off-chain multi-signature oracle cluster, generate verifiable qualification certificates based on zk-SNARK or zk-STARK zero-knowledge proof protocol, and write the certificates into the blockchain identity registration smart contract in accordance with the ERC-725 standard. The certificate status is maintained through a Merkle state tree.

3. The private skill asset collaboration network system based on AI Agent and blockchain technology as described in claim 1, characterized in that... The specific configuration of the demand routing layer is as follows: receiving the demand digest after local de-identification processing by the demand party node, encrypting the demand digest with the ciphertext policy attribute base encryption algorithm, generating a ciphertext demand packet that can only be decrypted by nodes holding the corresponding qualification category attribute key, writing the SHA-256 hash value of the demand digest onto the chain for notarization, and broadcasting the ciphertext demand packet to the distributed P2P network through the Kademlia DHT protocol.

4. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 1, characterized in that... The asset delivery layer is specifically configured as follows: the AI ​​Agent of the supply-side node generates a one-time structured output asset for a specific need based on a local private skill module. The content of the private skill module does not enter the network. Only the final output asset is uploaded to the InterPlanetary File System (IPFS) or Arweave permanent storage network after end-to-end encryption to obtain a Content Addressing Identifier (CID). The NFT is then minted using the ERC-721 standard. The metadata of the NFT includes the CID, the demand chain transaction hash (Tx Hash), the supply-side node public key hash, and the generation timestamp block height.

5. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 1, characterized in that... The verification layer is specifically configured as follows: N verification nodes are randomly selected from the nodes holding corresponding qualified ZKP certificates by the on-chain verifiable random function VRF. The AI ​​Agent of the verification node performs logical consistency and format compliance checks on the output asset based on the on-chain verification rule set, generates a multi-dimensional scoring vector, and the verification results of each node are aggregated into Merkle proofs by the M-of-N multi-signature contract and written into the block. The verification rule set is deployed on-chain contracts for different professional qualification categories. The verification content includes: the integrity of the argument based on the logical chain, compliance with industry standard format, consistency of cited data and declaration hash, and filtering of prohibited statements.

6. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 1, characterized in that... The specific configuration of the settlement layer is as follows: the funds of the demand side are held in custody by the ERC-20 functional token. After receiving the verification report that meets the M-of-N multi-signature threshold and the demand side confirms the acceptance on the chain, the contract automatically releases the main token reward to the supply side node, releases a fixed proportion of protocol token fees to each verification node, and injects protocol fees into the DAO treasury. The key parameters of the protocol are updated by the DAO governance contract through voting by token holders.

7. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 6, characterized in that: The settlement module also includes a reputation staking incentive mechanism based on token economics: before participating in verification, a verification node must stake a certain amount of ERC-20 reputation tokens; if the assets passed by the verification node are found to have major defects in subsequent dispute arbitration, the corresponding staked tokens will be destroyed proportionally; the accumulated reputation points on the node chain positively affect its weight coefficient in the subsequent VRF selection.

8. The private skill asset collaboration network system based on AI Agent and blockchain technology according to claim 6, characterized in that... The settlement module also includes an on-chain dispute arbitration sub-process: when the requesting party sends a dispute request transaction within the protection period block height, the VRF will select M expert nodes in the same field to participate in the arbitration, and the arbitration nodes must add pledged tokens; The arbitration result is determined by the majority rule. After the result is written into the block, the contract redistributes the escrow tokens according to the arbitration output, and the losing party bears the token fee of the arbitration node.

9. A method for collaborating on private skill assets based on AI Agent and blockchain technology, characterized in that... Includes the following steps: Step S1: The supply-side node publishes cryptographic credentials to the blockchain network, proving that it holds a specific category of professional qualifications, through a zero-knowledge proof protocol; Step S2: The demand-side node performs attribute-based encryption on the demand content that has been locally de-identified, generates a ciphertext demand packet, and broadcasts it to the peer-to-peer network. At the same time, the hash value of the demand digest is stored on the blockchain for evidence. Step S3: After receiving and decrypting the encrypted demand package, the supply-side node calls the private skill module of its local AI Agent to generate a one-time output asset for the demand, stores the asset in a decentralized storage network and obtains a content identifier, and at the same time mints a non-fungible token (NFT) representing the ownership of the asset. Step S4: Select multiple verification nodes through an on-chain verifiable random source. Each verification node performs formal verification on the output asset, generates verification results, and aggregates them into an auditable on-chain proof by signing multiple nodes. Step S5: The requesting node reviews the verification results and decides whether to send on-chain acceptance confirmation; Step S6: After receiving the acceptance confirmation and the verification result meets the preset conditions, the smart contract automatically completes the token settlement and the transfer of ownership of the NFT.

10. The method for private skill asset collaboration based on AI Agent and blockchain technology according to claim 9, characterized in that: In step S3, no component of the private skill module is transmitted over the network or stored on the blockchain. The components include the large language model fine-tuning weights, the private vector database, the judgment logic prompts, and the private dataset. Only the one-time structured output assets and their content addressing identifiers generated based on the private skill module are transmitted over the network.

Images