Skip to main content

​
Code Review

import { Injectable } from '@nestjs/common'
import { Document } from '@langchain/core/documents'
import {
  DocumentTransformerStrategy,
  IDocumentTransformerStrategy,
  IntegrationPermission,
  TDocumentTransformerConfig,
} from '@xpert-ai/plugin-sdk'
import { IconType, IKnowledgeDocument } from '@metad/contracts'
import { iconImage, LarkDocumentMetadata, LarkDocumentName, LarkName } from './types.js'
import { LarkClient } from './lark.client.js'

@Injectable()
@DocumentTransformerStrategy(LarkDocumentName)
export class LarkDocTransformerStrategy implements IDocumentTransformerStrategy<TDocumentTransformerConfig> {

  readonly permissions = [
    {
      type: 'integration',
      service: LarkName,
      description: 'Access to Lark system integrations'
    } as IntegrationPermission,
  ]

  readonly meta = {
    name: LarkDocumentName,
    label: {
      en_US: 'Lark Document',
      zh_Hans: '飞书文档'
    },
    description: {
      en_US: 'Load content from Lark documents',
      zh_Hans: '加载飞书文档内容'
    },
    icon: {
      type: 'image' as IconType,
      value: iconImage,
      color: '#14b8a6'
    },
    helpUrl: 'https://open.feishu.cn/document/server-docs/docs/docs-overview',
    configSchema: {
      type: 'object',
      properties: {},
      required: []
    }
  }

  validateConfig(config: any): Promise<void> {
    throw new Error('Method not implemented.')
  }

  async transformDocuments(
    files: Partial<IKnowledgeDocument<LarkDocumentMetadata>>[],
    config: TDocumentTransformerConfig
  ): Promise<Partial<IKnowledgeDocument<LarkDocumentMetadata>>[]> {
    const integration = config?.permissions?.integration
    if (!integration) {
      throw new Error('Integration system is required')
    }

    console.log('LarkDocTransformerStrategy transformDocuments', files, config)

    const client = new LarkClient(integration)
    
    const results: Partial<IKnowledgeDocument<LarkDocumentMetadata>>[] = []
    for await (const file of files) {
      const content = await client.getDocumentContent(file.metadata.token)
      results.push({
        id: file.id,
        chunks: [
          new Document({
            id: file.id,
            pageContent: content,
            metadata: {
              chunkId: file.id,
              source: LarkName,
              sourceId: file.id
            }
          })
        ],
        metadata: {
          assets: []
        } as LarkDocumentMetadata
      })
    }
    return results
  }
}

​
Logic Breakdown

​
1. Decorators and Dependency Injection

@Injectable()
@DocumentTransformerStrategy(LarkDocumentName)
  • @Injectable(): NestJS dependency injection decorator, marks this as an injectable service.
  • @DocumentTransformerStrategy(LarkDocumentName): Registers the class as a document transformation strategy with the unique name LarkDocumentName. 👉 This allows the system to automatically recognize and use this strategy.

​
2. Permission Definition

readonly permissions = [
  {
    type: 'integration',
    service: LarkName,
    description: 'Access to Lark system integrations'
  } as IntegrationPermission,
]
  • The plugin requires Lark integration permission to call the API and fetch documents.
  • IntegrationPermission declares the dependent service, here it’s LarkName (Lark).

​
3. Metadata (meta)

readonly meta = {
  name: LarkDocumentName,
  label: {
    en_US: 'Lark Document',
    zh_Hans: '飞书文档'
  },
  description: {
    en_US: 'Load content from Lark documents',
    zh_Hans: '加载飞书文档内容'
  },
  icon: {
    type: 'image' as IconType,
    value: iconImage,
    color: '#14b8a6'
  },
  helpUrl: 'https://open.feishu.cn/document/server-docs/docs/docs-overview',
  configSchema: { ... }
}
  • Plugin UI display info: name, icon, description, help documentation link.
  • configSchema: Defines configuration options (empty here, meaning no extra parameters required).

​
4. Configuration Validation

validateConfig(config: any): Promise<void> {
  throw new Error('Method not implemented.')
}
  • Placeholder method for future configuration validation.
  • For example: check if document ID or token is provided.

​
5. Core Document Transformation Logic

async transformDocuments(
  files: Partial<IKnowledgeDocument<LarkDocumentMetadata>>[],
  config: TDocumentTransformerConfig
): Promise<Partial<IKnowledgeDocument<LarkDocumentMetadata>>[]> {
  const integration = config?.permissions?.integration
  if (!integration) {
    throw new Error('Integration system is required')
  }

  const client = new LarkClient(integration)
  
  const results: Partial<IKnowledgeDocument<LarkDocumentMetadata>>[] = []
  for await (const file of files) {
    const content = await client.getDocumentContent(file.metadata.token)
    results.push({
      id: file.id,
      chunks: [
        new Document({
          id: file.id,
          pageContent: content,
          metadata: {
            chunkId: file.id,
            source: LarkName,
            sourceId: file.id
          }
        })
      ],
      metadata: {
        assets: []
      } as LarkDocumentMetadata
    })
  }
  return results
}
Line-by-line explanation:
  1. Get Integration Info 
    const integration = config?.permissions?.integration
if (!integration) throw new Error('Integration system is required')
    • Retrieves Lark integration credentials from config.
    • Throws error if credentials are missing.
  2. Initialize Client 
    const client = new LarkClient(integration)
    • Constructs LarkClient with credentials to access Lark API.
  3. Process Files in a Loop 
    for await (const file of files) {
  const content = await client.getDocumentContent(file.metadata.token)
}
    • Iterates over the list of documents to process.
    • Calls client.getDocumentContent to fetch document content by token.
  4. Build Transformed Document 
    results.push({
  id: file.id,
  chunks: [
    new Document({
      id: file.id,
      pageContent: content,
      metadata: {
        chunkId: file.id,
        source: LarkName,
        sourceId: file.id
      }
    })
  ],
  metadata: {
    assets: []
  } as LarkDocumentMetadata
})
    • Each Lark document is converted to an IKnowledgeDocument.
    • Main content is placed in the chunks array.
    • metadata stores extra info (currently only assets).

​
Overall Execution Flow

  1. Input: A batch of Lark document metadata (file ID / token).
  2. Permission Validation: Ensure Lark integration config is present.
  3. API Call: Use LarkClient to fetch the content of each document.
  4. Transform to Knowledge Base Format:
    • Wrap as IKnowledgeDocument
    • Content is chunked into Document (for later vectorization)
  5. Output: Returns an array of documents usable by Xpert AI Knowledge Base.

​
Core Value

  • Decoupling: The strategy class does not call the API directly, but relies on LarkClient.
  • Generality: All documents are ultimately converted to IKnowledgeDocument, seamlessly integrating with the platform’s knowledge base.
  • Extensibility: In the future, you can add to transformDocuments:
    • Text cleaning (remove empty lines/formatting)
    • Content chunking
    • Metadata enhancement (author, tags, update time)