Agent Skills: FastGPT SDK Factory

FastGPT SDK Factory

Use this skill when working on a FastGPT Tool Plugin that imports @fastgpt-plugin/sdk-factory or the local sdk/factory package.

Core Pattern

Plugin entry files default-export the instance returned by defineTool() or defineToolSet().

import {
  createToolHandler,
  defineTool,
  type InputSchemaMetaType,
  type OutputSchemaMetaType
} from '@fastgpt-plugin/sdk-factory';
import z from 'zod';

const handler = createToolHandler({
  inputSchema: z.object({
    text: z.string().meta({
      title: 'Text'
    } satisfies InputSchemaMetaType)
  }),
  outputSchema: z.object({
    result: z.string().meta({
      title: 'Result'
    } satisfies OutputSchemaMetaType)
  }),
  handler: async (input) => ({
    result: input.text.toUpperCase()
  })
});

export default defineTool({
  manifest: {
    pluginId: 'uppercase',
    version: '1.0.0',
    name: {
      en: 'Uppercase',
      'zh-CN': '转大写'
    },
    description: {
      en: 'Convert text to uppercase',
      'zh-CN': '将文本转换为大写'
    },
    versionDescription: {
      en: 'Initial version',
      'zh-CN': '初始版本'
    },
    tags: ['tools']
  },
  handler
});

Handler Rules

• Define handlers with createToolHandler({ inputSchema, outputSchema, secretSchema?, handler }).
• Use z.object(...) for inputSchema and outputSchema so TypeScript can infer input and return types.
• Import InputSchemaMetaType, OutputSchemaMetaType, and SecretSchemaMetaType from @fastgpt-plugin/sdk-factory when schema metadata is used.
• Add .meta({ ... } satisfies InputSchemaMetaType) to input fields and .meta({ ... } satisfies OutputSchemaMetaType) to output fields that need UI/manifest metadata.
• Set toolDescription in an input field's .meta() when that field should be available as an automatically filled tool-call parameter; without toolDescription, users must specify the field manually.
• Add .meta({ isSecret: true | false, ... } satisfies SecretSchemaMetaType) to every secretSchema field; set isSecret: true for values that must be encrypted at rest.
• Return an object that matches outputSchema; throw errors for failed operations.
• Add secretSchema when plugin configuration needs secrets such as API keys.
• Read secrets from ctx.secrets; the value is typed from secretSchema.
• Use ctx.streamResponse({ type: 'answer', content }) to emit incremental visible output before the final response.
• Use ctx.systemVar only for FastGPT-provided runtime variables.

const secretSchema = z.object({
  apiKey: z.string().meta({
    title: 'API Key',
    isSecret: true
  } satisfies SecretSchemaMetaType)
});

const handler = createToolHandler({
  inputSchema: z.object({
    query: z.string().meta({
      title: 'Query',
      toolDescription: 'Search query'
    } satisfies InputSchemaMetaType)
  }),
  outputSchema: z.object({
    answer: z.string().meta({
      title: 'Answer'
    } satisfies OutputSchemaMetaType)
  }),
  secretSchema,
  handler: async (input, ctx) => {
    ctx.streamResponse({
      type: 'answer',
      content: `Searching: ${input.query}`
    });

    return {
      answer: `Result for ${input.query}`
    };
  }
});

Manifest Rules

manifest must include these fields unless the surrounding package already supplies them:

• pluginId: stable unique plugin id.
• version: plugin version, usually semver.
• name: i18n object shaped as { en, 'zh-CN' }.
• description: i18n object shaped as { en, 'zh-CN' }.

Common optional fields:

• versionDescription
• author
• repoUrl
• tutorialUrl
• tags
• permission
• icon
• toolDescription

Use permission to declare host capabilities required by the plugin. It is an array of permission strings. Declare only the minimum permissions that the plugin actually uses. When using ctx.invoke to call host capabilities, declare the matching permission.

Supported permissions:

• userInfo:read: read user information, for example through ctx.invoke.userInfo().
• teamInfo:read: read team information.
• model:read: read model information.
• dataset:read: read dataset information.
• file-upload:allow: upload files, for example through ctx.invoke.uploadFile().

Keep pluginId, child tool id, input field names, and output field names stable for compatibility.

Tool Sets

Use defineToolSet() for one plugin with multiple child tools. Put shared metadata in the top-level manifest; put each child tool's id, name, description, optional icon, optional toolDescription, and handler in children.

import {
  createToolHandler,
  defineToolSet,
  type InputSchemaMetaType,
  type OutputSchemaMetaType,
  type SecretSchemaMetaType
} from '@fastgpt-plugin/sdk-factory';
import z from 'zod';

const secretSchema = z.object({
  apiKey: z.string().meta({
    title: 'API Key',
    isSecret: true
  } satisfies SecretSchemaMetaType)
});

const searchHandler = createToolHandler({
  inputSchema: z.object({
    query: z.string().meta({
      title: 'Query'
    } satisfies InputSchemaMetaType)
  }),
  outputSchema: z.object({
    items: z.array(z.string()).meta({
      title: 'Items'
    } satisfies OutputSchemaMetaType)
  }),
  secretSchema,
  handler: async (input) => ({ items: [input.query] })
});

export default defineToolSet({
  secretSchema,
  manifest: {
    pluginId: 'text-tools',
    version: '1.0.0',
    name: {
      en: 'Text Tools',
      'zh-CN': '文本工具集'
    },
    description: {
      en: 'Search and summarize text',
      'zh-CN': '搜索和总结文本'
    }
  },
  children: [
    {
      id: 'search',
      name: {
        en: 'Search',
        'zh-CN': '搜索'
      },
      description: {
        en: 'Search text',
        'zh-CN': '搜索文本'
      },
      toolDescription: 'Search text by query',
      handler: searchHandler
    }
  ]
});

Host Invocation

Use ctx.invoke for host capabilities. The SDK exposes userInfo() and uploadFile(). Declare the corresponding manifest.permission item before using a host capability; for example, ctx.invoke.uploadFile() requires file-upload:allow. These methods return a Result tuple shaped as [result, err]; check err before using result. When err is present, throw or return that original err so host-side error details are preserved.

const uploadHandler = createToolHandler({
  inputSchema: z.object({
    content: z.string().meta({
      title: 'Content'
    } satisfies InputSchemaMetaType)
  }),
  outputSchema: z.object({
    accessURL: z.string().meta({
      title: 'Access URL'
    } satisfies OutputSchemaMetaType),
    fileName: z.string().meta({
      title: 'File Name'
    } satisfies OutputSchemaMetaType),
    size: z.number().meta({
      title: 'Size'
    } satisfies OutputSchemaMetaType)
  }),
  handler: async (input, { invoke }) => {
    const [result, err] = await invoke.uploadFile({
      fileName: 'result.txt',
      contentType: 'text/plain',
      file: Buffer.from(input.content, 'utf-8')
    });

    if (err) {
      throw err;
    }
    if (!result) {
      throw new Error('Failed to upload file');
    }

    return {
      accessURL: result.accessURL,
      fileName: result.fileName,
      size: result.size
    };
  }
});

Local Development

• In external plugins, import from @fastgpt-plugin/sdk-factory.
• Inside this monorepo package tests or fixtures, imports may point at sdk/factory/src or ../../src/index when matching existing local patterns.
• Build the SDK with pnpm --filter @fastgpt-plugin/sdk-factory build.
• When changing SDK behavior, check sdk/factory/src/index.ts, sdk/factory/src/tool-factory.ts, sdk/factory/src/invoke.client.ts, and fixtures under sdk/factory/test/fixtures/.