Some checks failed
App Prod-Parity E2E Dispatch / dispatch (push) Failing after 10s
CD Storybook UI / build (push) Failing after 3m25s
CD Storybook UI / dispatch (push) Has been skipped
CD deploy main / website-changed-files (push) Successful in 18m8s
CI Agent Skills / changed-files-check (push) Successful in 18m9s
CI Create App E2E minimal / changed-files-check (push) Successful in 24s
CI Create App / changed-files-check (push) Successful in 26s
CI Docs / changed-files-check (push) Successful in 26s
CI Emails / changed-files-check (push) Successful in 28s
CI Example App Hello World / changed-files-check (push) Successful in 23s
CI Front / changed-files-check (push) Has been skipped
CI Example App Postcard / changed-files-check (push) Successful in 27s
CI UI / changed-files-check (push) Has been skipped
CI Utils Test / changed-files-check (push) Successful in 45s
CI Twenty Apps / discover (push) Successful in 1m3s
CD deploy main / deploy-main (push) Failing after 3s
CI Agent Skills / agent-skills-validate (push) Has been skipped
CI Create App / create-app-test (lint) (push) Has been skipped
CI Emails / emails-test (push) Has been skipped
CI Front / front-task (lingui:compile) (push) Has been skipped
CI Front / front-task (lint) (push) Has been skipped
CI Example App Postcard / example-app-postcard (push) Has been skipped
CI UI / ui-task (lint) (push) Has been skipped
CI UI / ui-task (pack-size) (push) Has been skipped
CI UI / ui-task (size) (push) Has been skipped
CI UI / ui-task (test) (push) Has been skipped
CI UI / ui-task (typecheck) (push) Has been skipped
CI Utils Test / utils-test (push) Has been skipped
CI Twenty Apps / ${{ matrix.app.name }} (push) Has been skipped
CI Emails / ci-emails-status-check (push) Successful in 2s
CI Example App Hello World / ci-example-app-hello-world-status-check (push) Successful in 2s
CI Example App Postcard / ci-example-app-postcard-status-check (push) Successful in 3s
CI UI / ui-sb-test (push) Has been skipped
CI Utils Test / ci-utils-test-status-check (push) Successful in 2s
CI Twenty Apps / ci-twenty-apps-status-check (push) Successful in 2s
CI UI / ci-ui-status-check (push) Successful in 2s
CI Desktop / desktop (push) Failing after 1m8s
Push docs to Crowdin / Push documentation to Crowdin (push) Failing after 55s
Push translations to Crowdin / Extract and upload translations (push) Failing after 5s
Push website translations to Crowdin / Extract and upload website translations (push) Failing after 5s
CI Create App / create-app-test (test) (push) Has been skipped
CI Create App / create-app-test (typecheck) (push) Has been skipped
CI Docs / docs-lint (push) Has been skipped
CI Example App Hello World / example-app-hello-world (push) Has been skipped
CI Front / front-task (lingui:extract) (push) Has been skipped
CI Front / front-task (test) (push) Has been skipped
CI Front / front-task (typecheck) (push) Has been skipped
CI UI / ui-sb-build (push) Failing after 1m32s
CI Twenty Apps / ${{ matrix.app.name }} (${{ matrix.server-source }}) (push) Has been skipped
CI Agent Skills / agent-skills-publish (push) Has been skipped
CI Create App / ci-create-app-status-check (push) Successful in 2s
CI Agent Skills / agent-skills-remote-source (push) Has been skipped
CI Create App E2E minimal / create-app-e2e-minimal (push) Has been cancelled
CI E2E Main / e2e-test (push) Has been cancelled
CI Front / front-sb-build (push) Has been cancelled
CI Front / front-build (push) Has been cancelled
CI Create App E2E minimal / ci-create-app-e2e-minimal-status-check (push) Has been cancelled
CI E2E Main / ci-e2e-main-status-check (push) Has been cancelled
CI E2E Main / notify-main-ci-failure (push) Has been cancelled
CI E2E Main / qa-scout-comment (push) Has been cancelled
CI Front / front-sb-test (push) Has been cancelled
CI Front / front-sb-screenshots (push) Has been cancelled
CI Front / ci-front-status-check (push) Has been cancelled
DPA Sub-Processor Sync / sync-subprocessors (push) Failing after 3s
AI Catalog Sync / sync-catalog (push) Failing after 3s
180 lines
8.9 KiB
Plaintext
180 lines
8.9 KiB
Plaintext
---
|
||
title: 技能与智能体
|
||
description: 为你的应用定义 AI 技能和智能体。
|
||
icon: robot
|
||
---
|
||
|
||
<Warning>
|
||
技能和智能体目前处于 Alpha 阶段。 该功能可用,但仍在演进中。
|
||
</Warning>
|
||
|
||
应用可以定义存在于工作区内的 AI 能力——可复用的技能指令以及具有自定义系统提示词的智能体。
|
||
|
||
<AccordionGroup>
|
||
<Accordion title="defineSkill" description="定义 AI 智能体技能">
|
||
|
||
技能定义了可复用的指令和能力,AI 智能体可在你的工作区中使用。 使用 `defineSkill()` 定义带内置校验的技能:
|
||
|
||
```ts src/skills/example-skill.ts
|
||
import { defineSkill } from 'twenty-sdk/define';
|
||
|
||
export default defineSkill({
|
||
universalIdentifier: 'a1b2c3d4-e5f6-7890-abcd-ef1234567890',
|
||
name: 'sales-outreach',
|
||
label: 'Sales Outreach',
|
||
description: 'Guides the AI agent through a structured sales outreach process',
|
||
icon: 'IconBrain',
|
||
content: `You are a sales outreach assistant. When reaching out to a prospect:
|
||
1. Research the company and recent news
|
||
2. Identify the prospect's role and likely pain points
|
||
3. Draft a personalized message referencing specific details
|
||
4. Keep the tone professional but conversational`,
|
||
});
|
||
```
|
||
|
||
关键点:
|
||
* `name` 是该技能的唯一标识字符串(推荐使用 kebab-case)。
|
||
* `label` 是在 UI 中显示的人类可读名称。
|
||
* `content` 包含技能指令——这是 AI 智能体使用的文本。
|
||
* `icon`(可选)设置在 UI 中显示的图标。
|
||
* `description`(可选)提供有关技能用途的更多上下文。
|
||
|
||
</Accordion>
|
||
<Accordion title="defineAgent" description="使用自定义提示词定义 AI 智能体">
|
||
|
||
智能体是在你的工作区内驻留的 AI 助手。 使用 `defineAgent()` 来创建带有自定义系统提示词的智能体:
|
||
|
||
```ts src/agents/example-agent.ts
|
||
import { defineAgent } from 'twenty-sdk/define';
|
||
|
||
export default defineAgent({
|
||
universalIdentifier: 'b3c4d5e6-f7a8-9012-bcde-f34567890123',
|
||
name: 'sales-assistant',
|
||
label: 'Sales Assistant',
|
||
description: 'Helps the sales team draft outreach emails and research prospects',
|
||
icon: 'IconRobot',
|
||
prompt: 'You are a helpful sales assistant. Help users with their questions and tasks.',
|
||
});
|
||
```
|
||
|
||
关键点:
|
||
* `name` 是该智能体的唯一标识字符串(推荐使用 kebab-case)。
|
||
* `label` 是在 UI 中显示的名称。
|
||
* `prompt` 是定义智能体行为的系统提示词。
|
||
* `description`(可选)提供有关智能体功能的上下文。
|
||
* `icon`(可选)设置在 UI 中显示的图标。
|
||
* `modelId`(可选)会覆盖该智能体使用的默认 AI 模型。
|
||
* `responseFormat`(可选)控制代理输出的结构形式。 对于自由格式文本,默认值为 `{ type: 'text' }`。 使用 `{ type: 'json', schema }` 来强制生成结构化 JSON 输出。
|
||
|
||
默认情况下,代理返回自由格式文本。 要获取结构化输出,将 `responseFormat` 设置为 `{ type: 'json' }`,并提供一个 `schema`:
|
||
|
||
```ts src/agents/structured-agent.ts
|
||
import { defineAgent } from 'twenty-sdk/define';
|
||
|
||
export default defineAgent({
|
||
universalIdentifier: 'c4d5e6f7-a8b9-0123-cdef-456789012345',
|
||
name: 'lead-scorer',
|
||
label: 'Lead Scorer',
|
||
prompt: 'Score the lead and explain your reasoning.',
|
||
responseFormat: {
|
||
type: 'json',
|
||
schema: {
|
||
type: 'object',
|
||
properties: {
|
||
score: { type: 'number', description: 'Lead score from 0 to 100' },
|
||
summary: { type: 'string', description: 'Short reasoning for the score' },
|
||
},
|
||
required: ['score', 'summary'],
|
||
additionalProperties: false,
|
||
},
|
||
},
|
||
});
|
||
```
|
||
|
||
架构说明:
|
||
* 该架构是一个扁平对象:每个属性的 `type` 必须是原始类型(`string`、`number` 或 `boolean`)。 不支持嵌套对象和数组。
|
||
* 每个属性上的 `description`(可选)用于引导模型应在此处填入什么内容。
|
||
* `required`(可选)列出模型必须始终返回的属性。
|
||
* `additionalProperties: false`(可选)禁止任何未在 `properties` 中声明的属性。
|
||
|
||
</Accordion>
|
||
<Accordion title="runAgent" description="从逻辑函数中运行代理">
|
||
|
||
`runAgent()` 允许逻辑函数运行你的应用的某个代理(及其技能和工具)。 通过你传递给 `defineAgent()` 的 `universalIdentifier` 来标识该代理。 只传入 `prompt` 字符串或 `messages` 会话历史其中之一,不要同时传入两者:
|
||
|
||
```ts src/logic-functions/run-enricher.ts
|
||
import { runAgent } from 'twenty-sdk/logic-function';
|
||
|
||
const { result, error, success } = await runAgent({
|
||
agentUniversalIdentifier: 'b3c4d5e6-f7a8-9012-bcde-f34567890123',
|
||
prompt: 'Enrich House Ad <recordId>: fill empty fields from its listing URL.',
|
||
});
|
||
```
|
||
|
||
对于多轮对话机器人(Slack、Discord、Teams 等),将线程历史作为 `messages` 传入,而不是单个 `prompt`:
|
||
|
||
```ts src/logic-functions/reply-in-thread.ts
|
||
import { runAgent } from 'twenty-sdk/logic-function';
|
||
|
||
const { result, error, success } = await runAgent({
|
||
agentUniversalIdentifier: 'b3c4d5e6-f7a8-9012-bcde-f34567890123',
|
||
messages: [
|
||
{ role: 'user', content: 'Who owns Acme?' },
|
||
{ role: 'assistant', content: 'Sarah owns the Acme account.' },
|
||
{ role: 'user', content: 'What was the last touchpoint?' },
|
||
],
|
||
});
|
||
```
|
||
|
||
关键点:
|
||
* 提供 `prompt`(字符串)或 `messages`(1 到 100 个 `{ role: 'user' | 'assistant', content: string }` 条目)中的 **恰好一个**。
|
||
* 代理以**同步**方式运行,并且可以通过其自身的工具读取/更新记录 —— `runAgent()` 会在运行完成后才 resolve。
|
||
* 一个应用只能运行它自己的代理。
|
||
* 应用的[默认角色](/l/zh/developers/extend/apps/config/roles)必须授予 `AI` 权限标记 —— 在其 `permissionFlagUniversalIdentifiers` 中添加 `SystemPermissionFlag.AI`(或设置 `canAccessAllTools: true`)。
|
||
否则,`runAgent()` 会因权限错误而失败。
|
||
* 在逻辑函数上设置较大的 `timeoutSeconds` 值 —— 代理运行可能需要数秒时间。
|
||
* 当运行完成时,`success` 为 `true` 且 `result` 为非空;失败时,`success` 为 `false`,`result` 为 `null`,并且 `error` 保存失败原因(例如,当工作区在运行过程中耗尽 AI 额度时)。
|
||
|
||
```ts src/roles/default-role.ts
|
||
import { defineApplicationRole, SystemPermissionFlag } from 'twenty-sdk/define';
|
||
|
||
export default defineApplicationRole({
|
||
universalIdentifier: 'b648f87b-1d26-4961-b974-0908fd991061',
|
||
label: 'Default function role',
|
||
// runAgent() requires the AI permission flag on the app's default role.
|
||
permissionFlagUniversalIdentifiers: [SystemPermissionFlag.AI],
|
||
});
|
||
```
|
||
|
||
<Warning>
|
||
\*\*避免循环:\*\*如果你从 `*.updated` 数据库事件触发器中调用 `runAgent()`,而代理又会更新同一条记录,请将触发器的 `updatedFields` 限定为代理永远不会写入的字段(例如来源 URL),或者在调用 `runAgent()` 之前检查任何目标字段是否仍为空。
|
||
</Warning>
|
||
|
||
### 以工作区成员身份运行
|
||
|
||
当运行是由某个用户触发时(例如,一个聊天机器人在回复消息),传入 `runAsWorkspaceMemberId`,这样智能体就会以该成员的身份而不是以应用自身的身份执行:
|
||
|
||
```ts src/logic-functions/answer-question.ts
|
||
const { result } = await runAgent({
|
||
agentUniversalIdentifier: 'b3c4d5e6-f7a8-9012-bcde-f34567890123',
|
||
prompt: 'How many open opportunities do we have?',
|
||
runAsWorkspaceMemberId: '20202020-0687-4c41-b707-ed1bfca972a7',
|
||
});
|
||
```
|
||
|
||
然后,此运行将以**成员自身的角色**执行:它可以执行该成员能够执行的所有操作且仅限于此,它创建的记录将归属于该成员,并且会应用该成员的行级权限。 智能体的角色不会参与——它是针对没有任何成员在背后参与的运行所使用的应用默认角色。 对于自主运行(计划任务、数据库事件触发器),省略该字段:这类运行会保持智能体自身的角色。 为保持来源追溯,执行操作的应用会附着在该运行的上下文中,但不会收窄该成员的权限。
|
||
|
||
你的应用负责将触发运行的用户映射到一个工作区成员。 要为某个成员命名,需要一个应用访问令牌,而令牌可以为谁命名,取决于它是否携带用户:
|
||
|
||
* **未绑定任何用户** 的令牌可以为任意成员命名。 逻辑函数使用这类令牌运行,通过 `client_credentials` 或由 API 密钥签发的令牌也属于此类。
|
||
* **代表某个用户签发的令牌**(例如前端组件接收到的令牌),只能为该用户自己的成员命名。
|
||
|
||
除此之外的任何情况——普通用户会话、没有应用访问令牌的 API 密钥——都不能为任何成员命名。
|
||
|
||
<Warning>
|
||
当工作区成员无法解析时——例如未知或已移除的成员,或者没有角色的成员——`runAgent()` 会抛出异常。 它绝不会回退到代理自身的角色,因为那样会授予比调用方请求更多的访问权限。
|
||
</Warning>
|
||
|
||
</Accordion>
|
||
</AccordionGroup>
|