Back to Copilotkit

@copilotkit/bot-slack

packages/bot-slack/README.md

1.60.29.7 KB
Original Source

@copilotkit/bot-slack

The Slack PlatformAdapter for @copilotkit/bot. It connects a Slack workspace to any AG-UI agent: ingress via Bolt (Socket Mode), egress as Block Kit rendered from the @copilotkit/bot-ui JSX vocabulary, plus text streaming, opaque-id interactions, and HITL.

You write your UI as JSX once (@copilotkit/bot-ui) and drive the bot with @copilotkit/bot; this package is the only one that talks to Slack.

Install

sh
pnpm add @copilotkit/bot-slack @copilotkit/bot @copilotkit/bot-ui

Quickstart

ts
import { createBot } from "@copilotkit/bot";
import {
  slack,
  defaultSlackTools,
  defaultSlackContext,
} from "@copilotkit/bot-slack";

const bot = createBot({
  adapters: [
    slack({
      botToken: process.env.SLACK_BOT_TOKEN!, // xoxb-…
      appToken: process.env.SLACK_APP_TOKEN!, // xapp-… (Socket Mode)
    }),
  ],
  agent: (threadId) => makeAgent(threadId),
  tools: [...defaultSlackTools, ...appTools], // lookup_slack_user + your tools
  context: [...defaultSlackContext, ...appContext], // tagging/mrkdwn/thread guidance
});

bot.onMention(({ thread }) => thread.runAgent());

await bot.start();

slack(opts) returns a SlackAdapter. By default it runs in Socket Mode (socketMode: true) — outbound WebSocket only, no public URL needed. HTTP mode (socketMode: false) needs signingSecret and a port. The Slack listener pre-filters ingress to the turns the bot should answer (@-mentions, replies in threads it owns, DMs), so a single onMention handler usually covers everything.

Required env

VarTokenPurpose
SLACK_BOT_TOKENxoxb-Bot token for the Web API.
SLACK_APP_TOKENxapp-App-level token for Socket Mode.

What it provides

JSX → Block Kit rendering

renderSlackMessage(ir) / renderBlockKit(ir) translate the @copilotkit/bot-ui vocabulary to Block Kit: Message → blocks, Header → header, Section → section (mrkdwn), Markdown → markdownToMrkdwn, Field(s) → section.fields, Context → context, Actions → actions, Button → button (action_id = minted opaque id), Select → static_select, Input → plain_text_input, Image → image, Divider → divider.

Per-element budget

Slack caps every element. The renderer degrades by truncate-with-overflow / clamp — it never silently drops content. Limits live in SLACK_LIMITS:

LimitValueElement
blocksPerMessage50blocks per message
sectionText3000section body chars
headerText150header chars
fieldsPerSection10fields per section
fieldText2000field chars
actionsElements25controls per actions row
contextElements10elements per context block
buttonText75button label chars
actionId255action_id chars
buttonValue2000button value chars
selectOptions100options per select

Colored cards

<Message accent="#RRGGBB"> renders as a Slack attachment with a colored left bar (Block Kit blocks have no native accent, so accented messages are posted as attachments: [{ color, blocks }]).

Streaming

By default, replies stream via Slack's native streaming API (chat.startStream / appendStream / stopStream) wherever the reply target is a thread — a true streaming UI rendering raw markdown (so real tables and fenced code render natively), with the same throttle budget and fence-aware multi-message continuation as the legacy path. Flat DMs (no thread) and any workspace where the streaming API is unavailable fall back automatically to the shipped chat.update transport (throttled edits, multi-message chunking, mid-stream bracket auto-close, Markdown → mrkdwn translation). Pass streaming: "legacy" to force the chat.update transport everywhere. The fallback is transparent — opting in can never break a bot: the first startStream failure marks the workspace legacy and redoes the stream the old way.

Assistant pane (agent-native, default-on)

When the Slack app has the Agents & AI Apps toggle (an assistant_view manifest block + the assistant:write scope and assistant_thread_* events), the adapter activates Slack's assistant pane with zero config:

  • Opening the pane posts a greeting + tappable prompt chips, and each pane conversation is its own thread (replies stay in-thread).
  • While the agent runs, native composer status is shown (assistant.threads.setStatus: "is thinking…", "is using `tool`…") instead of placeholder/:wrench: messages.
  • The pane thread is auto-titled from the first message.

Customize via the assistant option, or set assistant: false to disable pane handling entirely. Apps without the toggle behave exactly as before — the pane machinery lies dormant.

ts
slack({
  botToken,
  appToken,
  assistant: {
    greeting: "Hi! I can triage issues, search docs, and more.",
    suggestedPrompts: [
      { title: "Triage my open issues", message: "Triage my open issues" },
    ],
  },
});

// Dynamic behavior when a user opens the pane (layers on top of the defaults):
bot.onThreadStarted(async ({ thread, user }) => {
  await thread.setSuggestedPrompts(promptsFor(user));
  // await thread.setTitle(...) is also available
});

Interactions (ack-first)

Every Slack block_actions click is acked immediately (within the ≤3s deadline, ackDeadlineMs = 3000), then decodeInteraction extracts the opaque minted id (ck:…), any tiny bind() value, and the message ref, and hands an InteractionEvent to the engine. The token carries only the opaque id — no props or secrets. Unrelated clicks decode to events the bot harmlessly ignores.

Human-in-the-loop

Use thread.awaitChoice(<Picker .../>) to post an interactive message and block until a click resolves it; the resolved value is the clicked control's value. Agent interrupts (on_interrupt) are captured by the run renderer and dispatched to your onInterrupt handler, which posts a picker; the click resumes the agent via thread.resume(value).

Sender-profile resolution & file download

The adapter resolves each turn's Slack user id to a richer PlatformUser ({ id, name?, email? }), cached per id. Inbound files can be downloaded and delivered to the agent as multimodal content parts (buildFileContentParts); a tool can post a file back out via thread.postFile(...).

Built-ins

  • defaultSlackTools — ships lookup_slack_user so the agent can resolve a name/handle/email to a <@USERID> mention. Spread into tools.
  • defaultSlackContext — tagging procedure, Markdown-vs-mrkdwn guidance, and the Slack thread/DM conversation model. Spread into context.

Tool context

There is no Slack-specific tool context. Tools receive the single shared BotToolContext from @copilotkit/bot ({ thread, message?, user?, signal?, platform }) and reach Slack power only through capability-gated thread methods, which this adapter backs:

  • thread.getMessages() — the current thread's messages (via conversations.replies), each a ThreadMessage ({ user?, text, ts?, isBot? }).
  • thread.lookupUser(query) — resolve a name/handle/email to a PlatformUser.
  • thread.postFile({ bytes, filename, title?, altText? }) — upload a file back into the thread (files.uploadV2).

This keeps tools portable: define them with defineBotTool({...}) and they work against any adapter that advertises the same capabilities.

Running the demo

This package is the library. A runnable end-to-end demo wiring all of the above against a real workspace lives in examples/slack.

Slash commands

The adapter forwards every slash command Slack delivers to the engine, which routes it to the matching bot.onCommand handler (and ignores unregistered ones). Register handlers on the engine — see @copilotkit/bot:

ts
bot.onCommand({
  name: "triage",
  description: "Summarize the thread and propose issues.",
  async handler({ thread, text, user }) {
    await thread.runAgent({ prompt: `Triage: ${text}` });
  },
});

You must also declare each command in the Slack app config ("Slash Commands" / app manifest) with the same name — Slack won't deliver an unregistered command, even over Socket Mode. Args arrive as free text (ctx.text); the optional options schema is for surfaces with native structured args (e.g. Discord) and is unused on Slack. The adapter does not implement registerCommands, so the engine skips it (Slack matches commands dynamically rather than registering them up front).

What's NOT in v1

  • Modals / true batched form submit
  • OAuth / multi-workspace install (single bot token only)
  • Durable (Redis/DB) ActionStore — in-memory only; actions expire on restart
  • Proactive posting (bot replies only to turns it's part of)
  • Reactions

Exports

slack, SlackAdapter, SlackAdapterOptions, SlackAssistantOptions; createRunRenderer; decodeInteraction, conversationKeyOf; renderBlockKit, renderSlackMessage, SLACK_LIMITS; defaultSlackTools, lookupSlackUserTool, defaultSlackContext (+ the individual context entries); markdownToMrkdwn; and the preserved mechanics (SlackConversationStore, MessageStream, ChunkedMessageStream, NativeMessageStream, attachSlackListener, attachAssistant, SanitizingHttpAgent, buildFileContentParts, autoCloseOpenMarkdown, and supporting types).