Skip to content
KloudMate Docs

Summarize a Slack thread with AI

Long Slack threads are where decisions get made and then lost. This guide builds a workflow that summarizes a thread on request: anyone reacts to a message in the thread with 📝, and an AI Prompt step reads the whole thread and replies in it with the decisions, the open questions, and who agreed to do what. The owner of each action item is mentioned, so they get a notification.

The finished workflow has these steps:

Slack: Reaction added, 📝 only
├─ Read the thread               Slack: Get Thread Replies
└─ Is it a thread?               Branch
   └─ Then
      ├─ Summarize the thread    AI Prompt
      └─ Reply with the summary  Slack: Post Message

Before you start

Section titled “Before you start”
  • You need the Developer role in a KloudMate workspace whose plan includes workflows.
  • You need a Slack connection with the Workflows capability. The workflow works in every channel the KloudMate bot is in, so invite the bot to the channels where you want summaries.
  • Each summary uses workflow credits, and the AI Prompt step uses more. See Check workflow usage.

Step 1: Import the workflow

Section titled “Step 1: Import the workflow”
  1. Copy the YAML below.
  2. Open Workflows, click Import, paste the YAML, and click Import.
  3. Click Open workflow. On the trigger, pick your Slack connection under Account, and on the two Slack steps, pick your Slack connection.
kind: workflow
uid: guide-summarize-slack-thread
spec:
  name: Summarize a Slack thread
  description: When someone reacts to a message in a thread with a memo emoji, reads the whole thread and replies in it with a summary of the decisions, open questions, and action items.
  definition:
    schema_version: 1
    trigger:
      type: integration_event
      config:
        connection_id:
          $input: slack
        trigger_key: slack.reaction_added
        props:
          emoji: memo
    steps:
      - id: thread
        type: action
        action: slack.get_thread_replies
        display_name: Read the thread
        connection_id:
          $input: slack
        with:
          channel: "{{ trigger.event.item.channel }}"
          ts: "{{ trigger.event.item.ts }}"
          limit: 200
      - id: has_replies
        type: branch
        display_name: Is it a thread?
        if:
          all:
            - field: steps.thread.output.messages.1
              op: is_not_empty
        then:
          - id: summary
            type: action
            action: ai.prompt
            display_name: Summarize the thread
            with:
              prompt: |-
                Summarize this Slack thread for someone who wasn't part of it. Use Slack formatting, and keep it under 150 words, with these sections:
                *Decisions*: what the thread agreed on.
                *Open questions*: what is still undecided.
                *Action items*: who does what. Name each owner as they appear below, in the form <@USER_ID>.
                Leave out a section that would be empty. Don't add anything the thread doesn't say.

                The thread, oldest message first. Each message starts with its author:
                {% for m in steps.thread.output.messages %}
                <@{{ m.user }}>: {{ m.text }}
                {% endfor %}
          - id: reply
            type: action
            action: slack.post_message
            display_name: Reply with the summary
            connection_id:
              $input: slack
            with:
              channel: "{{ trigger.event.item.channel }}"
              thread_ts: "{{ steps.thread.output.messages.0.ts }}"
              text: ":memo: *Summary, as <@{{ trigger.event.user }}> asked*\n{{ steps.summary.output.text | replace: '<!', '&lt;!' }}"
inputs:
  slack:
    kind: connection
    name: Slack
    type: slack

How each step works

Section titled “How each step works”

The trigger

Section titled “The trigger”

The Slack Reaction added trigger has Only this emoji set to memo, Slack’s name for 📝, so any other reaction is ignored. Only in channel is empty, so a 📝 in any channel the bot is in starts a run. The event carries the reacted message’s channel and timestamp as trigger.event.item.channel and trigger.event.item.ts, and the person who reacted as trigger.event.user.

Read the thread

Section titled “Read the thread”

A Slack Get Thread Replies step reads the thread, with Channel and Thread set from the event. Thread can be the timestamp of any message in the thread, so a 📝 on a reply works as well as one on the first message. Max messages is 200.

Slack returns the thread in messages, oldest first, with the message that started the thread at position 0.

A Branch checks that steps.thread.output.messages.1 is not empty, which means the thread has at least one reply. A 📝 on a message with no replies does nothing.

Summarize the thread

Section titled “Summarize the thread”

An AI Prompt step writes the summary. The Prompt sets the sections and the length, and then lists the thread with a Liquid for loop, one message after another:

{% for m in steps.thread.output.messages %}
<@{{ m.user }}>: {{ m.text }}
{% endfor %}

Each message starts with its author’s user ID in Slack’s mention syntax, <@U02ABCDEF>. The prompt asks the model to name the owner of each action item the same way, so the owners are mentioned when the summary is posted, and Slack shows their names.

The prompt also tells the model not to add anything the thread doesn’t say. A summary that invents a decision is worse than no summary, so check a few summaries against their threads when you start.

AI Prompt reads the messages as part of its instructions, so whoever writes in the thread can steer the summary. That’s acceptable here, because the people in the thread are your colleagues, and the summary only goes back into the same thread. For text from outside your company, use AI Extract instead. See AI Prompt.

Reply with the summary

Section titled “Reply with the summary”

A Slack Post Message step replies in the thread, with Reply to thread set to {{ steps.thread.output.messages.0.ts }}, the first message of the thread. The reply names who asked for the summary:

:memo: *Summary, as <@{{ trigger.event.user }}> asked*
{{ steps.summary.output.text | replace: '<!', '&lt;!' }}

The replace filter turns <! into &lt;!. That keeps the summary from notifying a whole channel with <!channel> or <!here>, even if a message in the thread tricked the model into writing one. User mentions, which start with <@, still work.

Step 2: Test and publish

Section titled “Step 2: Test and publish”
  1. Open the trigger’s Test tab and click Test trigger. Then react with 📝 to a message in a thread in a channel the bot is in. The reaction becomes the trigger’s sample.
  2. Test Read the thread from its Test tab, then test Summarize the thread to see the summary without posting it.
  3. Click Publish. The first publish also switches the workflow on.

Adapt the workflow

Section titled “Adapt the workflow”
  • Save the summary in Jira. For threads about a Jira issue, add an AI Extract step that returns the issue key mentioned in the thread, and a Jira Add Comment step that posts the summary on that issue.
  • Summarize only some channels. Import the workflow once for each channel, each with a different uid, and set Only in channel on each trigger.
  • Write a different summary. Change the sections in the prompt, for example to a timeline for an incident thread, or to a list of open questions only.
Section titled “Related”