slack-thinking-steps

OpenClaw plugin that exposes Slack's native Thinking Steps API as agent tools. The model publishes a plan upfront, updates each step with collapsible reasoning as it works, then writes its final answer normally.

What it does

Instead of the model dumping its reasoning inline (cluttering the Slack conversation), this plugin gives it tools to stream thinking as native Slack collapsible blocks — task cards with expand chevrons that users can click to see details.

Before: Wall of reasoning text followed by the answer. After: Clean checklist of steps (collapsed by default) followed by the answer.

Tools

Requirements

• OpenClaw >=2026.3.24-beta.2
• Slack app with Agents & AI Apps enabled and chat:write scope
• For channel threads (not DMs): recipient_user_id and recipient_team_id are required by Slack's API

Installation

# Local development
openclaw plugins install /path/to/slack-thinking-steps -l

# From ClawHub (when published)
openclaw plugins install clawhub:@your-org/slack-thinking-steps

Configuration

No plugin-specific config needed. The plugin reads the Slack bot token automatically from channels.slack.botToken in openclaw.json, falling back to the SLACK_BOT_TOKEN environment variable. The model extracts the channel and thread from conversation metadata.

{
  plugins: {
    entries: {
      "slack-thinking-steps": {
        enabled: true
      }
    }
  }
}

Sample prompt

Add this to your agent's system prompt or AGENTS.md to teach the model how to use the tools:

Slack Thinking Steps

You have 3 tools for showing your thinking as collapsible Slack blocks. Use them on every non-trivial request. Never output intermediate reasoning as regular messages — all thinking, narration, and progress updates MUST go through slack_stream_task updates.

When you output thinking as regular text, it gets delivered to Slack as a visible message that clutters the conversation. Instead, wrap ALL intermediate reasoning in slack_stream_task output fields — they collapse behind a chevron, keeping the conversation clean. The only regular message you should write is your final answer after calling slack_stream_stop.

Tools:

• slack_stream_start — Opens a thinking stream. Use display_mode: "plan".
• slack_stream_task — Adds or updates a task card. task_id to reuse, title always visible, status (pending/in_progress/complete/error), output for ALL reasoning and narration (collapsed behind chevron).
• slack_stream_stop — Closes the stream. Call when done, then write your final answer.

Extracting Slack metadata from every inbound message:

Small tasks (single step):

1. Open a stream with a single task:

   slack_stream_start(channel="...", thread_ts="...", display_mode="plan",
     initial_task_id="task", initial_task_title="Checking Redis cluster health")
   

2. Do the work. Put all reasoning in output:

   slack_stream_task(task_id="task", title="Checking Redis cluster health", status="in_progress",
     output="Queried 6 PromQL metrics. All 11 nodes responding, memory at 64-68%...")
   

3. Mark complete and stop:

   slack_stream_task(task_id="task", title="Checking Redis cluster health", status="complete",
     output="All 11 nodes up. Memory 64-68%. No evictions.")
   slack_stream_stop()
   

4. Write your final answer normally.

Large tasks (multi-step plan):

1. Break the task into steps.
2. Open a stream with all steps as pending:

   slack_stream_start(channel="...", thread_ts="...", display_mode="plan",
     initial_task_id="step1", initial_task_title="Check Redis health")
   slack_stream_task(task_id="step2", title="Check MongoDB health", status="pending")
   slack_stream_task(task_id="step3", title="Review alert rules", status="pending")
   

3. Work through each step — set in_progress before, complete after with reasoning in output.
4. Call slack_stream_stop, then write your final answer.

Rules:

• NEVER output intermediate thinking as regular messages. All reasoning goes in slack_stream_task output fields.
• Keep task titles short — they're the checklist items users scan.
• Update output as you go, not just at the end.
• Always call slack_stream_stop before your final answer.
• In channels (C...), always pass recipient_user_id and recipient_team_id.

Development

npm install
npm test           # run tests
npm run dev        # watch mode
npm run build      # compile to dist/

License

MIT