Problem

When models emit thinking/reasoning blocks (e.g., Anthropic extended thinking, <think> tags), Slack sessions stream the raw reasoning text inline with the response. This creates long, verbose messages that obscure the actual answer.

Desired Behavior

Render thinking/reasoning content as collapsible sections in Slack messages — similar to how Cursor shows reasoning in a collapsed "Thinking..." block that users can click to expand.

Slack supports this via:

• Block Kit attachments with expandable content
• context blocks for metadata
• Collapsible rich text sections

Implementation Notes

This lives in the Slack provider's outbound message rendering pipeline — the same area as thread context handling (prepare.ts, Slack message formatting). Thinking blocks are already parsed and separated from response content; the change is in how they're presented rather than discarded.

Approach options:

1. Slack attachment — render thinking as a collapsed attachment with a summary line
2. Block Kit — use Slack's rich text blocks with a disclosure pattern
3. Separate message — post thinking as a threaded reply to the main response (most native)

Related

• PR fix(slack): thread history on subsequent turns, inbound meta context, and heartbeat thread leaking #16186 — Slack thread context (inbound side of the same pipeline)
• Issues [Bug]: Slack thread sessions don't load prior thread messages (including thread parent) #12742, Slack: Auto-fetch thread context on inbound thread replies #12586 — Thread context gaps

🤖 AI-assisted (Claude via OpenClaw). Author understands the proposed changes and Slack API constraints.