Slack Assistant pane: every message spawns a new cloud agent, with no way to follow up on an existing one

Where does the bug appear (feature/product)?

Cloud Agent (GitHub, Slack, Web, Linear)

Describe the Bug

When using Cursor through Slack’s AI Assistant pane (the app’s side panel, not a channel), every message I send inside a single assistant thread spawns a brand-new cloud agent instead of following up on the existing one.

A single assistant thread with three of my messages produced three separate cloud agents, each auto-titled after the message that created it. All three appear as distinct agents in cursor.com/agents.

Answer quality is not the problem. Each new agent reads the Slack thread, so context carries over. The problems are:

  1. There is no way to follow up on a specific agent from this surface. The documented mechanisms don’t exist here. @Cursor <prompt> is inert, because in an assistant thread Cursor is already the responder and the mention is just plain text. And the ⋯ overflow menu on agent messages has no “Add follow-up” item. Cursor’s messages in this surface carry inline “Open in Web / Open in Desktop / Switch repository” buttons instead. The only way to continue a specific agent is to leave Slack via “Open in Web”.

  2. Agent sprawl and cost. Each throwaway question becomes a full cloud agent against my default repository.

I know I can use a private channel instead, where @Cursor follow-ups work as documented. But then I cannot take advantage of the Assistant pane, which is the reason I use Cursor from Slack in the first place: one click from the sidebar, with its own Chat and History tabs. The workaround costs me the surface I actually want.

The Slack docs at Slack | Cursor Docs describe @Cursor <prompt> as adding follow-up instructions to an existing agent, and describe an “Add follow-up” item in the ⋯ context menu. As far as I can tell both of these are channel-only behaviors. Nothing in the docs indicates the Assistant pane behaves differently, which is what made this confusing to diagnose.

Possibly the same root cause as the GitHub Issues variant in #158505, where staff described per-mention session creation as intended. If per-message agent creation is also intended for the Assistant pane, then the docs are misleading and the surface needs some way to target an existing agent.

Steps to Reproduce

  1. Install the Cursor app in a Slack workspace and open it from the Slack sidebar so the AI Assistant pane appears (the panel with Chat / History / About / Files & links tabs).
  2. In the Chat tab, send a message that starts an agent. Cursor creates an assistant thread, auto-titled from your message, and replies in it.
  3. Reply in that same thread with a follow-up question. Send it as plain text.
  4. Reply again, this time prefixing the message with @Cursor.
  5. Open cursor.com/agents.

Observed: three separate cloud agents, one per message sent, each auto-titled from that message. Step 4’s @Cursor prefix makes no difference. Hovering the ⋯ menu on any of Cursor’s replies shows no “Add follow-up” item.

Expected Behavior

Replies within a single assistant thread should attach as follow-ups to that thread’s existing cloud agent, rather than each spawning a new one, matching the follow-up behavior the docs describe for channels.

Failing that, the Assistant pane should expose some way to target an existing agent (an “Add follow-up” action, or an equivalent to @Cursor), and the docs should state plainly that the Assistant pane creates one agent per message.

Operating System

MacOS

Version Information

This is a server-side Slack integration behavior, not specific to a desktop build, but for completeness:

Version: 3.12.10
VS Code Extension API: 1.128.0
Commit: 24a12dbd9cabf48956ce5bb3dbd234e41385b3d0
Date: 2026-07-16T03:15:48.538Z
Layout: Agent Window
Build Type: Stable
Release Track: Default
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
OS: Darwin arm64 25.5.0

Observed 2026-07-16, around 10:47 to 11:01 PDT. Model on all three agents: Cursor Grok 4.5 High Fast.

For AI issues: which model did you use?

Cursor Grok 4.5 High Fast (all three agents)

Additional Information

Related reports, which look like adjacent surfaces rather than duplicates:

  • “Cursor Background Agent from Github creates new session on every follow up post” (#158505). Same per-message session creation, on GitHub Issues. Staff said it was intended there and suggested reclassifying as a feature request.
  • “Follow ups not allowed by teammates in slack when team follow-ups enabled” (#155430). A permission-check bug on the follow-up path. Not my case; I’m following up on my own agents and I’m not on a team.

Note that “Manage” on the Slack row of the dashboard Integrations page doesn’t open a management view. It jumps straight into a fresh Slack OAuth install flow. There appears to be no surface anywhere for inspecting the current Slack connection.

Does this stop you from using Cursor

No - Cursor works, but with this issue

You’re right: in the Slack AI Assistant pane, each message currently starts its own cloud agent, and the follow-up mechanisms the docs describe (@Cursor <prompt> and the “Add follow-up” overflow item) are channel-only today. They aren’t surfaced in the Assistant pane, so there’s no way to target an existing agent from there. It’s the same shape as the GitHub Issues behavior you referenced.

For now, the only way to continue one specific agent from the pane is Open in Web and following up there. Using a regular channel is the other option if you want in-thread follow-ups, though I know that costs you the pane you actually wanted.

I’ve reported this to our team - the missing follow-up path in the Assistant pane, the channel-centric docs, and the Slack “Manage” link reopening the install flow. I’ll update here as that progresses.