Slack: chat.startStream fails with team_not_found on Enterprise Grid workspaces

[patrick-chinchill]

Symptom

chat.startStream fails with team_not_found on Enterprise Grid workspaces, even with a non-empty thread_ts. Observed in chinchill-api production (ray-dev, deploy 594817568):

slack_sdk.errors.SlackApiError: (url: https://slack.com/api/chat.startStream, status: 200)
The server responded with: {'ok': False, 'error': 'team_not_found'}

Traceback reaches SlackAdapter.stream() → super().stream() → streamer.append() → chat.startStream.

Workspace context

• team_id: T0917KZ3BTK
• stored_enterprise_id: E0917L87ZCH
• is_enterprise_install: False
• Non-streaming chat.postMessage succeeds on the same workspace
• Only chat.startStream returns team_not_found

Likely root cause

chat.startStream requires an explicit team_id for Grid orgs (the per-workspace token alone isn't sufficient), but the adapter's AsyncWebClient (SlackAdapter._get_client()) is created without one. Non-streaming Web API calls succeed because they don't enforce the same constraint.

Relationship to existing PRs

• Distinct from fix(adapters): bug-fix sweep from upstream 4.27.0 (slack/discord/telegram) #89's empty-thread_ts fix (commit 916e532 — that handles DM-root threads via post_message fallback; this path skips that branch because decoded.thread_ts is truthy).
• Potentially uses the recipient_team_id plumbing landed in fix(adapters): Slack streaming team_id + Teams DM Graph conversation IDs (vercel/chat#330, #403) #85 ([chat] fix Slack streaming team ID for interactive payloads vercel/chat#330 port — the interactive-payload team.id extraction). Worth checking whether that value is plumbed all the way through to streamer.append() / chat.startStream.

Suggested fix path (pending verification)

1. Thread team_id from the streaming context through to the AsyncWebClient call, either:
   • Via the existing recipient_team_id (preferred — already extracted on the inbound path), or
   • Via an explicit team_id parameter on the chat.startStream call
2. Add a regression test mocking an Enterprise Grid workspace where chat.postMessage succeeds without team_id but chat.startStream requires it.

Workaround in flight

chinchill-api is shipping an interim error-handling fix so a failed final-post on an otherwise-successful streaming run no longer surfaces a false "something went wrong" to the end user. The SDK-side fix is still needed for streaming to actually work on Grid workspaces.

Follow-up

@tony-chinchill-ai offered to attach the full traceback in a follow-up comment. Filed by report-on-PR #89.

[tony-chinchill-ai]

Full production traceback, as requested. Source: chinchill-api on ray-dev, deploy 594817568b99de748aa17bc9a1d9891fe688ae57, 2026-05-16 04:05:03 UTC. Path prefixes collapsed to .../ for readability (the deploy SHA above is the working-dir root); frame file:line:function are verbatim.

Workspace context (Enterprise Grid): the inbound event resolved team_id=T0917KZ3BTK; a later team_id-only installation fallback in the same session logged stored_enterprise_id=E0917L87ZCH, is_enterprise_install=False, bot_user_id=U0ASZ9VK29Z. Note the cache-hit lookup line reports enterprise_id=None even though the stored installation carries the enterprise id:

2026-05-16 04:05:02,101 - chinchill-api - DEBUG - Cache hit for installation lookup: team_id=T0917KZ3BTK, enterprise_id=None
2026-05-16 04:05:02,101 - chinchill-api - DEBUG - Found Slack bot credentials for team_id=T0917KZ3BTK
...
2026-05-16 04:07:17,143 - chinchill-api - INFO - Found installation via team_id-only fallback: team_id=T0917KZ3BTK, stored_enterprise_id=E0917L87ZCH
2026-05-16 04:07:17,143 - chinchill-api - INFO - Found valid installation for team_id=T0917KZ3BTK, enterprise_id=None, user_id=None, is_enterprise_install=False, bot_user_id=U0ASZ9VK29Z

Primary traceback — chat.startStream → team_not_found:

2026-05-16 04:05:03,070 - app.integrations.chat_sdk.handlers - ERROR - Chat SDK: Slash command streaming failed
Traceback (most recent call last):
  File ".../app/integrations/chat_sdk/handlers.py", line 1444, in _run_agent_streaming_simple
    sent_message = await _post_stream_if_non_empty(
  File ".../app/integrations/chat_sdk/handlers.py", line 420, in _post_stream_if_non_empty
    return await thread.post(_replay())
  File ".../chat_sdk/thread.py", line 523, in post
    return await self._handle_stream(message)
  File ".../chat_sdk/thread.py", line 646, in _handle_stream
    raw_result = await self.adapter.stream(self._id, _wrapped_stream(), options)
  File ".../app/integrations/chat_sdk/__init__.py", line 247, in stream
    return await super().stream(thread_id, text_stream, options)
  File ".../chat_sdk/adapters/slack/adapter.py", line 2386, in stream
    await push_text_and_flush(chunk)
  File ".../chat_sdk/adapters/slack/adapter.py", line 2381, in push_text_and_flush
    await flush_markdown_delta(delta)
  File ".../chat_sdk/adapters/slack/adapter.py", line 2339, in flush_markdown_delta
    await streamer.append(markdown_text=delta)
  File ".../slack_sdk/web/async_chat_stream.py", line 116, in append
    return await self._flush_buffer(**kwargs)
  File ".../slack_sdk/web/async_chat_stream.py", line 195, in _flush_buffer
    response = await self._client.chat_startStream(
  File ".../slack_sdk/web/async_client.py", line 2902, in chat_startStream
    return await self.api_call("chat.startStream", json=kwargs)
  File ".../slack_sdk/web/async_base_client.py", line 175, in api_call
    return await self._send(
  File ".../slack_sdk/web/async_base_client.py", line 215, in _send
    return AsyncSlackResponse(**{**data, **res}).validate()
  File ".../slack_sdk/web/async_slack_response.py", line 203, in validate
    raise e.SlackApiError(message=msg, response=self)
slack_sdk.errors.SlackApiError: The request to the Slack API failed. (url: https://slack.com/api/chat.startStream, status: 200)
The server responded with: {'ok': False, 'error': 'team_not_found'}

Note frame app/integrations/chat_sdk/__init__.py:247 is return await super().stream(...) inside the if decoded.thread_ts: branch — i.e. thread_ts was non-empty, so chinchill-api's empty-thread_ts fallback was correctly skipped and the native chat_stream path was taken. This is the non-empty-thread_ts Grid case that the #94/#89 empty-thread_ts fix does not address.

Secondary traceback — downstream consequence (the broken stream surfaces as anyio.BrokenResourceError; repeated pydantic-ai capability-wrapper frames collapsed):

2026-05-16 04:05:03,275 - asyncio - ERROR - Task exception was never retrieved
future: <Task finished name='Task-634' coro=<AbstractAgent._run_stream_events.<locals>.run_agent() done, defined at .../pydantic_ai/agent/abstract.py:1166> exception=BrokenResourceError()>
Traceback (most recent call last):
  File ".../pydantic_ai/agent/abstract.py", line 1168, in run_agent
    return await self.run(
  File ".../pydantic_ai/agent/abstract.py", line 395, in run
    node = await agent_run._run_node_with_hooks(node, _stream_step)
  File ".../pydantic_ai/run.py", line 278, in _wrap_and_advance
    result = await cap.on_node_run_error(run_context, node=node, error=e)
  ... (repeated pydantic-ai combined/abstract capability wrapper frames) ...
  File ".../pydantic_ai/agent/abstract.py", line 1164, in event_stream_handler
    await send_stream.send(event)
  File ".../anyio/streams/memory.py", line 256, in send
    raise BrokenResourceError from None
anyio.BrokenResourceError

Non-streaming chat.postMessage succeeds on this same workspace (the bot posts other messages fine in the same session) — only chat.startStream returns team_not_found, which lines up with the _get_client() / AsyncWebClient-has-no-team_id hypothesis and the #85 inbound recipient_team_id plumbing. Happy to pull more log lines if a wider window helps.

[patrick-chinchill]

Fixed in 0.4.31.1 (PR #176). stream() now threads a workspace team_id (= the already-plumbed recipient_team_id, the per-workspace T… id) into chat.startStream, which Enterprise Grid orgs require. Verified end-to-end against slack_sdk (3.42) that the value reaches the chat.startStream API call; append/stop correctly don't need it. Live-tested on a Grid workspace (E0917L87ZCH): chat.startStream → 200, no team_not_found, no regression. Caveat: that workspace doesn't reproduce the original failure (the production report was workspace T0917KZ3BTK), so the flip-from-team_not_found couldn't be reproduced here — @tony-chinchill-ai, please confirm streaming now works on the original Grid workspace; reopen if it still fails.