Slack DM streaming replies are dropped — stream() raises for empty thread_ts but _handle_stream never falls back

[tony-chinchill-ai]

Summary

Native Slack streaming silently drops the reply for top-level DMs. The agent runs, produces text, and nothing reaches the user. Channel @-mentions and threaded DM replies are unaffected.

Root cause

1. _handle_message_event (src/chat_sdk/adapters/slack/adapter.py) encodes a top-level DM thread id with an empty thread_ts — intentional, so open_dm subscriptions match incoming DMs.
2. Draft PR fix(adapters): bug-fix sweep from upstream 4.27.0 (slack/discord/telegram) #89 makes SlackAdapter.stream() raise ValidationError for an empty thread_ts instead of passing it to chat.startStream, which rejects it.
3. Thread._handle_stream (src/chat_sdk/thread.py) calls adapter.stream() with no try/except. Its _fallback_stream post+edit path is reached only when the adapter has no stream method at all — not when stream() raises.

Net effect: after PR #89 a DM streaming reply fails clean instead of failing confusingly, but it is still never delivered.

Proposed fix

For an empty thread_ts, SlackAdapter.stream() should accumulate the stream and post_message once instead of raising. DMs degrade to a single non-streamed message; channels keep native streaming. This can replace the raise introduced in draft PR #89:

decoded = self.decode_thread_id(thread_id)
thread_ts = decoded.thread_ts or None
if not thread_ts:
    accumulated = ""
    async for chunk in text_stream:
        if isinstance(chunk, str):
            accumulated += chunk
        elif isinstance(chunk, dict) and chunk.get("type") in ("text-delta", "markdown_text"):
            accumulated += chunk.get("text", "")
        elif getattr(chunk, "type", None) == "markdown_text":
            accumulated += getattr(chunk, "text", "")
    return await self.post_message(thread_id, PostableMarkdown(markdown=accumulated))

post_message already normalizes thread_ts or None, so an empty DM thread_ts posts as a normal message (Slack accepts that for chat.postMessage; only chat.startStream requires one).

Note

Production-tested: chinchill-api currently runs exactly this as a MultiTenantSlackAdapter subclass override. Folding it into the SDK lets that override be deleted.

[tony-chinchill-ai]

@patrick-chinchill flagging this for the chat-sdk-python side. This is the Slack DM streaming gap that surfaced on dev-tony — top-level DM replies are dropped after the empty-thread_ts guard. chinchill-api is running the fix as a MultiTenantSlackAdapter subclass override for now; the proposed stream() change above can fold into draft PR #89 so that override can be retired.

[patrick-chinchill]

Folded into PR #89 as commit 916e532. The vercel/chat#292 empty-thread_ts port now accumulates the stream and calls post_message once for empty thread_ts; non-empty (channel) path unchanged. Regression tests in tests/test_slack_api.py::TestEmptyThreadTsGuards.

Snippet was adapted from this issue's body to match the in-tree types — text_stream carries MarkdownTextChunk dataclasses rather than dicts, so the dict branch collapsed to a hasattr(chunk, "type") check matching the google_chat/adapter.py stream() pattern. PostableMarkdown(markdown=...) is the right wrapper since post_message accepts the postable union and already normalizes thread_ts or None.

Will auto-close when #89 merges. Once it does, the MultiTenantSlackAdapter.stream() override in chinchill-api can come out.

---

Generated by Claude Code