Skip to content

fix(adk): release streaming buffer repair in sdk#449

Merged
max-parke-scale merged 1 commit into
nextfrom
jr/repair-agentex-sdk-0-15-1
Jun 29, 2026
Merged

fix(adk): release streaming buffer repair in sdk#449
max-parke-scale merged 1 commit into
nextfrom
jr/repair-agentex-sdk-0-15-1

Conversation

@jromualdez-scale

@jromualdez-scale jromualdez-scale commented Jun 29, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Adds an ADK-scoped change so release-please cuts a new agentex-sdk patch release.
  • Documents that the SDK wheel is assembled from src/agentex/lib/**, which is where the already-merged streaming buffer fix lives.
  • Ships the StreamTaskMessageFull buffer fix from fix(streaming): StreamTaskMessageFull closes the coalescing buffer #426, which was merged to main in source but mis-routed into the agentex-client 0.16.1 release instead of a new SDK release.

Test plan

  • Not run; documentation-only release repair.

Notes

This is intended to repair the failed agentex-client v0.16.1 release state: the client published, but the SDK did not bump even though its wheel contents changed. A manual workflow dispatch alone cannot repair this because the repo still says agentex-sdk==0.15.0; publishing again would rebuild the changed 0.15.0 wheel and collide with the immutable PyPI artifact.

Document the SDK wheel source boundary so release-please attributes an ADK patch release that includes the already-merged streaming buffer fix.

Co-authored-by: Cursor <cursoragent@cursor.com>
@max-parke-scale

Copy link
Copy Markdown
Contributor

LGTM.

@max-parke-scale max-parke-scale merged commit 20795cb into next Jun 29, 2026
50 checks passed
@max-parke-scale max-parke-scale deleted the jr/repair-agentex-sdk-0-15-1 branch June 29, 2026 18:10
@stainless-app stainless-app Bot mentioned this pull request Jun 29, 2026
max-parke-scale added a commit that referenced this pull request Jun 29, 2026
…l event

test_send_event_and_stream_with_reasoning broke out of the stream loop on
the first `done` event. A single turn emits several messages — user echo,
reasoning, agent text — each ending in a `full` or `done`. When a reasoning
message's terminal event arrived before the agent's text `done`, the loop
exited with agent_response_found still False, failing at the assertion
("Agent response not found in stream") rather than timing out.

The failure signature confirms this: an AssertionError (loop broke early),
not a TimeoutError (latency). Consume terminal events until both the user
echo and the agent's text reply are seen, keying off message content rather
than the first terminal signal.

Test-side only: the producer (streaming.py) correctly emits a terminal
event per message and is keyed by message id by real consumers; it was
just repaired in #449, so the fix stays in the test.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants