release: 0.9.10

[stainless-app]

Automated Release PR

0.9.10 (2026-04-07)

Full Changelog: v0.9.9...v0.9.10

Features

• adk: Revamp run_claude_agent_activity to use more streaming (#309) (25069d3)

---

This pull request is managed by Stainless's GitHub App.

The semver version number is based on included commit messages. Alternatively, you can manually set the version number in the title of this pull request.

For a better experience, it is recommended to use either rebase-merge or squash-merge when merging this pull request.

🔗 Stainless website
📚 Read the docs
🙋 Reach out for help or questions

Greptile Summary

This release bumps the version to 0.9.10 and revamps run_claude_agent_activity to use streaming for all content block types (text, thinking, tool use), with tool results handled via a new TemporalStreamingHooks class with pre/post hook callbacks. The cost_info None guard (previously flagged) is correctly implemented on line 382.

• The success path after the message loop does not clean up any remaining subagent_spans (unlike the error path which does) — if a Task tool starts a span but its PostToolUse hook never fires, those spans will be abandoned on a clean exit.

Confidence Score: 4/5

Safe to merge; one minor cleanup gap on the success path

All findings are P2. The subagent span leak on success depends on an edge-case SDK behavior (PostToolUse not firing after a Task tool call) and won't cause data loss or incorrect results, but the success path should mirror the error path's cleanup for robustness.

src/agentex/lib/core/temporal/plugins/claude_agents/activities.py — success-path cleanup around line 397

Important Files Changed

Filename	Overview
src/agentex/lib/core/temporal/plugins/claude_agents/activities.py	Revamped streaming activity handling TextBlock/ThinkingBlock/ToolUseBlock; subagent spans not cleaned up on normal success path
src/agentex/lib/core/temporal/plugins/claude_agents/hooks/hooks.py	New TemporalStreamingHooks class with pre/post tool use hooks for streaming tool events; logic is correct
tests/lib/test_claude_agents_activities.py	Comprehensive tests for activity helpers and message handling; all key scenarios covered
tests/lib/test_claude_agents_hooks.py	Good coverage of all three hook methods and the factory function
src/agentex/_version.py	Version bump to 0.9.10
pyproject.toml	Version bump to 0.9.10

Sequence Diagram

sequenceDiagram
    participant W as Temporal Workflow
    participant A as run_claude_agent_activity
    participant SDK as ClaudeSDKClient
    participant H as TemporalStreamingHooks
    participant UI as AgentEx UI

    W->>A: execute_activity(prompt, workspace, tools)
    A->>SDK: async with ClaudeSDKClient(options)
    A->>SDK: client.query(prompt)
    loop receive_response()
        SDK-->>A: AssistantMessage(TextBlock)
        A->>UI: stream_text_delta()
        SDK-->>A: AssistantMessage(ThinkingBlock)
        A->>UI: stream_reasoning()
        SDK-->>A: AssistantMessage(ToolUseBlock)
        A->>UI: stream_tool_request()
        Note over A: opens subagent span if Task tool
        SDK-->>H: PostToolUse hook
        H->>UI: stream tool response
        Note over H: closes subagent span
        SDK-->>A: SystemMessage(init)
        A->>A: capture session_id
        SDK-->>A: ResultMessage
        A->>A: capture cost/usage
    end
    A->>A: close_text_stream()
    Note over A: subagent_spans NOT cleaned up here
    A-->>W: {messages, session_id, usage, cost_usd}

Loading

Prompt To Fix All With AI

This is a comment left during a code review.
Path: src/agentex/lib/core/temporal/plugins/claude_agents/activities.py
Line: 396-398

Comment:
**Orphaned subagent spans not cleaned up on success path**

The `subagent_spans` dict is fully cleaned up in the `except` block (lines 412–417), but the normal success path calls `close_text_stream()` and returns without touching it. If a `Task` tool use opens a span in `stream_tool_request` and the SDK completes the session without firing `PostToolUse` (e.g., a partial response or SDK-level soft termination that doesn't raise a Python exception), those span context managers are abandoned. The success path should mirror the error path:

```suggestion
        logger.debug("Message loop completed, cleaning up...")
        await close_text_stream()
        for _ctx, _span in list(subagent_spans.values()):
            try:
                await _ctx.__aexit__(None, None, None)
            except Exception:
                pass
        subagent_spans.clear()
```

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (3): Last reviewed commit: "release: 0.9.10" | Re-trigger Greptile}

[stainless-app]

Release version edited manually

The Pull Request version has been manually set to 0.9.10 and will be used for the release.

If you instead want to use the version number 0.10.0 generated from conventional commits, just remove the label autorelease: custom version from this Pull Request.

[stainless-app]

🤖 Release is at https://github.com/scaleapi/scale-agentex-python/releases/tag/v0.9.10 🌻