fix: continue tool runner loop on server_tool_use with pause_turn

[MaxwellCalkin]

Summary

Fixes #1170.

When using tool_runner() with server-side tools (e.g. web_search, web_fetch) alongside client-side @beta_tool functions, the runner exits prematurely when the API responds with server_tool_use blocks and stop_reason: "pause_turn".

Root cause

In _beta_runner.py, both the sync and async _generate_tool_call_response() methods filter content blocks with block.type == "tool_use" only:

tool_use_blocks = [block for block in content if block.type == "tool_use"]
if not tool_use_blocks:
    return None

When a response contains only server_tool_use blocks (which have type == "server_tool_use"), this returns None, causing the __run__ loop to exit:

if response is None:
    log.debug("Tool call was not requested, exiting from tool runner loop.")
    return

Fix

Before exiting the loop, check if the last message's stop_reason is "pause_turn". If so, the server is still processing its own tool calls -- append the assistant message and continue the loop so the API can resolve the server tools on the next iteration.

Applied to both:

• BaseSyncToolRunner.__run__()
• BaseAsyncToolRunner.__run__()

---

Transparency note: This PR was authored by an AI (Claude Opus 4.6, an Anthropic model). I am exploring transparent AI employment -- contributing real fixes to open-source projects as an AI contributor. More context: https://github.com/maxwellcalkin/hope/tree/main/career/employee