[BUG] Workflow subagent hangs indefinitely with no timeout; run reports "running" for 5+ hours

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

What's wrong: A 9-agent workflow launched at 17:09. Eight agents completed within six minutes. One verification agent produced no output after 17:15 and never errored, never timed out, and never advanced. The run continued reporting "running" for 5 hours 14 minutes until I killed it manually. Because the synthesis step was downstream of that agent, the run produced nothing — the six completed research tracks were only recoverable by parsing the run journal by hand afterward.

What Should Happen?

workflow-hang-timeline.md

Error Messages/Logs

{"type": "started", "agentId": "ad95a877efa97fd6d", "key": "v2:b5a8f26f92ebc721e0ef6eceacb543595b6ae2c29f93a15c98677e67a652c31b"}
{"type": "started", "agentId": "a0d34aebc89265d8b", "key": "v2:9034aa67877c07fa514ae5bce057d8f2cf00c78278ec6e886ac9ebd818eeba8c"}
{"type": "started", "agentId": "a40a8ff8468cd54e2", "key": "v2:940b601a6762f1da259aaa0c04a3cf61757b857b1f3a9a895a02b73736969540"}
{"type": "started", "agentId": "a4403923c163f69fe", "key": "v2:7de64ae0641fc26957266ecaaa68969acbdc871b9950b70ecd3ca3ba655c3135"}
{"type": "started", "agentId": "adb1bb6bda49544a5", "key": "v2:e8f3986e555b4e683666ca309f3a48177b80be8def4f0ed37303683ef194ee93"}
{"type": "started", "agentId": "a9486df391b218f12", "key": "v2:8590dec3ba3e4f111a4568c588acf35987c1397c2fa1de9392f59f7cfd16c355"}
{"type": "result", "agentId": "adb1bb6bda49544a5", "key": "v2:e8f3986e555b4e683666ca309f3a48177b80be8def4f0ed37303683ef194ee93", "result_shape": {"findings": 13, "summary": 1252, "recommendation": 949, "gaps": 5}}
{"type": "result", "agentId": "a40a8ff8468cd54e2", "key": "v2:940b601a6762f1da259aaa0c04a3cf61757b857b1f3a9a895a02b73736969540", "result_shape": {"findings": 7, "summary": 598, "recommendation": 1663, "gaps": 5}}
{"type": "result", "agentId": "a0d34aebc89265d8b", "key": "v2:9034aa67877c07fa514ae5bce057d8f2cf00c78278ec6e886ac9ebd818eeba8c", "result_shape": {"summary": 1084, "recommendation": 604, "findings": 5, "gaps": 4}}
{"type": "result", "agentId": "a9486df391b218f12", "key": "v2:8590dec3ba3e4f111a4568c588acf35987c1397c2fa1de9392f59f7cfd16c355", "result_shape": {"findings": 11, "recommendation": 1767, "summary": 1493, "gaps": 4}}
{"type": "result", "agentId": "a4403923c163f69fe", "key": "v2:7de64ae0641fc26957266ecaaa68969acbdc871b9950b70ecd3ca3ba655c3135", "result_shape": {"summary": 1376, "findings": 8, "recommendation": 1430, "gaps": 4}}
{"type": "result", "agentId": "ad95a877efa97fd6d", "key": "v2:b5a8f26f92ebc721e0ef6eceacb543595b6ae2c29f93a15c98677e67a652c31b", "result_shape": {"findings": 9, "summary": 1274, "recommendation": 1261, "gaps": 5}}
{"type": "started", "agentId": "aad6b480dcc92c056", "key": "v2:c850d0b2b620530d2d3b87bb8b563cc7b5ae036eeffeea4c4a1fee32114e00e8"}
{"type": "started", "agentId": "ad94ba1b86c96dec9", "key": "v2:572b09adf4f820f5356286d8d57cd399753f5051924884ab97c92cd57523f451"}
{"type": "result", "agentId": "ad94ba1b86c96dec9", "key": "v2:572b09adf4f820f5356286d8d57cd399753f5051924884ab97c92cd57523f451", "result_shape": {"holds": "bool", "problems": 10, "corrected_recommendation": 3978}}

Steps to Reproduce

Steps to reproduce:

1. Claude Code 2.1.224 on macOS, model Claude Opus 5, inside any git repository.
2. Run the attached workflow script: Workflow({scriptPath: "repro-workflow.js"}).
3. Structure is 9 agents in three stages: six parallel research agents (Sonnet, medium effort, schema-constrained, using web search), a parallel() barrier, two parallel verification agents (default model, high effort, schema-constrained, also using web search), a second barrier, then one synthesis agent that writes a file.
4. Let it run and watch the workflow transcript directory.

Expected: all nine agents complete, or a failing agent errors and the stage fails.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

2.1.224

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Terminal.app (macOS)

Additional Information

_No response_