Windows: Claude Code process crashes repeatedly while background Workflow subagents are running (v2.1.63)

Status Open
Maintainer reply None cached
Activity 1 comment · opened Jul 22, 2026

Environment

  • Claude Code v2.1.63 (desktop app session on Windows)
  • Windows 11 Pro 10.0.26200
  • Model: claude-fable-5
  • MCP servers connected (Notion connector, browser, session tools)

Symptom

The Claude Code process crashed/exited three times in one day, each time while a background Workflow was running with multiple parallel subagents doing heavy WebSearch/WebFetch work (4-8 research agents per workflow, up to ~16 concurrent with a verify phase).

On restart, the session receives:

<task-notification> status: stopped
No completion record was found for background workflow "..." from the previous session.
It may have been stopped ... or it may have been running when the previous Claude Code process exited.

MCP servers report disconnect/reconnect around each crash.

Timeline (local time, same day)

  1. Workflow with 8 parallel web-research agents — process died ~8 min in (~17:04). Resume relaunched agents; process died again ~10 min in (~17:17).
  2. Second workflow (4 parallel web-research agents) — process died ~7 min in (~18:45).

In the first crash, 4 of 8 agents had completed and their results were recoverable from journal.jsonl; in the later crashes no agent had finished, so all results had to be mined from the partial agent-*.jsonl transcripts.

Pattern / suspected repro

Launch the Workflow tool with several parallel agent() calls that each perform many WebFetch/WebSearch calls, let it run in the background for 5-10 minutes while the main conversation is idle or interacting. The host process exits (no visible error in the chat UI; the session resumes fresh afterwards).

Impact

  • Background workflows are killed with the process and report "stopped" with no completion record.
  • Partial token spend is lost unless manually recovered from the workflow journal/transcripts on disk.

Happy to provide the workflow scripts, journal files, or any local logs if you tell me where the crash logs live on Windows.

Filed via Claude Code on behalf of the user after the third occurrence.

View original on GitHub ↗

This issue has 1 comment on GitHub. Read the full discussion on GitHub ↗