Issue Station
Sign In
Tools hereToolbox

claude-code - 💡(How to fix) Fix [BUG] Single web search request spawned 280 parallel subagent sessions, consuming entire daily limit + $51 extra usage [1 comments, 2 participants]

Official PRs (…)
ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

Helpful · Quick feedback

Loading…
GitHub stats
anthropics/claude-code#49275•Fetched 2026-04-17 08:45:53
View on GitHub
Comments
1
Participants
2
Timeline
6
Reactions
0
Author
ALXFRUALXFRU
Participants
ALXFRUALXFRUgithub-actions[bot]github-actions[bot]
Timeline (top)
labeled ×4commented ×1cross-referenced ×1

Root Cause

  1. This is a clear runaway agent bug. Please refund the $51.20 extra usage charge caused by this incident
  2. Consider implementing maxParallelAgents (ref: #15487) or a circuit breaker for subagent spawning
RAW_BUFFERClick to expand / collapse

Environment

  • Claude Code version: 2.1.110
  • OS: macOS Darwin 25.3.0
  • Model: claude-opus-4-6 (1M context)
  • Plan: Max with $50 extra usage allowance

What happened

I asked Claude Code a single question: to research the practical risks of P2P cryptocurrency trading in Russia. A normal research task that should spawn 3-8 subagent sessions.

Instead, Claude Code spawned 280 parallel subagent sessions between 14:46 and 15:44 (under 1 hour). Each session performed nearly identical web searches with minor wording variations (e.g., "Тинькофф блокировка P2P крипто 2025", "Tinkoff bank P2P crypto block Russia 2025", "115-ФЗ криптовалюта блокировка" repeated hundreds of times).

The system became unresponsive. I could not stop the runaway agents.

Evidence

Session log analysis from ~/.claude/projects/:

  • 280 sessions created in the 14:46-15:52 window
  • 11.1 MB of session log data generated
  • Each session is a separate .jsonl file with a unique UUID
  • First user messages are all auto-generated subagent prompts like "Ты исследователь. Используй WebSearch чтобы найти...", "Search the web for:...", "Search the web thoroughly for:..."
  • The cascade kept spawning new sessions even after 100+ were already running

Billing impact

  • /usage shows: 95% of usage was while 4+ sessions ran in parallel
  • /usage shows: 51% of usage came from subagent-heavy sessions
  • Daily limit: 100% consumed (from a single question)
  • $51.20 / $50.00 extra usage spent on this one incident
  • Weekly usage jumped to 43% in one day

Expected behavior

A web research request should spawn a bounded number of subagent sessions (3-8 max), collect results, synthesize, and return. Not 280.

Request

  1. This is a clear runaway agent bug. Please refund the $51.20 extra usage charge caused by this incident
  2. Consider implementing maxParallelAgents (ref: #15487) or a circuit breaker for subagent spawning

Related issues

  • #15487 (Add maxParallelAgents config)
  • #34449 (Stop button fails to halt runaway agent)
  • #13996 (Kill switch for runaway Task agents)
  • #34476 (No way to cancel spawned agent team)
  • #35741 (Interrupted firing across multiple parallel sessions)

extent analysis

TL;DR

Implementing a maxParallelAgents configuration or a circuit breaker for subagent spawning could prevent similar runaway agent incidents.

Guidance

  • Review the session log analysis from ~/.claude/projects/ to understand the pattern of subagent spawning and identify potential triggers for the runaway behavior.
  • Consider implementing a temporary workaround, such as manually limiting the number of parallel subagent sessions or adding a custom circuit breaker, until a permanent fix is available.
  • Investigate the related issues (#15487, #34449, #13996, #34476, #35741) to see if they are connected to this incident and if their resolution could prevent similar issues.
  • Monitor usage and billing impact closely to detect and respond to potential future incidents.

Example

No code snippet is provided as the issue does not contain sufficient technical details to create a specific example.

Notes

The provided information suggests a potential issue with the subagent spawning mechanism, but without further technical details, it is difficult to provide a definitive solution. Implementing a maxParallelAgents configuration or a circuit breaker could help prevent similar incidents, but its effectiveness depends on the underlying cause of the issue.

Recommendation

Apply a workaround, such as manually limiting parallel subagent sessions, until a permanent fix is available, as it can help prevent similar incidents and reduce potential billing impact.

Vote matrix · Quick signals

Works
Did the solution work? Tap to confirm.
Easy Fix
Was it a quick fix?
Time Saver
Did it save you time?
Blocking
Was it severely blocking?
Common Issue
Are others likely hitting this too?
Flaky / Intermittent
Is it intermittent?
Verified / Reproducible
Can you reproduce it reliably?
Loading…

FAQ

Expected behavior

A web research request should spawn a bounded number of subagent sessions (3-8 max), collect results, synthesize, and return. Not 280.

#training loop#device allocation#model download#tokenizer error#prompt formatting

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Back to top recommendations

TRENDING

claude-code - 💡(How to fix) Fix [BUG] Single web search request spawned 280 parallel subagent sessions, consuming entire daily limit + $51 extra usage [1 comments, 2 participants]