[BUG] SessionEnd hook creates infinite recursive loop when spawning a claude CLI subprocess

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?

Severity: Critical (can silently drain API credits to zero)

Product: Claude Code — Hooks system (settings.json)

---
Description

The SessionEnd hook in Claude Code does not guard against recursive self-triggering. When the hook spawns a claude CLI subprocess (e.g., to run a logging agent), that
subprocess is itself a Claude Code session. When it terminates, it fires SessionEnd again — spawning another agent, which fires another SessionEnd, ad infinitum.

The loop only terminates when the API credit balance reaches zero.

---
Steps to Reproduce

1. Add the following to .claude/settings.json:

{
"hooks": {
"SessionEnd": [
{
"hooks": [
{
"type": "command",
"command": "claude --print 'Log this session.' --allowedTools 'Write'"
}
]
}
]
}
}

1. Close any Claude Code session.
2. Observe: the hook fires, spawns a claude subprocess, which ends and fires SessionEnd again. The loop runs until credits are exhausted.

---
Observed Behavior

• 74 agent invocations in ~4 hours on June 4, 2026
• ~50,000,000 input tokens consumed (vs. ~82,000,000 for the entire previous month)
• Zero productive output
• Stopped only when API balance hit zero ("Credit balance is too low" in logs)

---
Expected Behavior

Either:

• (A) SessionEnd should not fire for subprocesses spawned by a SessionEnd hook itself, or
• (B) Claude Code should detect recursive SessionEnd triggering and break the loop after N iterations (e.g., 3), or
• (C) The documentation should explicitly warn: "Do not spawn a claude CLI subprocess inside a SessionEnd hook — this will cause infinite recursion."

---
Hook Configuration That Triggered the Bug

"SessionEnd": [
{
"hooks": [
{
"type": "command",
"command": "claude --agent session-logger --model claude-haiku-4-5 --print 'Log session.' --allowedTools 'Read,Glob,Write,Bash'",
"timeout": 130
}
]
}
]

---
Environment

• Platform: Linux VPS (Ubuntu), Claude Code CLI
• Shell: bash
• Hook type: SessionEnd
• Subprocess: claude CLI (claude-haiku-4-5)

---
Suggested Fix

Inject an environment variable (e.g., CLAUDE_HOOK_DEPTH=1) when Claude Code runs a hook command. If that variable is already set when a new session starts, suppress
SessionEnd hook execution for that session. This would cost zero user-visible behavior change for legitimate use cases.

What Should Happen?

Expected Behavior

Either:

• (A) SessionEnd should not fire for subprocesses spawned by a SessionEnd hook itself, or
• (B) Claude Code should detect recursive SessionEnd triggering and break the loop after N iterations (e.g., 3), or
• (C) The documentation should explicitly warn: "Do not spawn a claude CLI subprocess inside a SessionEnd hook — this will cause infinite recursion."

Error Messages/Logs

Steps to Reproduce

The SessionEnd hook in Claude Code does not guard against recursive self-triggering. When the hook spawns a claude CLI subprocess (e.g., to run a logging agent), that
subprocess is itself a Claude Code session. When it terminates, it fires SessionEnd again — spawning another agent, which fires another SessionEnd, ad infinitum.

The loop only terminates when the API credit balance reaches zero.

---
Steps to Reproduce

1. Add the following to .claude/settings.json:

{
"hooks": {
"SessionEnd": [
{
"hooks": [
{
"type": "command",
"command": "claude --print 'Log this session.' --allowedTools 'Write'"
}
]
}
]
}
}

1. Close any Claude Code session.
2. Observe: the hook fires, spawns a claude subprocess, which ends and fires SessionEnd again. The loop runs until credits are exhausted.

---
Observed Behavior

• 74 agent invocations in ~4 hours on June 4, 2026
• ~50,000,000 input tokens consumed (vs. ~82,000,000 for the entire previous month)
• Zero productive output
• Stopped only when API balance hit zero ("Credit balance is too low" in logs)

Claude Model

None

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v2.1.161

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

VS Code integrated terminal

Additional Information

_No response_