[BUG] Headless single-agent sessions die with SIGTERM after 3-10 minutes (Max subscription, no rate limits)

Environment

• Claude Code version: 2.1.63
• Platform: Max subscription (OAuth), NOT API key
• Operating System: macOS 25.2.0 (arm64) — Mac mini M4
• Use case: Node.js script spawning headless Claude Code for automation

Bug Description

Single headless agents die with SIGTERM (exit code 143) after 3-10 minutes of active work. This happens with:

• Only ONE agent running (not parallel)
• Max subscription (no billing rate limits)
• Agent actively working when killed (mid-web-fetch or mid-file-operation)

Command Used

claude -p "task prompt" \
  --output-format stream-json \
  --allowedTools "Bash,Read,Write,Edit,WebSearch,WebFetch" \
  --verbose 2>&1 | tee /tmp/agent.log

Observed Pattern

| Task Type | Result |
|-----------|--------|
| Short file tasks (<2 min) | ✅ Completes |
| Long file tasks (5-10 min) | ❌ Dies with SIGTERM |
| Tasks with WebSearch/WebFetch | ❌ Dies faster |

Expected Behavior

Headless agents should complete their tasks without being killed, or at minimum surface an error before termination.

Actual Behavior

• Process receives SIGTERM (exit code 143)
• No error message in output stream
• Agent was actively working when killed
• No obvious cause in logs

Related Issues

• #6594 describes shared AbortController killing all subagents on any error — but that's for parallel agents
• #28482 describes hanging (different symptom)

Hypothesis

Something is sending SIGTERM to long-running headless sessions. Possibilities:

1. Internal timeout/watchdog not documented
2. Shared AbortController firing on transient errors (even for single agents)
3. WebSearch/WebFetch tool errors triggering abort

Request

1. Is there a timeout for headless sessions?
2. Any flags to extend or disable it?
3. How can we keep long-running headless agents alive for automation use cases?