Request: Configurable timeout for long-running tasks

Problem Statement

We are experiencing systematic crashes at exactly 10 minutes 12 seconds (612 seconds) during accessibility analysis and other long-running tasks in Claude Code.

Hypothesis (NOT confirmed)

We believe the 612-second timeout may be hardcoded in Claude Code's binary as a resource protection mechanism (600s task timeout + 12s grace period). However, we have not verified this - it's our working hypothesis based on:

1. Exact duration consistency: All crashes occur at precisely 10m 12s
2. Configuration search: No user-accessible timeout configuration found in:

• Environment variables (TIMEOUT, AGENT_TIMEOUT, etc.)
• Config files (~/.claude/*)
• systemd/Docker settings
• OS limits (ulimit)

1. User experience: Hours-long sessions work fine → timeout appears task-specific, not global

Request

Could you clarify:

1. Is there a configurable timeout mechanism we've missed?
2. If hardcoded, is there a technical reason it cannot be made configurable?
3. Are there alternative approaches we should consider?

Our Planned Workaround (F03 Task Splitting)

While awaiting clarification, we plan to implement automatic task splitting:

• Break large tasks into ≤5-minute subtasks
• Checkpoint state every 30 minutes
• Parallel worker execution where possible

This mitigates crashes but doesn't address the root cause if our hypothesis is incorrect.

Evidence

• 15 crashes documented (13 at 612s, 2 at shorter durations during nested execution)
• Pattern: Accessibility tasks, strategic planning, multi-platform workflow design
• Crash logs available if helpful for diagnosis

Environment

• Claude Code version: 2.0.70 (Claude Code)
• OS: Linux 6.12.61-1-lts
• Session type: CLI (not VSCode extension)

Thank you for any guidance you can provide\!