Stop Coming Back to a Stalled Background Session: /goal Now Checks In After 30 Minutes

TL;DR: Claude Code's /goal command now checks in on background tasks after 30 minutes of waiting — you no longer come back to a session frozen at hour three.

Long /goal sessions used to be a gamble: if a background task stalled, the session waited without telling you. Week 34 of Claude Code (shipped August 21) adds automatic check-ins.

What changed: When you launch a session with /goal "ship the auth PR" and background tasks (--bg, /fork) are keeping it busy, Claude now checks in after those tasks have been running for 30+ minutes. The check-in reports what's done, what's blocked, and whether it needs your call before continuing. If you stay away, it checks in again at longer intervals rather than silently stalling.

How to use it:

/goal "Refactor the payments module, run tests, fix failures, open PR when green"

Step away. After 30+ minutes of background task activity, you get a push notification (Remote Control / Desktop app):

Goal check-in: 3/5 files refactored, 2 test failures blocking the PR.
API naming collision between old and new — need your call.

Two configuration knobs if the defaults don't fit:

# Disable check-ins entirely (shell env)
export CLAUDE_CODE_GOAL_CHECKIN_MINUTES=0

# Or per-repo in .claude/settings.json
{"goalCheckinMinutes": 60}

Also new in the same release: /goal now clears itself with a notice when a turn dies on an unrecoverable error (revoked auth, exhausted credit balance, context overflow) instead of staying armed and silently blocking future goals.

Where it doesn't help:

• API token quotas don't trigger check-ins — only claude.ai time-based usage limits support auto-continuation. A session paused at an API quota just waits for you.
• Tasks under 30 minutes never trigger a check-in. Short runs complete before the threshold.
• Check-ins arrive via push notification in Remote Control or the Desktop app. Terminal-only sessions show the check-in when you reattach — you need to be watching.

Why it matters: Before this change, a six-hour agentic task that stalled at hour two looked identical to one that finished cleanly — you only discovered the difference on return. Check-ins give you an intervention window at a reasonable cadence without polling the session yourself.

Sources: Claude Code Week 34 — code.claude.com/docs/en/whats-new/2026-w34 · /goal, /loop, and long-horizon agents — Pinggy Blog · Background Tasks guide — Developers Digest