Skip to main content
Features Cart

Guides

Auto-Nudge and Objective Mode — stop Claude Code from quitting early

Neural-LLM Admin 2 min read
Auto-Nudge and Objective Mode — stop Claude Code from quitting early

Agents often stop for social reasons — “want me to continue?”, multi-choice offers, or a fake checkpoint — while the real task is unfinished. Power Claude Auto-Nudge classifies those trivial stops and continues; Objective Mode keeps the north-star goal sticky across nudges.

TL;DR

ModePurpose
Auto-NudgeDeterministic (and optional LLM) continue-on-trivial-stop
Objective ModeKeep the user objective latched for long runs
Shadow modeLog would-fire decisions without resuming
Loop guardCap consecutive nudges so you don’t ping-pong

The problem it solves

Claude Code is polite. Polite agents:

  • Ask permission to continue
  • Offer multi-choice menus
  • Promise “next I’ll…” and stop
  • Restate the plan and park

That’s fine for interactive pair-programming. It’s death for unattended loops. Auto-Nudge is the productized answer to “just say continue.”

How to enable

  1. Settings → powerClaude.autoNudge.enabled (default on in current builds — verify).
  2. Set aggressiveness: conservative · balanced · aggressive.
  3. Review per-nudge toggles (powerClaude.autoNudge.nudges) — high-confidence patterns default on; lower-confidence off.
  4. Optional: shadowMode: true for a week; read ~/.power-claude/logs/events.jsonl.
  5. Optional: objectiveMode: true for long goal runs.
  6. Feature discovery can toast suggestions when disabled nudges would have helped.

Safety rails

  • Loop guard (loopGuardMax) stops infinite nudge ping-pong.
  • LLM fallback is optional, cost-bounded (Haiku-class by default).
  • Kill switch still wins: ~/.power-claude/state/emergency-off.
  • Nudge ≠ rewake after hard usage limits — see Auto-Resume.

Recommended first week

  1. Shadow mode on.
  2. Run real work.
  3. Inspect which nudges would fire.
  4. Enable only the patterns you trust.
  5. Turn shadow off.

FAQ

Will it skip real questions I wanted to answer?

That’s what aggressiveness + per-nudge toggles are for. Start balanced; disable patterns that steal control.

Is Objective Mode required?

No. It’s for multi-hour “finish this epic” runs. Interactive coding can leave it off.

Related

Primary sources

  • Settings under powerClaude.autoNudge.*
  • Video: media/video/scenes/auto-nudge/, objective-mode/
  • Feature notes: docs/features/auto/resume/nudges.md

Independent third-party software by Neural-LLM. Not affiliated with Anthropic.