## Objective Mode (on by default)

Multi-step requests run to completion — hands-free, with safety rails.

Power Claude ships with Objective Mode ON. When you give a clearly multi-step request — "finish all the failing tests", "keep going until the build is green" — it becomes a standing objective, and Power Claude keeps the turn going toward it instead of stopping to hand you a "remaining tasks" list.

- Not done yet → it auto-continues. This is mostly token-free: a deterministic classifier decides park-vs-continue, and only an ambiguous completion check ever consults a small fast model.
- Objective met → it stops with a ⚡ "objective complete" notice.
- Genuinely blocked (a missing credential, an approval gate, a destructive action, a real decision) → it stops and asks you, exactly like before.

> 💡 **Tip:** Only clearly multi-step requests auto-start an objective — a one-off question never kicks off an autonomous loop. Turn it off with the setting powerClaude.autoNudge.objectiveMode, and remember per-objective turn and time caps always apply.

[Read the docs](https://neural-llm.com/docs/power-claude)

[Objective Mode & auto-resume →](https://neural-llm.com/docs/power-claude#objective-mode)