Skip to main content

Guides

Objective Mode in Power Claude — keep multi-step work going

Objective Mode in Power Claude — keep multi-step work going

When you ask Claude Code to finish all the failing tests or keep going until the build is green, you want a standing goal — not another “remaining tasks” list every few minutes. Objective Mode is Power Claude’s harness-controllable equivalent of Claude Code’s /goal: multi-step requests become a standing objective the agent works toward, with token-light stop checks and full safety on real blockers.

TL;DR

QuestionAnswer
What is it?Auto-promote clearly multi-step requests into a standing objective
DefaultOn (powerClaude.autoNudge.objectiveMode)
What you feelFewer “here’s what’s left” stops mid-job
When it stopsObjective met (⚡), or a genuine blocker
NotBypassing Anthropic limits, inventing quota, or looping on one-off questions

How it works

  1. You send a clearly multi-step request (“finish all…”, “keep going until done”, “fix all…”).
  2. Power Claude auto-derives a standing objective and injects it as context so the model works toward it.
  3. After each stop, a deterministic stop-classifier decides park vs continue without an API call.
  4. Only a genuinely ambiguous “is it done?” check may use a small/fast model (via your rotation proxy) — cost is negligible.
  5. Outcomes:
    • Not met → auto-continues (no remaining-tasks dump).
    • Met → stops with a ⚡ objective complete notice.
    • Blocked (missing credential, approval gate, destructive action, real decision) → stops and asks, same as today.

One-off questions never start an autonomous loop. Hard-veto stops always park. Per-objective turn/time caps apply.

How to enable (VS Code)

  1. Install Power Claude.
  2. Settings → search objectiveMode or autoNudge.
  3. Confirm Objective Mode is on (default).
  4. Give a multi-step goal; watch for ⚡ complete when done.
  5. To require an explicit CLI objective only: turn the setting off, then use pc objective set when you want a standing goal.

Pair with Auto-Nudge / Auto-Resume

  • Objective Mode keeps multi-step intent alive.
  • Auto-Nudge / Auto-Resume clear trivial stop phrases and continue safe work.
  • Resume stalled finishes abandoned sessions in parallel worktrees.

Auto-Nudge how-to · Auto-Resume · Resume stalled

Safety & compliance

  • Official Claude Code transport; seats/profiles you own.
  • Does not bypass Anthropic rate limits or invent quota.
  • Real blockers still stop and ask.
  • Emergency off: ~/.power-claude/state/emergency-off.

FAQ

Does a single question start a loop?

No. Only clearly multi-step language auto-starts an objective.

How do I force off?

Settings → powerClaude.autoNudge.objectiveMode → off.

Related

  • Self-help hub
  • Product page
  • Walkthrough: media/walkthrough/objective-mode.md in the extension repo
  • Video scene: media/video/scenes/objective-mode/

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