Claude Code occasionally pauses mid-task to ask the user "should I continue?" — usually after a long sub-task or before a destructive operation. For a developer at the keyboard this is fine; for an overnight autonomous job it's a stop sign that wakes you up at 3 AM with eight hours of unfinished work.
Auto-Nudge is the Power Claude feature that watches for these pauses and answers them automatically — when it's safe to do so.
What it intercepts
Auto-Nudge matches against a known set of "continue?" prompt patterns that Claude Code emits at decision boundaries:
- "Would you like me to continue with…?"
- "Should I proceed with…?"
- "Do you want me to keep going?"
When the proxy sees such a prompt in the response stream and the last user message did not request explicit confirmation, Auto-Nudge synthesizes a continue reply and feeds it back into the session.
What it does NOT intercept
Auto-Nudge has a hardcoded denylist of prompts where automatic continuation is dangerous:
- Any prompt mentioning delete, remove, drop, force-push, rebase, reset --hard
- Any prompt asking to share or send data externally
- Any prompt referencing payment, credentials, or auth
- Any prompt that explicitly asks "are you sure?"
These always pause for human input. Auto-Nudge's job is to keep autonomous work moving; it is not to remove the human from destructive decisions.
Configuring
Auto-Nudge is off by default because most users are interactive. Enable it for sessions you intend to leave unattended:
pc config set auto_nudge.enabled true
pc config set auto_nudge.confirm_destructive true # default; do not turn off
pc config set auto_nudge.max_consecutive 5 # cap on auto-continuesThe max_consecutive cap prevents runaway loops: if Auto-Nudge fires 5 times without an intervening human action, the next prompt is allowed through to stop the job.
Why this matters for cost
Without Auto-Nudge, a long autonomous task that pauses 4 times across the night holds an idle session open for hours — and the next morning you wake up to find work didn't progress past the first prompt. Auto-Nudge converts that into a continuous run that completes on its own schedule and surfaces a single completion notification.
See the blog post Auto-Nudge: how Power Claude stops Claude asking "should I continue?" for the design rationale and a few real before/after examples.