no-ask-human

Detects when Claude Code's output contains questions directed at the user ("Should I...?", "Would you like...?") and reminds the agent to decide autonomously instead of asking.

Essential for autonomous/unattended operation where no human is available to answer questions. Every question is a blocking call with no one on the other end.

Event Configuration

• Event Type: PostToolUse
• Tool Matcher: "" (every tool invocation)
• Category: automation

Patterns Detected

• "Should I...?"
• "Would you like...?"
• "Do you want...?"
• "Which approach/option...?"
• "What do you think...?"
• "Shall I...?"
• "Do you prefer...?"
• "Can you confirm...?"
• "Is that okay...?"
• "Let me know if..."

Decision Framework (provided to agent)

1. Technical choices → pick the standard/conventional option
2. Implementation details → follow existing code conventions
3. Ambiguous specs → follow common industry practices
4. Errors → investigate and fix (up to 3 attempts)
5. If truly blocked → document the blocker and move to the next task

Environment Variables

Variable	Default	Description
CC_NO_ASK_ENABLED	1	Set to 0 to disable
CC_NO_ASK_PATTERNS_FILE	(none)	Path to file with additional patterns (one regex per line)

Design Philosophy

This hook never blocks (always exit 0). It issues a reminder only. It can't unsend text, but it trains the agent via system message feedback to self-correct in future interactions.

Requirements

• jq for JSON parsing

Origin

Born from running Claude Code in autonomous 24/7 mode where every question hangs the session until a human checks in. Some sessions asked 3+ questions, burning 30+ minutes of idle time each. This hook reduced question frequency by ~90%.

Source

From claude-code-hooks — battle-tested hooks from 140+ hours of autonomous Claude Code operation.