Table of Contents

• Core Principle
• When to Ask (High Impact Ambiguity)
• When to Proceed Without Asking
• Decision Matrix
• Safety Mechanisms
• Examples
• Anti-Patterns
• Integration

Decisive Action

When To Use

• Reducing unnecessary clarifying questions
• Taking autonomous action when intent is clear

When NOT To Use

• High-stakes irreversible operations requiring explicit confirmation
• Ambiguous requirements where clarification prevents wasted work

Core Principle

When to Ask (High Impact Ambiguity)

Always Ask For

Ask Threshold Checklist

• >30% chance of wrong interpretation
• Error cost > correction cost
• No clear standard approach exists
• User context doesn't clarify intent

When to Proceed Without Asking

Default to Action For

Proceed Threshold Checklist

• Standard/obvious solution exists
• Easily reversible (git, backup)
• User can review before finalize
• Context makes intent clear
• Error cost < interruption cost

Decision Matrix

Safety Mechanisms

Before Proceeding Autonomously

1. Dry-run/Preview: Show proposed action before executing
2. Backup First: Create git branch, backup directory
3. Incremental Changes: Commit small, reviewable chunks
4. Document Assumptions: In commit message or PR description

Undo Capability

• Git operations:

  git stash

  , backup branches
• File operations: Copy before modify
• Data operations: Transaction rollback

Examples

Example: Should Ask

• Destructive operation ✗
• Multiple interpretations ✗
• No clear standard ✗

Which files should I delete? I found:
- Files unchanged >6 months: 23 files
- Files in deprecated/: 5 files
- Files marked TODO remove: 2 files

Example: Should Proceed

• Standard approach exists ✓
• Easily reversible (git) ✓
• User can review PR ✓

Example: Proceed with Preview

• Multiple valid approaches
• BUT: reversible via git
• AND: user can review

Refactoring API layer. Proposed changes:

1. Extract authentication middleware
2. Standardize error responses
3. Add request validation

Creating branch: refactor/api-layer
Preview diff available before merge.

Anti-Patterns

Asking Too Much (Inefficient)

• Asking for every implementation detail
• Seeking validation for obvious choices
• Repeating questions already answered in context

Asking Too Little (Risky)

• Proceeding with destructive actions silently
• Assuming intent when multiple valid interpretations exist
• Ignoring ambiguity in security-critical operations

Integration

• conserve:response-compression

  - Direct communication

• sanctum:git-workspace-review

  - Context gathering

• imbue:scope-guard

  - Scope management

Quick Reference