follow-instruction

Enforce a question-first workflow with pre-execution audits before acting.

7|1|Updated Apr 1, 2026

Search Tags:#workflow #audit #clarification #safety #boundary #ai-guardrails #follow-instruction

Ready to run

npx skills add https://github.com/gyy0592/claude-config --skill follow-instruction

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: follow-instruction
Source: https://github.com/gyy0592/claude-config/tree/main/skills/follow-instruction
Command: npx skills add https://github.com/gyy0592/claude-config --skill follow-instruction

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

AI often fills ambiguity with assumptions, leading to scope creep and unsafe actions. This skill enforces a disciplined pipeline to confirm understanding before acting.

Core Features & Use Cases

Enforces a four-stage clarify-confirm-audit-execute cycle for any user instruction.
Prevents overreach by requiring explicit questions and confirmations.
Use case: when a user asks to modify files or run tasks, the AI asks clarifying questions and documents intent before performing any action.

Quick Start

Always ask clarifying questions, confirm intent, and audit boundaries before taking any action.

Frequently Asked Questions about follow-instruction

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I prevent AI from making assumptions on ambiguous user requests?▼

To prevent AI from making assumptions on ambiguous requests, enforce a question-first workflow requiring the system to ask clarifying questions and confirm intent before taking action. This stops scope creep by documenting boundaries upfront.

What is a pre-execution audit in AI task workflows?▼

A pre-execution audit in AI task workflows is a verification step that reviews and documents system boundaries before any action is taken. It ensures understanding is validated against user intent, guarding against unsafe actions and scope creep.

How do I add AI guardrails for modifying files or running system tasks?▼

You can add AI guardrails for modifying files by implementing a strict understand, clarify, confirm, and audit pipeline. This enforces explicit user confirmation and boundary checks before the AI executes any file modifications or system tasks.

Does enforcing a clarify-confirm-execute cycle slow down task execution?▼

Enforcing a clarify-confirm-execute cycle introduces a deliberate verification phase before task execution. While it adds an initial clarification step, it prevents costly errors and unsafe actions by ensuring the AI fully understands the instruction boundaries first.

When should I use a question-first workflow for AI instructions?▼

You should use a question-first workflow for AI instructions whenever a user request is ambiguous, involves changes to files or systems, or triggers slash-commands where understanding must be verified to prevent overreach and unsafe actions.

Why does AI fill ambiguity with assumptions during scope creep?▼

AI fills ambiguity with assumptions because it attempts to complete tasks without explicit boundaries, leading to scope creep and unsafe actions. Implementing a visible pre-execution audit forces the system to ask clarifying questions instead of guessing intent.