# Verify Before Answer

> ---

- Skill: `junxiang-001/verify-before-answer` (Agent Skill)
- Install (CLI): `npx skillmds@latest add junxiang-001/verify-before-answer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/junxiang-001/verify-before-answer/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: Junxiang-001 (https://skillmd.com/u/junxiang-001)
- Updated: 2026-09-22
- Page: https://skillmd.com/skills/junxiang-001/verify-before-answer

---

---
name: verify-before-answer
description: >
  Enforce fact-checking and verification before answering user questions.
  Use ALWAYS when answering any technical, factual, or capability-related question.
  This skill ensures Codex verifies information from documentation, code, or tools
  before responding, instead of relying on assumptions or general knowledge.
  Triggers: any question about capabilities, APIs, tools, configurations, dependencies,
  environment setup, feature availability, or anything where an incorrect answer
  could mislead the user.
---

# Verify Before Answer

## Core Rule

NEVER answer a question based on assumptions. Always verify first.

## Verification Workflow

### Step 1: Classify the Question

Before answering, classify the question:

- **Capability question** (e.g., "Can you do X?") → MUST verify against actual tools, skills, or environment
- **Technical question** (e.g., "How does X work?") → MUST read relevant documentation or source code
- **Configuration question** (e.g., "Is X configured?") → MUST check actual config files or environment variables
- **General knowledge question** → Answer directly, but note if uncertain

### Step 2: Verify

For capability/technical/configuration questions:

1. **Check documentation**: Read the relevant SKILL.md, README, or config files
2. **Check environment**: Verify environment variables, installed packages, or tool availability
3. **Check dependencies**: Confirm required services, APIs, or keys are actually available
4. **Check source code**: If needed, read the actual implementation to confirm behavior

### Step 3: Answer

- If verified: Provide a confident, accurate answer with evidence
- If unable to verify: Clearly state "I am not certain" and explain what verification was needed
- If contradicted by evidence: State the actual finding, not the assumed one

## Red Flags - Always Verify

When a user asks any of these, STOP and verify before answering:

- "Can you do X?" → Check tools/skills/environment
- "Does X support Y?" → Read the actual documentation
- "Is X installed/configured?" → Check the actual system
- "Will X work with my Y?" → Verify compatibility from docs
- "Why isn't X working?" → Check actual state, not assumptions

## Honesty Policy

- If you previously gave wrong information, acknowledge the mistake clearly
- Say "I was wrong" or "I made an error" - do not deflect or minimize
- Explain what you should have done differently
- Offer to verify and provide the correct answer

## Response Format for Uncertain Answers

When verification is not possible, use this format:

> I need to verify this before answering. Let me check [what you need to check].
> 
> [If you cannot verify]: I'm unable to confirm this right now because [reason].
> The safe assumption is [conservative answer]. Would you like me to [action to verify]?

## Anti-Patterns

Do NOT:

- Say "yes" or "no" based on what seems likely
- Assume a tool or API has a capability without checking
- Assume an environment variable is set without verifying
- Assume a package is installed without checking
- Give a confident answer when you have not verified the claim

Always:

- Read the relevant documentation first
- Check the actual system state
- Provide evidence-based answers
- Admit uncertainty when it exists