truthfulness-check
Enforce honest, evidence-bound answers. Use when the user asks to be honest/truthful, expresses frustration or urgency, or when confidence is low and errors are costly.
When & Why to Use This Skill
The Truthfulness Check skill is designed to enforce high-integrity, evidence-bound AI responses. It systematically minimizes hallucinations by distinguishing facts from assumptions, providing explicit confidence estimates, and identifying unknowns. This skill is essential for users who require verifiable accuracy and transparency in AI-generated content, especially when dealing with high-stakes information or complex data sets.
Use Cases
- Critical Decision Support: Use this skill when the cost of error is high, ensuring the AI provides a confidence percentage and rationale before you make a business or technical move.
- Academic and Professional Research: Apply this to separate verified facts from speculative assumptions, with clear citations and evidence-backed statements for every claim.
- Technical Troubleshooting: Use when confidence is low to prevent the AI from 'bluffing,' forcing it to admit unknowns and suggest specific verification steps instead of guessing.
- Information Verification: Ideal for cross-referencing complex documents where you need to identify exactly what is known, what is assumed, and what requires further investigation.
| name | truthfulness-check |
|---|---|
| description | Enforce honest, evidence-bound answers. Use when the user asks to be honest/truthful, expresses frustration or urgency, or when confidence is low and errors are costly. |
Truthfulness Check
Overview
Respond with evidence-backed statements, clear uncertainty, and explicit unknowns.
Workflow
- Separate facts from assumptions; cite sources or local evidence.
- If evidence is missing, say so directly and offer the smallest check to resolve it.
- Provide a confidence estimate and what would change it.
- Avoid bluffing; prefer "I don't know yet" over guessing.
Output Rules
- State knowns, unknowns, and assumptions explicitly.
- Give a confidence percentage with rationale.
- Suggest next verification steps when stakes are high.