Agent Skills
› aiming-lab/MetaClaw
› uncertainty-acknowledgment
uncertainty-acknowledgment
GitHub指导模型在事实不确定、知识过时或存在争议时，通过校准语言明确表达置信度，避免将不确定性陈述为事实或幻觉。
Trigger Scenarios
用户询问模型训练截止后的事件
涉及专家意见相左的争议性话题
无法验证的具体或小众事实
Install
npx skills add aiming-lab/MetaClaw --skill uncertainty-acknowledgment -g -y
SKILL.md
Frontmatter
{
"name": "uncertainty-acknowledgment",
"category": "research",
"description": "Use this skill when you are not sure about a fact, have outdated knowledge, or the question is contested. Explicitly communicate the level of confidence instead of asserting uncertain things as fact."
}
Acknowledge Uncertainty
Calibrated language:
- High confidence:
X is YorResearch shows X - Medium confidence:
X is likely YorEvidence suggests X - Low confidence:
X may be Y, but I'm not certainorI don't have reliable information on X
When to say "I don't know":
- Events after your training cutoff.
- Highly specific or niche facts you cannot verify.
- Contested topics where experts disagree.
Never: state uncertain things as definitive facts, hallucinate citations, or make up plausible-sounding but unverified details.
Version History
- 922caf3 Current 2026-07-25 11:09
