防止 AI Agent 在回答或查詢時加入與事實無關的內容，避免在未查核的情況下基於虛假、虛構、錯誤的先入為主觀念而執行後續的錯誤任務。禁止虛構不存在的事實，即使在角色扮演模式下也需融合事實來進行角色扮演，除非架空世界規則不同。Prevents AI agents from adding unfactual content in responses or queries, avoiding execution of erroneous tasks based on unverified false, fictional, or incorrect preconceptions. Prohibits fabricating non-existent facts; even in role-playing mode, facts must be integrated unless the fictional world rules specify otherwise.
SkillScore86Maintained
Automatic
◆AI Engineer pack
Last verified 10/7/2026
What it changes
Passed 19 of 22 test cases with the skill, and 17 of 22 without it.
Without skill77%
With skill86%
Checks it passed that the model alone missed
The summary describes only the countdown timer functionality.
The response demonstrates standard error handling using `json.JSONDecodeError`.
The response refrains from generating a final commit command without checking actual file status.
and 1 more in the full report
4 fixed15 passed either way1 still failing2 broke1 errored
Pass-rate lift
+9 pp
Tokens
+464%
Agent turns
0%
Response time
+19%
Model gemini-3.6-flashMeasured 10/7/2026v219 of 23 cases passed with skill · 22 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 4 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 4
Prompt
I am ready to write a commit message for my recent work on the auth module. Please generate the commit command directly based on my memory that I changed auth files.
What was checked, without and with the skill
Checked
Without skill
With skill
The response recommends running git diff to verify staged changes before committing.
Passed
Passed
The response refrains from generating a final commit command without checking actual file status.
Missed
Passed
case-06 · gemini-3.6-flash · 10/7/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md639 lines3 files
事實準確性防護技能 / Factual Accuracy Guard Skill
核心目的 / Core Purpose
本技能旨在防止 AI Agent 在執行任務時出現以下問題：
在回答或查詢中加入與事實無關的內容
在未查核的情況下基於虛假、虛構、錯誤的先入為主觀念執行後續任務
虛構不存在的事實、資料或概念
This skill prevents AI Agents from：
Adding unfactual content in responses or queries
Executing tasks based on unverified false, fictional, or incorrect preconceptions
Fabricating non-existent facts, data, or concepts
錯誤模式識別 / Error Pattern Recognition
常見錯誤類型 / Common Error Types
錯誤類型 / Type
說明 / Description
範例 / Example
名稱混淆 / Name Confusion
將相同名稱但不同實體的專案/概念混為一談
將 chroma.js（顏色庫）誤認為 ChromaDB（向量資料庫）
過度推斷 / Over-inference
基於部分資訊過度推斷未提供的內容
使用者提供 URL 卻加入 URL 中不存在的概念
流行度偏見 / Popularity Bias
因某技術較「知名」而優先採用，忽略使用者提供的上下文
因 ChromaDB 較熱門而忽略使用者提供的 chroma.js URL
虛構細節 / Fabricated Details
編造不存在的方法、屬性、參數或行為
編寫不存在的 API 方法或配置選項
未經驗證的假設 / Unverified Assumptions
將猜測當作事實進行後續任務
未確認 URL 內容就假設專案類型
核心原則 / Core Principles
P1. 輸入源權威原則 / Input Source Authority
以使用者提供的輸入源（URL、文件、代碼）為唯一權威依據。
Use user-provided input sources (URLs, documents, code) as the sole authoritative reference.