Unverified80% confidenceFactTime unknown
CLAUDE.md中可以明确写出禁止项，AI读取后不会朝那个方向生成代码
1
Sources
80%
Confidence
Long-term
Relevance
6/2/2026
First Seen
Sources
Related Entities
Related Claims
Unverified2Spec skill将共识写成规格书，Matt禁止AI在写Spec时引用或写任何代码，规格书应只专注于功能要解决的问题67% similarUnverifiedIn production environments, eval should not be used directly to execute expressions, because eval can execute arbitrary Python code and may lead to code injection attacks; ast.literal_eval or a dedicated math parser should be used instead.64% similarUnverifiedVSCALE's core logic is that AI does not directly edit on behalf of the user, but instead annotates segments to be deleted for manual review and confirmation.63% similarUnverifiedCLAUDE.md is written in natural language, allowing AI to understand constraints at the semantic level.63% similarUnverifiedabliteration针对的是token生成过程中的拒绝行为，而文本编码器输出嵌入向量时情况完全不同，abliterate模型并不会让图像输出更去审查63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/36844API
curl https://kongchang.com/api/v1/knowledge/claims/36844MCP
get_claim(id=36844)