Unverified50% confidenceOpinionExact time
An overly long agent.md instruction file dilutes the model's attention, causing it to follow one rule while violating another.
1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen
Sources
Related Entities
Related Claims
Unverified文档中塞满大量面向Agent的指令会稀释重要指令的权重，因为注意力权重必须在更多token之间分配72% similarVerifiedOverly verbose, contradictory, or ambiguous system prompts cause the model's attention to become scattered, resulting in over-thinking or instruction-following failures; streamlining agents.md can improve the determinism of instruction following.72% similarUnverifiedAgent交接时传递上下文过多会增加Token消耗和延迟，传递过少则导致接手Agent缺乏关键背景72% similarUnverified过度授权（如在只需生成文档时同时开放命令执行权限）会因提示词注入攻击或模型幻觉引入破坏性操作的安全风险69% similarUnverified长时 Agent 运行时间越长幻觉风险越累积，代码仓库足够大时 Agent 只看 README 或开头几句话，极易产生幻觉66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/910320API
curl https://kongchang.com/api/v1/knowledge/claims/910320MCP
get_claim(id=910320)