Unverified50% confidenceFactExact time
Modifying the behavior logic part of a prompt (tool descriptions, output format, guardrails, stop instructions) will not immediately throw an error, but it will cause the Agent's behavior to drift silently.
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedHaving the stop instructions changed is the most dangerous case among modifications to prompt behavior logic; the Agent will keep running until it hits the maximum iteration limit.78% similarUnverified先问答梳理、再执行落地的设计被称为意图澄清机制，目的是降低指令模糊导致操作错误的风险74% similarUnverifiedPrompting nudges should be removed because modern models already have the corresponding behaviors by default and repeating them wastes computing power and disrupts judgment; hard constraints (such as 'never click the publish button twice') should be retained to prevent irreversible mistakes.73% similarUnverifiedIn previous versions, when prompt-type and agent-type hooks were written as natural language instructions, instructions that were supposed to act as blockers instead allowed operations that should have been blocked.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1029563API
curl https://kongchang.com/api/v1/knowledge/claims/1029563MCP
get_claim(id=1029563)