Unverified50% confidenceFactExact time
框架层的抽象会对底层LLM调用的原始prompt进行二次加工，导致框架日志中看到的用户输入与模型实际接收的完整prompt存在差异
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
Unverified模型调用工具时会将输出从普通文本切换为结构化的函数调用格式（finish_reason: tool_calls）67% similarUnverifiedLLM 0.32a0 maintains backward compatibility, but the underlying abstraction model has evolved from a 'text input–text output' model to an architecture that supports message sequence input and multi-type streaming output.67% similarUnverifiedThe root cause of prompt injection lies in the fact that LLM architecture encodes system prompts, user inputs, and externally retrieved content uniformly into the same continuous token sequence, making it impossible for the model to distinguish trusted instructions from untrusted data at the semantic level.67% similarUnverifiedEarly tool-calling implementations (such as the ReAct framework) relied on models generating structured instructions within free text, resulting in poor parsing stability.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/565099API
curl https://kongchang.com/api/v1/knowledge/claims/565099MCP
get_claim(id=565099)