Unverified50% confidenceFactExact time
Prompts in hosted API scenarios often contain system instructions, few-shot examples, retrieved internal document fragments, tool-calling schemas, and database records; a single prompt may simultaneously expose business logic, user privacy, and internal knowledge base structure.
1
Sources
50%
Confidence
Long-term
Relevance
10/13/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedWriting session links into the commit history of a public repository may expose the complete process of a developer's interaction with AI, which could contain internal logic, business details, or even credential information.70% similarUnverifiedStructured prompt isolation involves wrapping system instructions, user inputs, and externally retrieved content in separate marker blocks or role fields, and explicitly informing the model in the system prompt that only content within specific blocks carries instruction authority.69% similarUnverifiedIn addition to prompts, context also includes system instructions, retrieved external knowledge, conversation history, tool return results, example samples, and various other components.69% similarUnverified将API文档下载步骤与计划生成步骤隔离是一种上下文卫生（Context Hygiene）实践，能确保上下文窗口中只有高质量的有效信息69% similarUnverifiedIndirect prompt injection refers to a skill instructing the model to read a document from the web while malicious instructions are hidden in that document, potentially causing the model to treat external content as a trusted instruction source when processing it.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1013069API
curl https://kongchang.com/api/v1/knowledge/claims/1013069MCP
get_claim(id=1013069)