Unverified50% confidenceFactExact time
The model itself does not actually operate files or send requests; it only generates invocation intents, and what actually carries out execution is the Harness.
1
Sources
50%
Confidence
Long-term
Relevance
11/7/2026
First Seen
Sources
DeepSeek Open-Sources Agent Foundation: How a Plugin Architecture Lets AI Actually Get Work Done
bilibili啊年的AI笔记11/6/2026
Related Entities
Related Claims
Unverified工具调用机制中,模型不会直接执行任何代码,而是返回结构化的调用请求,由本地程序执行后再回传结果;因此安全边界掌握在开发者手里,模型只能请求,不能擅自读取文件或发送邮件。75% similarUnverifiedIn the "LLM + tool calling" paradigm, the model does not directly manipulate files itself, but instead outputs structured instructions, with deterministic code handling the actual execution.74% similarUnverifiedA model-agnostic harness is a runtime framework that decouples the reasoning model from tool calling, file reading and writing, and terminal execution, such as OpenCode or Py, and is not tied to a specific model vendor.73% similarUnverifiedThe model never directly calls code; it simply returns a message requesting that a tool be run, and it is the loop written by the developer that actually decides what gets executed.73% similarUnverifiedIn the tool calling loop, execution authority always stays with the caller's code; the model is only responsible for "requesting" the call.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1145420API
curl https://kongchang.com/api/v1/knowledge/claims/1145420MCP
get_claim(id=1145420)