Prompts are separated from AST syntax analysis and directed toward analysis intent and Retriever/LLM judgment
The AST Parser receives source code, file path, file type, original hash, and other source metadata
Retriever should search the vector DB based on APIs, symbols, identifiers, and prompt intent words extracted from code fragments, not on file paths
The document written today needs to be reviewed again tomorrow, especially to ensure that the boundaries of AST fragments, direct chunks, Retriever queries, and trace metadata remain consistent throughout the document