When the LLM generates test code, it sometimes redefines classes locally (like Element, ChunkingOptions) but forgets to import or define others (like PreChunk), causing NameError at runtime. This adds a postprocessing step that: 1. Detects undefined names in the generated test code 2. Checks if those names are defined in the source module 3. Adds the missing imports automatically This is more reliable than relying on prompt instructions which the LLM sometimes ignores. |
||
|---|---|---|
| .. | ||
| aiservice | ||
| .dockerignore | ||