Prompt Iteration

The current prompt (give me the "goals" of the source file) is literally the first thing the author tried. Some more thought could go into the combination of what we ask GPT on indexing ("the goals") vs. what we prompt the user for at search time ("find code in $repo which..."). Even subtle iteration on these two prompts may provide dramatically better results.

Locate the prompts used during indexing and at search time, then read how each is assembled and passed to GPT. Experiment with their wording and interaction, and compare the resulting semantic search quality; the issue does not name files, tests, or a concrete completion criterion.

Written by the indexing model from the issue text.