The problem
Currently, Minuet appears to limit its context strictly to the text of the current active buffer (via minuet--get-context). While this works well for large single files, it drastically reduces the quality of suggestions for projects composed of many small, interdependent files. For example, if I'm editing main.js which imports a function from utils.js(an open buffer or file in the same project), Minuet has no visibility into utils.js and guesses the function signature, often incorrectly.
Suggestion
I would like an option to include "Project Context" in the prompt sent to the LLM. Since modern models (Gemini 1.5/2.0, Claude 3, GPT-4o etc.) have massive context windows (128k - 1M+ tokens), we can afford to send significant chunks of the project structure or even full file contents.
Possible Ideas
- Automatically append the content of other open buffers (in the same mode/project)
- Projectile integration + .gitignore consideration
The problem
Currently, Minuet appears to limit its context strictly to the text of the current active buffer (via
minuet--get-context). While this works well for large single files, it drastically reduces the quality of suggestions for projects composed of many small, interdependent files. For example, if I'm editingmain.jswhich imports a function fromutils.js(an open buffer or file in the same project), Minuet has no visibility intoutils.jsand guesses the function signature, often incorrectly.Suggestion
I would like an option to include "Project Context" in the prompt sent to the LLM. Since modern models (Gemini 1.5/2.0, Claude 3, GPT-4o etc.) have massive context windows (128k - 1M+ tokens), we can afford to send significant chunks of the project structure or even full file contents.
Possible Ideas