text: human code: AI

I just saw Vercel release a lock file for their v0 product. The idea being that their AI won’t modify those files but only read them.

It makes sense that IDE’s implement something similar. And I assume the best name for it would be .llmignore.

Even for my own personal workflows I think the agents would have far better outputs if they know they can’t touch certain files of mine, and that they should work around them as best they can. So I would love to see this implemented, lots of other potential benefits;

• Keep optimized algorithms intact
• Preserve specialized implementations
• LLM’s encouraged to build off your code not change fundamentals
• Consistent behavior across LLM tools

Implementation

Create a .llmignore file in your project root:

# .llmignore
      # Highly optimized modules
      src/core/rendering-engine.js
      src/utils/critical-algorithms/*
      
      # Legacy code
      legacy/*
      
      # Pattern matching
      *.asm
      src/ml/trained-models/*.py
      

No access?

Maybe it’s not required but no access could be useful. Something like;

# .llmignore
      # Files LLM cannot read or modify
      [no-access]
      src/ml/models/*.pth    # Model weights
      src/config/*-keys.json # API keys and secrets
      *.asm                  # Assembly files
      *.bin                  # Binary files
      
      # Files LLM can read but not modify
      [read-only]
      legacy/*                   # Legacy code
      src/ml/trained-models/*.py # Trained model scripts
      

Let’s do it!