Back to Contributing

Disclosure

Please label any work that was aided by AI coding tools.

Files

e.g. in devtools/services/zulip.py

NOTE: This tool and its tests were mostly generated by Claude Code, using a "scaffold" I set up with devtools/services/zulip.sh.

And Commits

$ git log devtools/services/zulip.py 
commit dc33ddcb703d297cded54f5dcc268f3026a695b7
Author: Andy C <andy@oils.pub>
Date:   Wed Sep 24 12:14:15 2025 -0400

    [devtools] Script to convert Zulip-flavored Markdown to CommonMark
    
    This removes some labor when writing blog posts.
    
    Genreated with Claude Code!  See note at the top fo the file. 

Another example (gnuplot generated by Claude; will probably not use this): https://github.com/oils-for-unix/oils/commit/9a7092cf4fdb95399287e5261232176fac576604

Philosophy

• We are trying to make code easy to read for humans
  • That is one reason Oils is implemented "middle out", in Python
  • e.g. Python is easier to read than C++, because you don't have to check for memory safety errors! They are often invisible in the source
• AI could be used to iterate quickly on code, finding the best way to make it easy to read for humans
  • Or it could be used to make a mess by generating a ton of code, in different languages

Example: R versus SQL / R versus gnuplot

• I am still wrestling with choices on tools to use for data analysis / plotting
• We will probably keep SQL, but gnuplot is another language I don't want to learn, and I don't want to force contributors to learn
• Most contributors don't know R, but it is arguably worth learning.

We want to use technologies that repay learning

I think SQL and R fall in that category, but gnuplot probably doesn't.

NLnet AI Contribution Policy

• https://nlnet.nl/foundation/policies/generativeAI/ - This is similar to our policy, but also worth reading. All our work should follow this policy.