Auto fix all rubocop offenses with chatgpt / openai / local llm

gem install rubofix

• Get openai key and export as RUBOFIX_API_KEY= (for non-openai see "Options" below)
• Break something for rubocop (remove # rubocop:disable comment or remove Enabled: false from .rubocop.yml)

MAX=2 rubofix
# Fixing MAX=2 of 35 warnings with MODEL=gpt-4o-mini ...
# Fixing Rakefile:89:19: W: Lint/AssignmentInCondition: Use == if you meant to do a comparison ... with:
#  unless (template = args[:template])
# Fixing Rakefile:103:22: W: Lint/AssignmentInCondition: Use == if you meant to do a comparison ... with:
#    next unless (spec = e["spec"])
git diff
# Rakefile
#  -unless template = args[:template]
#  +unless (template = args[:template])
git commit -am 'fixing rubocop warnings'

• only fix given files rubofix file1.rb file2.rb
• DEBUG=1 show prompt and answers
• CONTEXT=10 feed 10 lines of context around the offense to the model
• RUBOFIX_URL= defaults to https://api.openai.com
• RUBOFIX_API_KEY=

• rake to run unit tests
• rake integration to run integration tests, they need an api key set
• rake bump:<major|minor|patch> to create a new version
• rake release to release a new version

• custom api endpoints
• local LLM support
• read rubocop.todo file and autofix everything in there
• retry on api failures
• parallel execution
• colored output
• try different temperatures to get better results
• try to send output back to llm with "check this makes sense" to fix bugs
• produce diffs and then apply them so we can fix multiple things in 1 file without changing line numbers

Michael Grosser
michael@grosser.it
License: MIT