Assumptions

• we have somewhat validated that we can get chatgpt to generate dark code
• even if chatgpt cannot be made good right now, we feel confident that ditching the technical debt and moving forward cleanly with the improved language, a VSCode-based editor,
• a VSCode backend will take some time, but a CLI/repl can be used in the short term, and will be valuable after that
• we all have the skills to work on all the components

Components

• chatgpt
  • prompts
  • api connection
  • which model to use
  • where is this called from?
• vscode extension
  • split between ts/dark
  • feature list/spec/ordering of experiments
  • UX/UI design/mockups
  • UX of interacting with chatgpt
• dark-editor canvas
  • written in dark, bootstrapped using .dark files
  • all the apiserver endpoints
    • Initial-load, update-handler, etc.
  • locally we overwrite the dark-editor canvas with the .dark files
  • list of features, APIs
• language
  • Project Light darklang language
  • parser
  • dark interpreter executable
  • trace improvements
• light repo codebase cleanup/tech debt removal
• infrastructure
  • cloudrun
  • new DB? cockroach?
  • clean up tables
• dark repo
  • abandon PRs?
  • DB tracing?

Set of initial goals:

• Dark-editor endpoints
  • add them to local canvas
• [ ] Initial-load endpoint
• [ ] add-op endpoint
• [ ] download latest gpt prompt
• [ ] vscode extension features
  • [ ] design UX // Ocean
  • [ ] write small spec // Ocean
  • [ ] process initial-load request
  • [ ] show handlers, fns, dbs, secrets, types, etc, in sidebar
  • [ ] click on sidebar element, load pane with code
  • [ ] create new handler, etc
• [ ] gpt feature:
  • [ ] prompt input
  • [ ] regenerate handler code from prompt