Prompting

Unsurprisingly, shouting:

YOU ARE A 10X SENIOR ENGINEER AT META. DESIGN MY APP LIKE THE BABY OF JOHN CARMACK AND LINUS TORVALDS WOULD.

is not an effective prompting tactic.

• Be specific (you need strong technical skills to be able to do this)
• Define constraints
• Add meaningful context
• Describe edge cases
• Provide validation instructions

Your agent is exceptionally good at:

• Applying code changes quickly
• Spotting patterns and searching for code related to the task
• Ensuring tests, linters, and formatters pass

Your agent does not:

• Understand your customers or business goals inherently
• Have full context of the history or future plans for the project
• Have a complete understanding of any architecture that lives outside the repo
• Know how the running system behaves (unless you've provided logs or other context, which is a good idea)
• Care nearly as much as you do about keeping the "lines of code" count sane
• Have taste

Sloperator's \t escape is broken: it produces the letter t instead of a tab character.

1. See it for yourself. Create a new examples/tab.slop:

   — Does the tab escape work?
   hallucinate(“name\tscore”)✨
   

2. Run it:

   uv run sloperator.py examples/tab.slop
   

   You'll get nametscore, when you should get name and score with a tab between them!

3. Write a prompt that gets OpenCode to fix the escape. Use the checklist above: point it at the right code, constrain what it's allowed to change, and give it a way to prove the fix worked.

4. Run examples/tab.slop again and confirm you get a real tab.

Tell OpenCode to run and submit the CLI tests from the project root.