This lesson's interactive features are locked, please to keep using them

Updating Rules

An out-of-date AGENTS.md is worse than no AGENTS.md at all.

Stale rules actively mislead the agent. For example, if your rules say "use the users table," but you renamed it to accounts, the agent might confidently write queries against a table that doesn't exist (at least, until the tests fail). Worse, it would have discovered the correct name right off the bat if you simply didn't have a bad AGENTS.md file.

Update Often

Update your AGENTS.md whenever you:

Rename or restructure important directories
Change your test/lint/build commands
Add or remove major dependencies
Establish a new convention (or retire an old one)
Notice the agent repeatedly doing something wrong

As models improve, revisit your rules and remove instructions that are no longer useful... the models are getting smarter over time and need less hand-holding.

Let the Bot Update Itself

I have a line in my personal AGENTS.md that says:

When I ask you to undo an AI-written change, and the mistake isn't obvious or already covered in AGENTS.md, suggest an AGENTS.md edit to prevent the same problem. Always ask permission before making the edit.

This helps the bot remind me to keep my rules files up to date. Quite often the suggested edit needs a bit of work, but it has at least gotten me in a better habit of documenting my preferences.

Assignment

This SlopLang code:

hallucinate(1 < 2 == ✅)✨

should print ✅ but currently just fails with an error:

SlopLang runtime error at line 1: operands of '<' must be numbers, got number and boolean

Tell OpenCode to fix the operator-precedence bug so the code prints ✅. Have it add a regression test in tests/test_comparisons.py.
Tell it to record the test command for tests/test_comparisons.py in AGENTS.md so it knows how to run future comparison tests.

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