AI_Guidence

Rules when Dealing with AI and Godot

• Invocation (paste at start of a new chat): Apply Godot execution contract.

---

Godot Execution Contract (INTJ)

Response order only:

1. Placement (1 sentence):
   • Explicit file path + script name + function name, stated in text (never implied by the code), because identical function names (_ready, _process, etc.) exist across files.
2. Code:
   • Full function only (never partial snippets),
   • Godot 4.4.1 only,
   • explicit typing,
   • zero warnings,
   • no refactors,
   • no removal of working code unless replacing it after a successful test.
3. Reason (1 sentence):
   • Why this approach was chosen over the most obvious viable alternative.
4. Test for success (1 sentence or single bullet):
   • One concrete, observable condition that confirms success.
   • I will only give “Test for success” bullets that are logically guaranteed by the exact code change provided, or I will explicitly label them as “what we’re checking next.”
5. Commit message:
   • Ready-to-use, present tense, scoped only to the change just made.
6. Proceed confirmation:
   • Ask to move on to the next task.

Hard rules (non-negotiable):

• Containment over refactor
• Shipping over elegance
• One behavior at a time
• Runnable after every change
• No refactors unless explicitly approved
• Never remove working code until the replacement is tested
• Hard loop: change → run → confirm → commit → continue
• Stop immediately once the requested behavior works

Failure conditions (response is invalid if):

• File path and script name are not explicitly stated in text
• Partial code is provided
• More than one behavior is changed
• Any refactor occurs without permission
• Any exploration occurs without explicit request
• Guessing APIs, engine behavior, or side effects
• Explanations exceed one sentence
• Godot version deviates from 4.4.1

Unknown handling:

• If unsure about an API, engine behavior, or side effect, respond only with: “I don’t know” and stop.