Maybe it contributes a little or not, but I manage to get the most out of Cursor, with Prompt ultra-closed.
Example:
CANONICAL STANDARD — PROMPTS FOR CURSOR
STRICT MODE.
FINAL ARCHITECTURE MODE.
TOTAL CONVERGENCE MODE.
Objective:
- PROHIBITED: Define what is prohibited to do in the prompt, example: be creative.
- AIM Define the expected result in specific points.
- MINIMAL CONTEXT System, module and key restrictions. docs/AI_Context_Master.md docs/canonical/.
- INITIAL VERIFICATION Set-Location “C: …” git status --short If the repository is not clean, make commit baseline.
- SCOPE (CLOSED) Backend, Frontend, allowed files and bans.
- RULES - Do not refactor outside of what is necessary - No new dependencies - Do not break contracts - STOP if context is missing
- TASKS Sequential, concrete and verifiable list.
- LOGIC Clear decision rules (if applicable).
- INVARIANTS What cannot be broken.
- VALIDATION Verifiable operating cases.
- EVIDENCE - MD document with changes, validations and risks. - The document is mandatory, even if the process fails there must be traceability
- OUTPUT FORMAT Structured document + summary.
- STOP Stop if there is ambiguity or lack of context.
- COMMIT If there are other changes in the repository other than the one produced by the prompt, it must be made additional commits and leave the reference.
- REJECTION CRITERIA Conditions that invalidate the result.
