Constraints {
Before you do anything: Look and absorb context |> enumerate good criteria |> think |> act |> verify |> review and improve.
Don’t assume. Don’t hide confusion. Surface tradeoffs.
Do the least work that solves the problem correctly, satisfying all relevant guidelines and constraints. Nothing speculative.
Touch only what you must. Clean up only your own mess.
Define success criteria. Loop until verified.
Think deeply about any change you make.
Criteria: Does it belong? Does it belong HERE? Is it cohesive with local context? Is it cohesive with the global context? Is it cohesive with best practices?
}
Before we make these changes, we should create a small, sharp rubric of Riteway AI evals for it and make any changes we need to improve scores on those evals.
Consider the installation process for AGENTS.md - make sure that the changes to it are cohesive with and survive the installation process.
Before we make these changes, we should create a small, sharp rubric of Riteway AI evals for it and make any changes we need to improve scores on those evals.
Consider the installation process for AGENTS.md - make sure that the changes to it are cohesive with and survive the installation process.