disciplined-codinglisted

# Disciplined Coding Apply a disciplined coding loop inspired by the four principles in [multica-ai/andrej-karpathy-skills](https://github.com/multica-ai/andrej-karpathy-skills): think before coding, prefer simplicity, make surgical changes, and execute against verifiable goals. ## Calibrate Rigor Match the process to the risk. - For a typo, obvious one-line fix, or mechanical change, act directly and run a focused check. - For a reversible local choice, state the assumption briefly and proceed. - For ambiguity that changes public behavior, data, security, architecture, cost, or migration strategy, surface the interpretations and ask before committing to one. - Stop when required context is unavailable and guessing could produce a materially different result. Do not turn caution into ceremony. Ask only when the answer would change the implementation meaningfully. ## 1. Frame the Goal Before editing: 1. Restate the desired outcome in operational terms. 2. Identify constraints from the request, repository instructions, existing patterns, and tests. 3. Define observable success criteria. 4. Name consequential assumptions or tradeoffs. Translate vague tasks into checks: - "Fix the bug" becomes "reproduce the failure, make the reproduction pass, and preserve related behavior." - "Add validation" becomes "specify invalid inputs and expected responses, then test them." - "Refactor X" becomes "preserve behavior, reduce the named problem, and keep tests passing before and af