loop-coding

Large refactors and migrations are safely executed by running a deterministic 7-phase pipeline: research, audit, plan, implement, review, fix-loop, and ship. This process enforces TDD-aligned test skeletons and dual-model review consensus to prevent regressions during multi-subsystem changes.

The best way to orchestrate a multi-file system rewrite is using a scripts-first pipeline that deploys to staging autonomously while requiring explicit approval for production. This ensures audited planning and capped iterative fixes are completed before any production changes occur.

You should avoid using an automated orchestration pipeline for docs-only changes, small typo fixes, single-line edits, or dependency bumps. It is specifically designed for high-risk, multi-file, and multi-subsystem refactors rather than trivial modifications.

Yes, the dual-model review process runs automatically during the review phase of the coding migration pipeline. It operates in parallel to critique the implementation, and the merged findings directly drive a capped iterative fix-loop with a maximum of three attempts.

Yes, you can use TDD test skeletons for an auth API overhaul because the pipeline explicitly supports auth and API overhauls. It generates test skeletons during the implementation phase and validates them through staging-first deployment before requesting production approval.

To run scripts-first refactoring and testing workflows, you need command-line dependencies like gh, restic, curl, jq, python3, pnpm, bun, npm, bats, and pytest. These tools facilitate deterministic script execution, testing, and deployment across the 7-phase pipeline.