experiment-iterative-coder

Iterative code refinement automates quality gates by running lint checks and tests within plan, code, evaluate, and refine cycles. You initiate it by starting a code task with MODE: MORE_EFFORT to progressively elevate correctness across multi-file projects.

For multi-file coding tasks requiring high quality, use an iterative phase decomposition approach. It breaks tasks into sequential phases with up to 3 iterations per phase and a maximum of 10 total iterations to ensure robustness through automated self-evaluation.

Automated failure diagnosis works by evaluating lint and test failures, then crafting precise, data-guided improvement plans. This targeted planning enables rapid code refinement by directly addressing the diagnosed errors within the iteration cycle.

Use an iterative coding approach over single-pass output for high-stakes coding tasks, complex refactoring efforts, or scenarios where incremental quality gains are prioritized to ensure robustness and correctness.

Yes, the iterative coding workflow supports ruff and pytest as automated quality gates. It runs ruff for lint checks and pytest for test execution during the evaluate phase to surface targeted improvements before the next refinement cycle.

The iteration limits for phase decomposition allow up to 3 iterations per phase, with a maximum of 10 iterations overall. This constraint ensures structured, progressive quality elevation without excessive cycles during complex coding tasks.