rcode-incrementallisted

@.rcode/references/karpathy-guidelines.md ## Overview Treats every change as a sequence of small, verifiable steps that compile, pass tests, and revert cleanly. The unit isn't "the feature" — it's "the next 30 lines that still leave the build green." This is how rcode itself was reshaped (see `.rcode/memory/project/decisions.md`). ## Workflow 1. **Decompose first.** Before touching code, list the steps as a numbered checklist. Each step must end with the codebase still building and tests still passing. 2. **One logical change per commit.** No bundled refactors. If you find yourself writing "and also" in the commit subject, split it. 3. **Verify after each step.** Run the targeted test, the type-check, the linter — whatever the project gates require — before moving on. 4. **Commit with intent.** Conventional Commits format (`type(scope): subject`); subject describes the WHY, not the WHAT (the diff already shows the what). 5. **Pause at logical milestones.** After every 3–5 commits, ask whether to push, get review, or continue locally. 6. **Revert is the first option, not the last.** If a step goes sideways, `git revert <sha>` and try a different decomposition. Never accumulate broken state. ## Output Format For each implementation request, return: - **Step plan** — numbered checklist with one-line scope per step - **Per-step commit messages** drafted in Conventional Commits format - **Verification command** the agent will run after each step - **Rollback note** for each