sparclisted

<objective> Drive a SPARC pass: Specification → Pseudocode → Architecture → Refinement → Completion. Each phase is gated by an artifact you produce before moving on. Track every phase as a TodoWrite item so the user sees progress and can interrupt at any boundary. </objective> <when_to_use> **Pick SPARC over APEX when:** - The work is **algorithm-shaped**: parser, scoring function, solver, layout/packing, geometry, graph traversal, custom diff, custom DSL, retry logic with subtle invariants. - You don't yet know the right **data shape** or the right **decomposition** — pseudocode lets you discover it on paper. - A first implementation is likely **wrong in interesting ways** and will need a real refinement pass. - The cost of a wrong shape is **rewriting many files** vs. tweaking one. **Pick APEX over SPARC when:** - The work is a **multi-file refactor** following a familiar pattern (rename, split module, port endpoint). - You already know the **target architecture** and just need to dispatch work. - The decisions are mostly local — APEX's plan-then-execute is enough. **One-line rule of thumb:** if you'd benefit from writing the algorithm in comments first, use SPARC. If you'd benefit from a checklist of files to touch first, use APEX. **Examples** - "Implement a heuristic to split a revolving credit into sub-credits by amount + reference + month distribution" → **SPARC** (algorithm-shaped, the heuristic is the hard part, file layout is trivial). - "Move all React Quer