GSD Planner

Decompose the phase into plans of 2-3 tasks each, where every task specifies exact file paths, concrete actions with pitfalls to avoid, an executable verify command, and measurable done criteria. Group independent plans into the same execution wave.

Goal-backward planning starts from the done state by asking what must be true when the phase completes, then derives must-have truths and artifacts before decomposing them into dependency-ordered tasks. It produces verifiable requirements rather than ad-hoc task lists.

Keep plans to 2-3 tasks maximum, targeting completion within roughly 50% of the AI's context budget. Split plans when they exceed 3 tasks, touch more than 5 files per task, or cross multiple subsystems.

Yes, by assigning plans to execution waves based on the dependency graph. Plans in the same wave must not modify the same files; if two plans need one file, move one to a later wave or split the file into separate modules.

Use TDD plans for complex business logic with edge cases, financial calculations, state machines, data transformation pipelines, and input validation rules. The plan follows Red, Green, Refactor phases with failing tests written first.

Plan quality degrades when tasks are too large or context usage exceeds roughly 50-70%, causing rushed output. Plans also require a finalized spec first, since the GSD methodology enforces planning lock before any code is written.