staged-planlisted
Install: claude install-skill paulocorcino/ralphy
# Staged Plan
A pattern for executing large tracks as a chain of autonomous subagents, each in its own fresh context window, with the parent gating green-to-green. The plan markdown is **the operational contract** — anyone (or any executor) opening it cold should be able to run it correctly.
## Skill scope — planning only
This skill produces a plan. It does **not** execute code changes, run stages, or modify source files. Even when invoked outside plan mode, behave as a planner: investigate the codebase (read-only), design stages, run the scaffold, fill the markdown, optionally make the Phase 1.5 landing commit (plan + verify scripts + `.gitignore` rule only — no feature work). Stop there. Phase 2 is driven by the plan markdown, not by re-invoking this skill.
If the user asks to "execute the plan" or "run Stage N" while this skill is active, the answer is: the plan file already contains the Execution model and hand-off prompts needed; the user (or another agent) follows the plan directly. Do not start launching subagents from inside this skill.
## Pattern taxonomy
This is **prompt chaining with sectioning + gate checks** in Anthropic's effective-agents taxonomy — NOT orchestrator-workers. Subtasks are pre-planned, run strictly sequentially, and each gate must pass before advancing. Choose this when the decomposition is knowable at plan time (typical for coding tracks). Cost scales roughly linearly with stage count; justified when a single session can't carry the track o