← ClaudeAtlas

write-planlisted

Use when turning an approved spec or a small clear goal into an executable implementation plan — ordered tasks, file list, test plan, agent routing, and parallel contracts if more than one agent will edit code. Phase = Plan.
nuttaruj/rolepod · ★ 3 · AI & Automation · score 76
Install: claude install-skill nuttaruj/rolepod
# Write Plan Convert an approved spec or a clear small goal into a plan another engineer (or specialist agent) can execute without re-asking the user. ## Iron Rule <EXTREMELY-IMPORTANT> 1. NEVER start editing before the plan names the files, the order, and the verification per task. 2. NEVER spawn more than one parallel agent on the same feature without a written cohesion contract that pins file ownership and merge order. 3. NEVER write a vague task like "add tests" — every task names a test or evidence that proves it done AND the exact runnable command that checks it, so the build loop verifies without guessing. 4. Pick the simplest viable approach. Complexity needs an explicit reason and user awareness. </EXTREMELY-IMPORTANT> ## Skip when - A one-line fix on a single file · a question / explanation only. - The router tiered the task **R2** (1 file + its own test, clear scope, ≈≤30 logic lines) → the plan is a 3-5 line inline checklist in chat, each step with its verify command; no artifact. - Scope grows past one file mid-flight (the task's own test file does not count) → stop, write the real plan here. - **Spec-as-plan R3 lane:** ≤3 tasks the approved spec already lists 1:1 (files, order, verify command, dependencies), single-agent, no high-risk surface → the same inline checklist; a parallel layout, a risk path, a 4th task, or a compaction mid-plan → write the artifact. ## Boundary Owns: HOW / WHO / WHERE / ORDER — file list, task order, test plan per task, age