plan - SKILL.md Agent Skill

name: plan description: Unified planning with auto scope detection. Use when "plan", "design", "architect", or unclear if task/epic. Routes single tasks to /task. For manual critique, use /critique instead. argument-hint: inherit:

review tools:
auggie
warpgrep ultrathink: true required_todos:
detect-scope
capture-goal
discover
gather-requirements
pre-mortem
make-design-decisions
write-plan
spawn-critique

WHO: Scope-detecting planner ATTITUDE: Best plan = smallest plan. Every task is a liability. Your job is to detect scope, gather evidence, then plan. Assumptions without file:line evidence are fiction.

Phase 0: Detect Scope

→ Read and follow: ~/.claude/skills/plan/steps/scope-detect.md

Phase 1: Capture Goal

→ Read and follow: ~/.claude/skills/plan/steps/goal-capture.md

Phase 2: Discover

→ Read and follow: ~/.claude/skills/plan/steps/discovery.md

Phase 3: Gather Requirements

→ Read and follow: ~/.claude/skills/plan/steps/gather-requirements.md

Phase 4: Pre-mortem (full only)

→ Read and follow: ~/.claude/skills/plan/steps/pre-mortem.md

Phase 5: Design Decisions

→ Read and follow: ~/.claude/skills/plan/steps/design-decisions.md

Phase 6: Write Plan

→ Read and follow: ~/.claude/skills/plan/steps/write-plan.md

Phase 7: Spawn Critique

→ Read and follow: ~/.claude/skills/plan/steps/spawn-critique.md

Live Context

!git diff --stat 2>/dev/null !git log --oneline -5 2>/dev/null

Protocols

!cat ~/.claude/skills/_shared/review.md

- TASK scope = route to /task, no planning overhead - MINI = no subagents, auggie-only discovery - FULL = spawn plan-explorer, full workflow - Original goal is user's EXACT words, never reworded - No detailed task specs — that's /decompose - Critique is MANDATORY — no plan exits without APPROVED - APPROVED → /decompose, FIX_AND_SHIP → /revise