What problem does it solve? Running a full multi-perspective review of a development plan normally means answering 15-30 intermediate questions across separate CEO, design, DX, and engineering review passes. This Skill automates that gauntlet: it reads each review skill from disk, runs them sequentially, auto-decides routine questions using 6 decision principles, and surfaces only genuine taste decisions at one final approval gate. ## Core Features & Use Cases - Sequential multi-phase review: Runs CEO (strategy/scope), design (7 dimensions, only when UI scope exists), DX (8 dimensions, only when developer-facing scope exists), and engineering (architecture, tests, security) reviews in order, each with dual voices from a Claude subagent and Codex. - Auto-decision engine: Applies 6 decision principles to resolve routine questions automatically, marking borderline calls as TASTE DECISION and agreed scope changes as USER CHALLENGE for the final gate. - Aggregated task output: Collects per-phase implementation task lists from JSONL files, dedupes them, and renders a prioritized task list plus TODOS.md updates at the Final Approval Gate. - Use Case: You have a draft implementation plan and want the full review gauntlet without babysitting it. Invoke autoplan, and it produces a fully reviewed plan with consensus tables, a test plan artifact, and one consolidated approval prompt. ## Quick Start Ask the assistant to run autoplan on your plan file to execute all review phases automatically and present the final approval gate.
