Multi Phase Plan is a skill published by agensi that takes finalized product requirements and design documents and produces a structured, executable multiphase implementation plan packaged for AI coding agents such as Cursor, Windsurf, or Bolt.new. Rather than organizing work into horizontal technical layers, it structures the build as vertical slices — each phase delivering a fully functional, testable feature rather than an isolated layer like a database schema or an API stub. The output is a downloadable multiphase-plan/ directory containing a master tasklist roadmap, a phase-specific test plan, implementation rules, and a series of sequentially ordered markdown prompts. Each prompt includes context, scope, and completion criteria, so an AI coding agent can consume it directly without additional prompt engineering. The skill enforces a no-mock-data rule, directing agents to build real data boundaries and error handling from the start rather than generating placeholder UIs. It integrates test-driven development by specifying the exact behaviors and states that must be validated before a phase is considered complete. The master tasklist is designed to track overall progress without flooding an LLM’s context window with irrelevant detail. The skill works from business model documents, design docs, and UX flows, so it requires those inputs to be finalized before use. There are no environment variables to configure and no runtime transport — it is invoked as a skill and returns a file package.

Use cases

Convert a finalized PRD and design system into a sequenced vertical-slice build plan for an AI coding agent

Generate ready-to-use phase markdown prompts with scope and completion criteria for Cursor or Windsurf

Enforce a no-mock-data policy across all implementation phases by embedding real data boundary rules in each prompt

Define per-phase test plans that specify exactly which behaviors must pass before moving to the next phase

Produce a master tasklist roadmap that an agent can check off without exceeding context window limits

Translate UX flow documentation into discrete, testable feature milestones rather than technical layer tasks

When to use it

Product requirements and design documents are fully finalized and ready to be handed off to an AI coding agent

The project should be delivered as incremental, demo-able features rather than layered technical infrastructure

The team wants pre-written markdown prompts instead of crafting per-phase agent instructions manually

Test-driven development checkpoints are needed to gate phase completion

When not to use it

Requirements are still exploratory or incomplete — the skill requires finalized inputs

The target coding environment is not an AI agent tool; it is designed specifically for agents like Cursor, Windsurf, or Bolt.new

A raw Kanban or ticket-based task list is needed rather than a packaged prompt directory

The project has no design docs or UX flows to analyze, as those are required inputs