plan-phases_skill

Overview

This skill helps create implementation plans broken into context-sized phases that sub-agents can execute. It enforces independent, verifiable phases, explicit dependencies, and phase sizing tuned to sub-agent context budgets. Use it to produce a clear, actionable plan document ready for review before implementation.

How this skill works

It asks targeted clarifying questions to establish scope, technical context, integrations, users, and constraints. It then applies phase-sizing heuristics (target 30–50k token work per phase), splits oversized phases by layer/entity/operation, and maps hard/soft/none dependencies. The output is a docs/plans/{feature-name}.md plan with phase specs, files to change, acceptance criteria, and an execution graph.

When to use it

• Creating an implementation plan for a new feature or system
• Breaking a large feature into sub-agent-executable phases
• Preparing phase documents before handing work to developer sub-agents
• When /workflow-plan-phases command is invoked
• Converting vague project goals into verifiable, testable phases

Best practices

• Always start by asking 3–5 focused clarifying questions before drafting a plan
• Target 30–50k tokens per phase: ~2–5 files and 1–2 related features
• Avoid sub-phases and use whole-numbered phases only
• Define interfaces or stubs early to minimize hard dependencies
• Write testable acceptance criteria and explicit out-of-scope items

Example use cases

• Plan a user authentication system split into model, auth service, and UI phases
• Break a monolithic feature into vertical slices so sub-agents can work in parallel
• Create a phased migration plan for a database schema and dependent services
• Prepare a rate-limiting feature plan with clear API changes and verification steps
• Draft an integration plan for an external API with stubs and final implementation phases