workflow-plan-phases_skill

Overview

This skill creates a structured implementation plan document that divides a feature into properly sized phases optimized for context-efficient sub-agent execution. It produces a markdown file under docs/plans/, with whole-number phases, clear acceptance criteria, dependency mapping, and execution recommendations. The skill only plans — it does not implement any code or run phases.

How this skill works

It first prompts targeted clarifying questions to lock scope, constraints, and technical context. After answers, it sizes phases to target 30–50k token sub-agent budgets, defines objectives, files to change, acceptance criteria, and dependencies, then outputs a markdown plan saved to docs/plans/{slugified-name}.md. Finally, it presents the plan and asks if changes are needed; it never implements phases.

When to use it

• When you need an implementation plan that sub-agents can run independently
• When a feature is large and must be split to respect LLM context limits
• When you want explicit acceptance criteria and dependency mapping
• Before handing work to automation or distributed developer agents
• When you must ensure phases are verifiable and non-overlapping

Best practices

• Always ask 3–5 clarifying questions and wait for answers before planning
• Target each phase to 30–50k tokens (approx. 2–5 files, 1–2 related features)
• Use whole-number phases only; avoid sub-phases like 1.1 or 2.a
• Write testable acceptance criteria (HTTP responses, behaviors, or checks)
• Minimize dependencies: define interfaces early, stub or use feature flags to enable parallel work

Example use cases

• Plan a user authentication system split into model, auth service, session management, password reset, and UI phases
• Create API feature rollout where schema, endpoints, tests, and docs are separate phases
• Break a monolith refactor into safe, verifiable vertical slices sized for sub-agents
• Prepare a compliance-focused feature with explicit security and testing acceptance criteria
• Draft a multi-service integration plan with clear hard vs soft dependency mapping