planning_skill

Overview

This skill produces interactive, detailed implementation plans through iterative research and collaboration. It emphasizes reading code and documentation fully, surfacing assumptions, and structuring phased work with explicit success criteria. The output is a machine- and human-readable plan file that supports review and optional commit integration.

How this skill works

The skill first gathers full context by reading all referenced files and spawning targeted research tasks (codebase locators, analyzers, pattern finders). It presents findings, solicits preferences and approvals via structured AskUserQuestion interactions, and builds a phased plan following a strict template. Each phase includes automated and manual success criteria and the plan is written to a time-stamped path for review and iteration.

When to use it

• When you invoke /create-plan to scope or spec an implementation
• When another agent requests a planning sub-skill for a feature
• When you need a phased, testable technical plan with clear verification steps
• When you want a plan that integrates with file-review and commit workflows

Best practices

• Establish user preferences and autonomy level before deep work
• Always read all referenced files completely before deciding or spawning sub-tasks
• Prefer small, incremental phases with clear automated checks and manual steps
• Use AskUserQuestion for every clarification, design choice, or commit decision
• Include file:line references for important findings and assumptions

Example use cases

• Create a migration plan for a shared TypeScript library with phased compatibility checks
• Design an implementation plan for a new agent feature, including tests and review steps
• Break down a large refactor into verifiable phases with commit points after manual validation
• Prepare a spec that integrates with a file-review plugin for inline feedback