plan-harder_skill

Overview

This skill creates detailed, phased implementation plans when a user explicitly asks to "plan harder." It breaks work into research, clarification, sprinted implementation, testing, and review, producing atomic, testable tasks and a single markdown plan file. The output is focused on runnable increments, validation criteria, and known gotchas.

How this skill works

The skill inspects the request and the codebase context, runs a short research phase to map architecture and dependencies, and asks targeted clarification questions to remove ambiguity. It then generates a multi-sprint plan with atomic tasks, file paths, dependencies, and validation steps, saves the plan to a kebab-case filename with a -plan.md suffix, and lists potential risks and a rollback approach.

When to use it

• When you explicitly instruct the agent to "plan harder" for a bug, feature, or task
• When you need a phased implementation with demoable increments and tests
• When you want atomic, committable tasks with clear file paths and acceptance criteria
• Before implementation to identify dependencies, risks, and validation steps
• When handing work to multiple engineers or preparing a review-ready plan

Best practices

• Answer up to 10 targeted clarification questions promptly to keep scope tight
• Provide repository access or a code snapshot so research can be accurate
• Prefer small, independent tasks that can be completed in a single commit
• Include expected file paths and test commands for each task to ensure verifiability
• Review the generated plan and run the suggested demo/validation checklists before coding

Example use cases

• Planning a multi-sprint rollout for a new feature with backend and frontend components
• Breaking a complex bugfix into atomic steps that are independently testable
• Creating onboarding tasks for a new engineer to implement a subsystem safely
• Preparing a migration or refactor with clear rollback and verification steps
• Generating a test-first implementation plan with explicit test cases per task