create-plan_skill

Overview

This skill creates a single, executable plan for a coding task that you can run without extra interpretation. It bounds scope, lists ordered atomic actions, includes explicit validation commands, and surfaces risks and open questions. The plan follows strict read-only discovery and prefers project-local profile data when available.

How this skill works

I inspect project files and the project profile (REPO_PROFILE.json) to determine stack shape, canonical paths, and validation commands. Then I produce a compact plan: scope, 6–10 verb-first action items, runnable validation steps, risks, and up to three open questions. If critical data is missing, I record it as unknown rather than guessing and proceed with stated assumptions.

When to use it

• You need a clear, step-by-step executable plan for a multi-step coding task.
• You want validation commands and explicit checks included with the plan.
• You require minimal ambiguity and want open questions surfaced before execution.
• You need the plan to respect local project conventions and profile data.
• You want a plan that can be executed by engineers or automation without extra interpretation.

Best practices

• Run the plan in a separate feature branch and keep steps atomic for easy review.
• Provide the project profile (REPO_PROFILE.json) to make commands authoritative.
• If a blocking unknown exists, answer at most one short clarifying question before proceeding.
• Include at least one fast lint/typecheck step and one tests-based correctness check.
• Record assumptions clearly when project-local values are unavailable.

Example use cases

• Add a new API endpoint with clear discovery, implementation, test, and rollout steps.
• Implement a cross-cutting refactor that requires coordinated validation and migration.
• Add a CI gate and associated quality checks following the project's profile conventions.
• Create a small feature with explicit commands for linting, testing, and build verification.
• Triaging a task where project conventions are partially unknown and must be documented.