think-it-through

Grill vague ideas into dependency-aware work-items through one-question-at-a-time planning interviews.

5|1|Updated Mar 7, 2026
One-click install
npx skills add https://github.com/ai-hero/hero-skills --skill think-it-through-ai-hero
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: think-it-through
Source: https://github.com/ai-hero/hero-skills/tree/main/skills/think-it-through
Command: npx skills add https://github.com/ai-hero/hero-skills --skill think-it-through-ai-hero

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Vague feature requests and unexamined assumptions cause wasted engineering work. This Skill turns a rough idea or ambiguous task into explicit shared understanding before any code is written, capturing the result as structured, dependency-aware work-items. ## Core Features & Use Cases - Structured grilling interview: Asks one question at a time, each with a recommended answer, walking a design tree until goals, non-goals, failure modes, reversibility, and success criteria are all explicit. - Work-item store: Writes approved plans as markdown files in a private, git-ignored .plans/ directory with depends_on ordering, status lifecycle, and Definition of Done checklists. - Feature and roadmap modes: Re-plans existing store items in place after verifying their premises against the codebase, or plans a set of related items in one shared pass. - Use Case: You receive a ticket saying "add device-flow login." The Skill interviews you about scope, edge cases, and reversibility, then emits ordered work-items with measurable success criteria ready for implementation. ## Quick Start Ask the assistant to think through your feature idea or task, for example by saying "think it through: add OAuth device-flow login to the CLI."

Frequently Asked Questions about think-it-through

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I plan a vague feature request before writing code?

Invoke the skill with the idea or task as an argument. It interviews you one question at a time, always proposing a recommended answer, until goals, non-goals, failure modes, and success criteria are explicit, then writes work-items to a .plans directory.

What is a dependency-aware work-item store for planning?

It is a git-ignored .plans folder of markdown files where each item has an id, status, and depends_on list of blocking ids. Readiness is derived by checking that every dependency is done, so blocked work is never started prematurely.

When should I skip the grilling interview for a task?

Skip it for typos, copy tweaks, dependency bumps, and small changes with one obvious approach in one area. The skill itself writes a one-line approach and moves on when a full grill would produce a plan nobody reads.

Does the planning skill write or commit code automatically?

No. It only produces understanding and work-items; implementation is delegated to the separate one-shot skill. The .plans store is private and git-ignored, and items only become buildable after the user explicitly marks them ready.

Why does planning ask one question at a time instead of batching?

Batched questions destroy the dependency order between decisions, since an early answer reshapes every later question. Single questions with recommended answers keep the design tree ordered and surface disagreement immediately.