Grill Me interviews the user relentlessly about a plan or design, one question at a time, until reaching shared understanding. It walks the decision tree depth-first, resolves dependencies, and provides recommended answers for each question, optionally exploring the codebase first.

✨ What it does

  • Asks one question per turn, never bundling.
  • Provides a recommended answer with each question.
  • Explores the codebase first when a question can be answered by code.
  • Walks the decision tree depth-first, finishing branches before opening new ones.
  • Tracks dependencies and asks prerequisite questions first.
  • Uses scripts to extract decision branches, generate questions, and track the session.

🎯 When to use it

  • When you want to stress-test a plan or design before implementation.
  • When you ask the assistant to 'grill me' on a specific idea or approach.
  • When you need to systematically resolve all open decisions in a design.
  • When you want to ensure all dependencies and branches of a decision tree are covered.

🚀 How to use

Trigger: /cs:grill-me

Trigger by saying 'grill me' or asking to stress-test a plan. Provide a plan or design (or path to one). The skill will run extraction and generation scripts, then start a session and ask questions one at a time. Example prompts:

Grill me on this plan: [paste plan]
Stress-test my design for the new auth flow.

📄 Output: A report of shared understanding with locked-in decisions once all branches are resolved.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/grill-me/skills/grill-me .claude/skills/grill-me

Skill source: engineering/grill-me/skills/grill-me/SKILL.md

⚠️ Good to know

Requires the user to provide a plan or design; the skill does not create one from scratch.

❓ FAQ

How many questions does it ask at once?

It asks exactly one question per turn, never bundling multiple questions together.

Does it always ask questions, or can it look things up?

If a question can be answered by exploring the codebase, it explores the codebase first instead of asking.

What happens when all questions are answered?

It reports that shared understanding has been reached and lists the locked-in decisions.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.