A technical alignment skill designed to drill down into requirements, architecture, and business rules before code implementation begins.
⚡ Quick Install
npx skills add https://github.com/kipperdev/skillpper --skill grill-me
🎯 Why Does the /grill-me Protocol Exist?
When starting a major feature, refactoring, or new system architecture, unspoken assumptions frequently cause rework. AI coding assistants typically fall into one of two traps:
- Blind Assumptions: The agent guesses architecture or library choices on its own, building solutions misaligned with the developer's tech stack.
- Text Slop in Chat: The agent prints long lists of questions in conversational prose (
❓ Question 1,➡️ Options:), forcing the user to type out tedious paragraphs to answer.
The /grill-me protocol uses the question interface available in the current agent. Ask focused multiple-choice questions when supported, then refine the decision tree with the user's answers.
🌐 Cross-Platform Support
This skill works on macOS, Linux, and Windows. It has no operating-system-specific scripts or paths. Use the active agent's question interface and respect its capabilities, question limits, and available response formats. See the interview guide.
⚙️ Mandatory Execution Guidelines
1. Zero Raw Text in Chat
- Prefer Native Input: Use the current agent's interactive question interface when one is available.
- Use a Clear Fallback: If the host has no interactive question interface, ask concise questions in the conversation. Do not claim to have opened a modal that is unavailable.
2. Interview in Bounded Rounds
- Build a decision tree and ask only questions whose prerequisites are settled.
- Respect the host's per-call question limit. A round may contain fewer questions; cover additional branches in later rounds only when the user's answers make them relevant.
- Prioritize the choices that materially change the implementation:
- Scope & Boundaries: What is included in the MVP and what is explicitly excluded.
- Architecture & Libraries: Target frameworks, persistence engines, and design patterns.
- Interface & UX: Visual layout behavior, loading states, and responsive viewports.
- Error Handling & Edge Cases: Failure modes, permissions, and security constraints.
- Testing & Delivery Strategy: Target test suites and verification criteria.
3. Option Formatting
- First-Person Voice: Formulate every option as the user's direct response (e.g., "Use SQLite in WAL mode" rather than "The agent should configure...").
- Recommended Option First: The technically optimal choice recommended by the agent must always appear first, prefixed with
(Recommended). - Deliberate Multi-Select: Use multi-select only when combining answers makes sense; use a single choice for mutually exclusive decisions.
- Respect Host Controls: Include a write-in option only when the interface does not provide one automatically.
4. Transition to Execution
- Wait for the user to answer decisions that materially affect implementation before committing to those choices. Continue independent inspection that does not depend on the answers.