About grill-me
grill-me is a skill for clarifying a plan, feature, or idea before implementation. It guides a user through five adaptive questions, then produces a requirements summary for confirmation; it does not implement the proposal.
Key features
- 01Five sequential, adaptive questions
- 02Analysis of goals, rules, and risks
- 03Constructive challenge to weak reasoning
- 04Requirements summary with confirmation
- 05No implementation or skill invocation
Use cases
- 01Clarifying a feature before coding
- 02Surfacing edge cases and scope limits
- 03Agreeing on acceptance criteria
- 04Recording open decisions before implementation
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ sztadii/ agent-skills --skill grill-meAfter installing, type/grill-mein Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMEGrill Me
Scope
Discuss a plan, feature or idea with the user before implementation. Analyze the initial prompt and context, then ask five questions about business goals, logic and features, one at a time, so each question builds on the previous answers. Challenge weak reasoning directly and constructively, without arguing.
Use the user or caller's proposal and relevant conversation context; ask what to discuss if none is clear. Invocation authorizes the discussion and requirements summary only, not implementation, publishing or invocation of another skill.
Dependencies
No skill dependencies
Workflow
Analyze
- Read the initial prompt and relevant code and documentation. Identify the business goal, users, intended behavior, rules and features, and list what is unclear, contradictory or risky.
- Answer verifiable repository details yourself; never ask the user for them.
Ask five questions, one at a time
- Ask about the gap that most affects business outcome or behavior: who it is for, rules and edge cases users will hit, feature scope and exclusions, failure behavior and how success is observed.
- Send exactly one question per message, numbered (for example
Question 2/5), with a short reason, a concrete recommendation and meaningful alternatives. - Wait for the answer before asking the next. Use it to choose the next question: follow up on surprises or contradictions, drop questions it made obsolete. "Your call" accepts the recommendation; silence is not an answer.
- Always ask all five. When the major gaps are settled, ask about progressively smaller details of the agreed behavior instead of ending early.
Confirm
- After the last question, send the complete summary described in Result, labeled as awaiting confirmation, and ask whether the user accepts it.
- If the user requests changes, explain conflicts with earlier answers, then send the complete revised summary and ask again. Repeat until the user explicitly accepts the latest complete summary.
Result
After acceptance, send one self-contained message labeled Confirmed requirements — ready for implementation in the user's language, detailed enough for another skill or agent to act on without the discussion:
- Goal, users, intended behavior and agreed scope, including exclusions.
- Business rules and observable acceptance criteria.
- Key decisions with their reasons and tradeoffs.
- Relevant code/documentation references and remaining non-blocking unknowns, kept separate from accepted requirements.
Include only confirmed requirements. If the user abandons the discussion or a material decision remains open, label the summary provisional and state what is missing.
End with this instruction in the user's language: "To start implementation, request it in a separate message, either directly or by invoking an implementation skill." Then stop; do not start implementation.