grill-me

Grill me

Use this skill to relentlessly interrogate a plan or design until every branch of the decision tree is resolved. Pure interrogation — no documentation side-effects.

Use this skill when

• The user wants to be grilled, stress-tested, or interviewed about a plan or design.
• The user explicitly says "grill me" and does not mention domain docs, CONTEXT.md, or ADRs.
• The user wants to think through a decision tree before committing to a direction.

Do not use this skill when

• The plan touches domain concepts that should be captured in CONTEXT.md or ADRs — route to grill-with-docs.
• The request is under-specified and needs sharpening before interrogation — route to reverse-prompt.
• A completed /plan needs formal reviewer approval — route to plan-review-loop.

Routing boundary

Situation	Use this skill?	Route instead
User says "grill me" about a plan, no domain-doc intent	Yes	-
User wants interrogation AND domain language / ADR maintenance	No	grill-with-docs
User's request is too vague to interrogate	No	reverse-prompt
Completed /plan needs Jason/Freddy review	No	plan-review-loop

Inputs to gather

Required before starting

• The plan, design, or decision the user wants stress-tested.

Helpful if present

• Prior conversation context about the feature.
• The repository codebase for verification of claims.

First move

1. Read the plan or design the user has presented.
2. Identify the first unresolved decision branch.
3. Ask the first question, providing your recommended answer using the ask_user tool

Workflow

1. Ask one question at a time. Walk down each branch of the design tree, resolving dependencies between decisions sequentially. For each question, provide your recommended answer.
2. Wait for feedback on each question before continuing to the next.
3. Explore instead of asking when a question can be answered by examining the codebase.
4. Repeat until all branches are resolved and you have reached a shared understanding.

Outputs

• A shared understanding of the plan or design, with all decision branches resolved.

Guardrails

• Ask questions one at a time. Never dump a wall of questions.
• Provide your recommended answer with every question.
• Explore the codebase when it can answer a question directly — do not ask the user what the code already tells you.
• Do not create or update any documentation files (CONTEXT.md, ADRs). If domain-doc maintenance is needed, tell the user to use grill-with-docs instead.

Validation

• Confirm the skill activates on "grill me" prompts without domain-doc intent.
• Confirm it routes to grill-with-docs when the user mentions CONTEXT.md, ADRs, or domain language maintenance.
• Run node skills/skill-authoring/scripts/validate-skill-library.mjs skills/grill-me/SKILL.md.
• Smoke test:
  • should trigger: "Grill me on my plan to refactor the notification service."
  • should not trigger: "Grill me on this design and build up CONTEXT.md as we go." (→ grill-with-docs)

Examples

• "Grill me on this plan to split the monolith into two services."
• "Stress-test my design for the new caching layer — I want to make sure I haven't missed anything."
• "Interview me about this migration approach before I start implementing."

Reference files

• references/interrogation-patterns.md — lightweight interrogation patterns: question sequencing, when to explore vs. ask, session completion criteria