diff --git a/Skills/Codex/grill-me/SKILL.md b/Skills/Codex/grill-me/SKILL.md
new file mode 100644
index 0000000..cd88705
--- /dev/null
+++ b/Skills/Codex/grill-me/SKILL.md
@@ -0,0 +1,53 @@
+---
+name: grill-me
+description: Interview the user relentlessly about a plan, design, architecture, proposal, feature idea, product decision, or implementation approach until shared understanding is reached. Use when the user asks to be grilled, stress-test a plan, pressure-test a design, find holes in an approach, or resolve an ambiguous decision tree before execution.
+---
+
+# Grill Me
+
+Use this skill to pressure-test the user's thinking through focused interrogation. The goal is not to debate for sport; it is to uncover assumptions, dependencies, missing requirements, tradeoffs, and unresolved branches until the plan is crisp enough to act on.
+
+## Workflow
+
+1. Identify the plan, design, or decision being examined.
+2. Build a mental decision tree of the unresolved questions.
+3. Ask exactly one question at a time.
+4. For each question, provide your recommended answer or default direction after the question.
+5. If a question can be answered by exploring available files, code, docs, or tools, investigate instead of asking the user.
+6. Use the user's answer to choose the next branch of the decision tree.
+7. Continue until the important branches are resolved and the remaining uncertainty is explicit.
+
+## Question Style
+
+Keep questions direct, specific, and hard to dodge. Prefer questions that force a concrete tradeoff, acceptance criterion, owner, constraint, fallback, or success metric.
+
+Avoid asking several questions at once. If multiple issues are related, pick the blocking one first and explain why it matters.
+
+After each question, add a concise recommendation:
+
+```text
+My recommended answer: ...
+```
+
+## Investigation Before Asking
+
+When the answer may already exist in the workspace, inspect it directly. Examples:
+
+- Read relevant code before asking about current architecture.
+- Search tests before asking what behavior is covered.
+- Check package scripts before asking how verification works.
+- Inspect docs before asking about documented constraints.
+
+Only ask the user when the answer depends on intent, priority, business context, or information unavailable from the workspace.
+
+## Exit Criteria
+
+Stop grilling when:
+
+- The plan has clear goals, non-goals, constraints, and success criteria.
+- Major alternatives and tradeoffs have been considered.
+- Dependencies and sequencing are understood.
+- Risks and verification steps are named.
+- The user has enough clarity to decide, implement, or write a plan.
+
+End with a short synthesis of the resolved decisions, open risks, and recommended next action.
diff --git a/Skills/Codex/grill-me/agents/openai.yaml b/Skills/Codex/grill-me/agents/openai.yaml
new file mode 100644
index 0000000..d0168f4
--- /dev/null
+++ b/Skills/Codex/grill-me/agents/openai.yaml
@@ -0,0 +1,4 @@
+interface:
+  display_name: "Grill Me"
+  short_description: "Stress-test plans with focused, relentless questions."
+  default_prompt: "Use $grill-me to stress-test my plan one decision at a time."