G

grill-me

Verified

by mattpocock · Updated Apr 16, 2026

opus
548in2.4Kout
$0.282m 3s

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

planningdesign-reviewinterviewingrequirementsdecision-making

See It In Action

1 turns·claude
The project codebase is in your current working directory. Explore the files to understand the code before making changes. engineering review’s this afternoon — grill my per-workspace API rate limit rollout plan against the current code and give me a resolved design review

Let me explore the codebase first, then grill you on your plan.

Good, I have a thorough understanding of the codebase. Now let me grill you on your plan.

Let me read the key files first to ground this properly.

Good. I've read every file. Let's start with the question that will dominate your eng review.


Question 1: Your enterprise limit doesn't solve the incident.

Your plan proposes enterprise tier at 2,000 req/min (burstable to 3,000). Acme Corp generated 12,000 req/min during the April 9 outage. That's 4x your proposed burst cap.

If this plan had been live on April 9, Acme would have been rate-limited to 3,000 req/min. Their quarterly sync would have taken roughly 4x longer (possibly 3+ hours instead of ~30 minutes), their retry logic would have been hammering 429s the entire time, and your $2.1M ARR customer's CSM would have been in your Slack within the hour.

How do you reconcile the enterprise limit with the actual traffic pattern that caused the incident?

My recommendation: You need a mechanism for temporary burst allocation or custom per-workspace overrides that sit above the tier defaults. The tier config is the floor, not the ceiling. Acme should get a custom limit (e.g., 15,000 req/min) that reflects their actual usage, possibly negotiated into their contract. The plan needs to distinguish between "tier defaults" and "workspace-specific overrides."

Claude Sonnet 4.6

Permissions

ScopeDescription
filesystem:read

SKILL.md

Interview me relentlessly about every aspect of this plan until we reach a shared understanding. Walk down each branch of the design tree, resolving dependencies between decisions one-by-one. For each question, provide your recommended answer.

Ask the questions one at a time.

If a question can be answered by exploring the codebase, explore the codebase instead.

FAQ

What does grill-me do?

Interview the user relentlessly about a plan or design until reaching shared understanding, resolving each branch of the decision tree. Use when user wants to stress-test a plan, get grilled on their design, or mentions "grill me".

When should I use grill-me?

Use it when you need a repeatable workflow that produces text response.

What does grill-me output?

In the evaluated run it produced text response.

How do I install or invoke grill-me?

Ask the agent to use this skill when the task matches its documented workflow.

Which agents does grill-me support?

Agent support is inferred from the source, but not explicitly declared.

What tools, channels, or permissions does grill-me need?

It uses no extra tools; channels commonly include text; permissions include filesystem:read.

Is grill-me safe to install?

Static analysis marked this skill as low risk; review side effects and permissions before enabling it.

How is grill-me different from an MCP or plugin?

A skill packages instructions and workflow conventions; tools, MCP servers, and plugins are dependencies the skill may call during execution.

Does grill-me outperform not using a skill?

About grill-me

When to use grill-me

When you want an agent to stress-test an architecture or implementation plan through structured questioning. When a design has unclear dependencies or unresolved tradeoffs that need to be surfaced one by one. When the agent should inspect the local codebase to answer factual questions before asking the user.

When grill-me is not the right choice

When you need the agent to implement, deploy, or integrate with external services rather than facilitate design clarification. When the task does not involve a plan, design, or decision tree that benefits from iterative questioning.

What it produces

Produces text response.