Grill Me
Grill Me is a relentless interviewing skill that stress-tests your plans, designs, and decisions through systematic questioning. Instead of telling you what's wrong, it asks questions — one at a time — until every assumption is surfaced and every decision tree is fully explored.
206,000+ installs. By Matt Pocock (opens in a new tab) (TypeScript educator, 104K GitHub stars).
Install:
npx skills add https://github.com/mattpocock/skills --skill grill-meThis skill is hosted on skills.sh (opens in a new tab), a separate registry from ClawHub. It has passed Gen Agent Trust Hub, Socket, and Snyk security audits.
What it does
Grill Me interviews you about every aspect of your plan. It works through each branch of your design tree — resolving dependencies between decisions one by one — until it reaches shared understanding of your intent.
For each question, it also provides a recommended answer. If the answer is already visible in your codebase or files, it explores there first rather than asking you redundantly.
When to use it
| Situation | What Grill Me does |
|---|---|
| Before building a new feature | Exposes missing requirements and hidden assumptions |
| Before a technical decision | Tests whether you've considered the tradeoffs |
| Architecture review | Walks every branch of your design until nothing is ambiguous |
| Pre-implementation planning | Surfaces blockers before you write a line of code |
| Presenting a plan to a team | Prepares you for every question you'll be asked |
Basic usage
Activate it by invoking the skill directly or just saying "grill me":
Grill me on this plan: [describe your plan]I want to add a caching layer to the API. Grill me.Here's my architecture for the new agent pipeline. Grill me on it before I start building.
[paste architecture doc]The skill asks questions one at a time. Answer each one. It won't move to the next branch until the current one is resolved.
Example session
You: I want to build a persistent research agent that works through a TASKS.md file on a cron schedule. Grill me.
Grill Me: What happens if two cron runs overlap — if run N is still working when run N+1 triggers?
Recommended answer: Use a named session (--session session:research) so the second run queues rather than spawning a parallel session. Add a max runtime check.
You: Good point — I'll use a named session.
Grill Me: How does the agent mark a task as "in progress" before it starts, so a restart doesn't re-claim a task that's already underway?
Recommended answer: Change [ ] to [~] at the start of each task before working it. If the agent crashes mid-task, [~] acts as a signal that the task was interrupted and needs human review.It keeps going until every decision is resolved or you end the session.
Why it's valuable
Most planning fails not because the plan is wrong but because the assumptions behind it were never examined. Grill Me makes the assumptions visible before they become bugs.
It's especially useful before building with agents — where a missed edge case (what if the tool call fails? what if the session times out?) can cause silent failures hours later in a cron job nobody's watching.
Run Grill Me before setting up any new standing task or persistent agent workflow. Five minutes of being grilled saves hours of debugging a cron job that silently failed at 3 AM.