Grill Me: Product — a one-question-at-a-time idea interview

TL;DR

• The skill turns an agent into a product griller: it walks your idea down a decision tree, researches the facts it can find, and puts every decision back to you, one question at a time.
• Ten areas are covered in dependency order — problem, audience, market, competitors, positioning, acquisition, monetization, unit economics, risks, proof — with depth adapted to where your answer is weakest.
• The split is the point: facts are researched, decisions are yours. Market size, competitors, platform fees and benchmarks get looked up and cited.
• It ends with a compact product brief, and it does nothing to the idea without explicit confirmation.

What it is and who it is for

Grill Me: Product is a skill for AI coding agents — Claude Code, OpenAI agents and other skill-compatible runtimes. It is inspired by grill-me and grill-with-docs, both by Matt Pocock, re-targeted from engineering plans to product and business development.

It is for the person with an idea that feels obviously good and has never been asked who exactly pays for it. The README's diagnosis is that most product ideas die not from a bad build but from an unexamined assumption: the wrong audience, a channel that never materializes, unit economics that break at scale, a moat that is not there.

What it does

• One question at a time. No question dumps: ask, listen, reflect, next.
• Researches facts, defers decisions. Market size, competitor landscape, platform fees, pricing benchmarks and regulatory constraints are looked up and cited; the decisions — which segment, which channel, how to price — stay yours.
• Dependency-ordered roadmap. Ten areas in order, with depth adapted to where your answer is weakest.
• Pushes on vagueness. "Everyone" becomes a named segment, "cheap" becomes a number, "viral" becomes a concrete channel.
• Ends with a product brief: problem, segment, positioning, channel, model, unit economics, top risks and first experiment.
• No action until you confirm. The skill only interviews.

The ten areas are: problem and pain; audience and segment; market and timing with TAM, SAM and SOM researched; competitive landscape researched; differentiation and positioning; acquisition channels with benchmarked CAC; monetization model and price point; unit economics with CAC against LTV, gross margin and scale breaks; risks and fatal flaws; and proof, meaning the smallest real-signal experiment.

How it works

1. You trigger the skill on your idea.
2. The agent asks you to state the idea in one or two sentences.
3. The interview begins: for each area, it researches what it can and cites sources, then asks one decision question, recommends an answer with reasoning, and waits.
4. After each answer it reflects the implication and surfaces the next dependency: "if X is true, then we need to settle Y next."
5. It pushes on vagueness and turns generalities into named, numbered and specific answers.
6. On confirmation it outputs the compact product brief.

Sample session

The README includes a real exchange. The idea is a tool that summarizes meeting transcripts for product managers. The agent starts with the pain — what a PM actually feels today, and whether it is a pain they are aware of or one they have learned to tolerate — with a recommended answer about the gap between "the meeting happened" and "the decisions are captured and trackable". After the person confirms, the agent narrows the segment: product managers at early-stage startups, at mid-size SaaS companies and at enterprises have very different workflows and willingness to pay, so which slice is most desperate now. It recommends PMs at 20-200 person SaaS companies running many cross-functional meetings with no dedicated note-taker, promises to research that slice's market size next, then actually searches.

That exchange is the whole method: one question, a recommendation to react to, and research in between.

Install and customize

# Clone into your skills directory
git clone https://github.com/artificialguybr/grill-me-product.git

Then register the skill path with your agent runtime, or copy the grill-me-product/ folder into your existing skills collection. The structure is SKILL.md plus agents/openai.yaml — the manifest for OpenAI-agent-style runtimes, while other runtimes read the frontmatter at the top of SKILL.md.

Four things are meant to be edited: add or reorder areas in the list under "Areas to cover" to match your domain, make the tone gentler or harsher in "Session conduct", change the final brief to a Lean Canvas or a YC application shape, and tighten or loosen the cite-sources requirement in "Facts vs. decisions".

When to use it, when not to

Use it before you build: when an idea has been in your head long enough to feel obvious, when someone asks what the business model is and you answer with adjectives, or when you are about to spend a month on a build with no evidence anyone wants it.

Do not use it as market research. The agent researches and cites, but the README is clear that the decisions stay yours — it interviews and never acts on the idea without explicit confirmation. And do not use it if you want a verdict; you get questions and a brief, and the yes or no is still yours.

Sources

1. Grill Me: Product — repository — features, the ten areas, session flow, sample session, customization and compatibility quoted above.
2. README.md — install command, structure, license (MIT), acknowledgements.

Project page: Grill Me: Product

The skill credits grill-me and grill-with-docs by Matt Pocock, and is MIT-licensed.

FAQ

How many questions does it ask at once? One. The README makes one question at a time the first feature: ask, listen, reflect, next.

Does the agent decide the strategy for me? No. Facts are researched and cited; decisions — segment, channel, price — stay yours, and it recommends an answer with reasoning rather than choosing it.

Does it act on the idea? No. The skill only interviews, and it does nothing without explicit confirmation.

Can I change the areas it covers? Yes. The README documents editing the numbered list under "Areas to cover", plus tone, output format and research depth.