# Grill Me: Product
**A relentless interview skill that stress-tests your product idea — market, audience, positioning, acquisition, monetization, and unit economics — through one sharp question at a time.**
[Features](#features) · [Why](#why-this-exists) · [Install](#install) · [How it works](#how-it-works) · [Sample session](#sample-session) · [Customize](#customize)
---
A [skill](https://github.com/mattpocock/skills) for AI coding agents (Claude Code, OpenAI agents, and other skill-compatible runtimes). It turns the agent into a **product griller**: an interviewer that walks your idea down its decision tree, researches the facts it can find, and puts every decision back to you — one question at a time — until the idea is sharp enough to build.
Inspired by [`grill-me`](https://github.com/mattpocock/skills/tree/main/skills/productivity/grill-me) and [`grill-with-docs`](https://github.com/mattpocock/skills/tree/main/skills/engineering/grill-with-docs), but re-targeted from **engineering plans** to **product & business development**.
## Features
- **One question at a time.** No bewildering question dumps. Ask → listen → reflect → next.
- **Researches facts, defers decisions.** Market size, competitor landscape, platform fees, pricing benchmarks, regulatory constraints — the agent looks these up and cites sources. The *decisions* (which segment, which channel, how to price) stay yours.
- **Dependency-ordered roadmap.** Works through ten areas in order, adapting depth to where your answer is weakest:
1. Problem & pain
2. Audience & segment
3. Market & timing (TAM/SAM/SOM, researched)
4. Competitive landscape (researched)
5. Differentiation & positioning
6. Acquisition channels (benchmarked CAC)
7. Monetization model & price point
8. Unit economics (CAC vs LTV, gross margin, scale breaks)
9. Risks & fatal flaws
10. Proof & next step (smallest real-signal experiment)
- **Pushes on vagueness.** "Everyone", "cheap", "viral" get pushed to specific, numbered, and named.
- **Ends with a product brief.** Problem, segment, positioning, channel, model, unit economics, top risks, first experiment — in a compact summary.
- **No action until you confirm.** The skill only interviews; it never acts on the idea without explicit confirmation.
## Why this exists
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 isn't. A grilling interview surfaces those assumptions *before* you invest — and an agent that can research the world (market data, competitors, benchmarks) while interrogating you is sharper than a static checklist.
This skill applies the relentless-interview pattern from Matt Pocock's `grill-me` to the product domain: the agent is your skeptical co-founder who happens to have web search.
## Install
### As a skill (Claude Code, OpenAI agents, compatible runtimes)
```bash
# 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.
### Structure
```
grill-me-product/
├── SKILL.md # The grilling prompt — product-focused
└── agents/
└── openai.yaml # Agent manifest (display name, invocation policy)
```
`SKILL.md` is portable across skill-compatible agent runtimes. `agents/openai.yaml` provides the manifest for OpenAI-agent-style runtimes; other runtimes read the frontmatter at the top of `SKILL.md`.
## How it works
1. **You trigger the skill** (e.g. `/grilling` on your idea, or by name once registered).
2. **The agent asks you to state the idea** in 1–2 sentences.
3. **The interview begins.** For each area in the roadmap:
- It researches what it can (market size, competitors, benchmarks, fees) and cites sources.
- It 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** — turns "everyone" into a named segment, "cheap" into a number, "viral" into a concrete channel.
6. **On confirmation**, it outputs a compact **product brief**: problem, segment, positioning, channel, model, unit economics, top risks, first experiment.
Throughout, **facts are researched; decisions are yours.**
## Sample session
> **You:** I want to build a tool that summarizes meeting transcripts for product managers.
>
> **Agent:** Let's start with the pain. What concrete problem does a product manager feel today when dealing with meeting transcripts? Is this a pain they're acutely aware of, or one they've learned to tolerate? *Recommended answer:* The pain is the gap between "the meeting happened" and "the decisions are captured and trackable" — PMs tolerate it by manually re-reading or re-listening, which is slow and lossy. Yours to confirm.
>
> **You:** Yeah, exactly — they re-read and miss things.
>
> **Agent:** Then the next decision is the segment. "Product managers" is broad — PMs at early-stage startups, at mid-size SaaS companies, and at enterprises have very different workflows and willingness to pay. Which slice is most desperate for this *now*? *Recommended answer:* PMs at 20–200-person SaaS companies who run many cross-functional meetings and have no dedicated note-taker. I'll research the market size for that slice next.
>
> *(agent searches, cites TAM/SAM, continues…)*
## Customize
- **Add or reorder areas.** Edit the numbered list in `SKILL.md` under "Areas to cover" to match your domain (e.g. drop "Regulatory" for consumer apps, add "Supply/demand dynamics" for marketplaces).
- **Tone.** Adjust "pushes on vagueness" to be gentler or harsher by editing the "Session conduct" section.
- **Output format.** Change the final brief structure in the "Session conduct" section to match your workflow (e.g. Lean Canvas, business model canvas, YC application fields).
- **Research depth.** Tighten or loosen the cite-sources requirement in the "Facts vs. decisions" section.
## Compatibility
Tested with skill-compatible agent runtimes that read `SKILL.md` frontmatter. The `agents/openai.yaml` manifest targets OpenAI-style runtimes; the frontmatter (`name`, `description`, `disable-model-invocation: true`) is runtime-agnostic.
## License
MIT — see [LICENSE](LICENSE).
## Acknowledgements
Built on the relentless-interview pattern from [`grill-me`](https://github.com/mattpocock/skills/tree/main/skills/productivity/grill-me) and [`grill-with-docs`](https://github.com/mattpocock/skills/tree/main/skills/engineering/grill-with-docs) by [Matt Pocock](https://github.com/mattpocock). This variant re-targets the pattern from engineering plans to product & business development.
*Star if it sharpens your next idea.*