Decide whether to build it.
A ready-to-use Claude Code workspace for taking a product idea from synthetic research to a launch plan. Clone it, open Claude Code inside, and start. It ships with two skills that chain: the first writes a research report, the second turns that report into a plan.
Build a small panel of made-up consumers for a product, capture how each one reacts in their own words, score their purchase intent (1–5), and write up an insights report — what blocks them, what wins them over, and what to do about it.
Each persona's reaction is written by its own subagent, in isolation, so the
personas don't copy each other and the panel stays realistic. Reports are saved
to reports/.
Based on Maier et al. (2025), arXiv:2510.08338. It's a quick, qualitative screen — no API keys, no Python. Treat the results as a directional first read, not a real consumer panel.
/consumer-personas <product or SaaS description>
You can add hints — segment, region, price tier, B2B/B2C, number of personas (default 6). If you leave them out, the skill assumes sensible defaults and tells you what it assumed.
Takes a research report from reports/ and writes a focused plan for getting the
product from zero to its first 100 customers — beachhead segment, MVP, how to
find customers by hand, pricing, and what would tell you to stop. Plans are saved
to plans/.
/startup-planner reports/<the-report>.md
Every choice in the plan traces back to a persona, objection, or value prop in the report.
No install step — the skills live in .claude/skills/, so they work the moment
you open Claude Code in this folder.
git clone https://github.com/sperelson/bellwhether.git
cd bellwhether
claudeThen run a skill, e.g. /consumer-personas <product>. Reports are saved to
reports/ and plans to plans/ (both kept out of git, so your work stays local).