Vets product ideas by collecting independent product boxes from multiple frontier AI models.
npx skills add https://github.com/line7works/tony-skills --skill jpb-line7works
Please help me install this Agent Skill. Skill: jpb Source: https://github.com/line7works/tony-skills/tree/main/plugins/jpb/skills/jpb Command: npx skills add https://github.com/line7works/tony-skills --skill jpb-line7works
/jpb resolve flow to record the human's verdicts and close the run. - Use Case: Before drafting a build plan, run the product box on a new app idea to see whether independent models converge on the same buyer and assumptions, then debate the flagged disagreements yourself. ## Quick Start Ask the assistant to run /jpb on your product idea or PRD file to collect independent product boxes from multiple AI models.High-intent search queries and answers about installing and using this skill.
Run /jpb with your idea text or a PRD file path. The skill scrubs out the working name, sends the identical brief to 3-6 isolated frontier models, and collects each one's filled product box verbatim into a single run doc.
A product box is a fixed template each model fills: a name, one-sentence pitch, single buyer, 3 things the product does, 9 distinctive properties tagged build/sell/run, and 3 load-bearing assumptions. A validator script rejects any box that deviates from the format.
Yes. All box and judge calls are dispatched through the readers plugin's roster, which owns model ids, transports, sandboxes, and parity settings. Without readers and the Workflow tool for the Claude rows, the run stops before any spend.
Only the OpenRouter-routed boxes (DeepSeek, Qwen) bill marginal dollars — roughly pennies for a short brief and tens of cents for a PRD-length one. The skill fetches actual billed USD per generation and records a per-box cost block in the run doc.
The approval step is a hard pause: it shows the scrubbed brief, the cost estimate, and the exact roster of calls, then waits for explicit approval. No box launches without it, and failed boxes are recorded rather than retried.
Skip it when the input is too thin to support 9 distinctive properties and 3 assumptions — the skill itself refuses bare one-liners and asks enrichment questions instead. It also never recommends; decisions remain a human debate step.
TECH-HYTransforms messy product notes and ideas into a structured discovery brief before PRD writing.
lwokerayConvert PRDs and user journeys into story maps with backbones and release slices.
m16khbReview plans and designs adversarially to identify flaws and complexity.
EvolutionAPIRun the same prompt through Claude, Codex, and Gemini in parallel and synthesize results.