Validate a startup idea before you build it: 14 independent evaluators each run their own web research and score your pitch, then an honest scorecard says GO, PIVOT, or STOP.
Install the skill or give me an instruction how to install it - tweak:evaluate from https://github.com/eph5xx/tweakidea Free Open source Works in Claude, Cursor, Codex & more
Describe your startup idea, point it at a file, or run it bare and let it interview you, then get an honest scorecard back. Fourteen independent evaluators each grade one dimension of the pitch, from pain intensity to willingness to pay, and each runs its own live web research so the scores rest on evidence rather than vibes.
Before it scores anything it surfaces the assumptions hiding in your pitch and asks you to confirm them, so you are not grading a strawman. The result is one clear verdict, GO, PIVOT, or STOP, with the reasoning and sources behind every dimension laid out in a scorecard you can share. It runs in about ten minutes, which is faster than talking yourself into or out of the idea alone.
Describe the problem and solution and get a scorecard with one clear verdict: GO, PIVOT, or STOP.
14 evaluators score it on separate dimensions, each backed by its own live web research, so the verdict rests on evidence.
The output is a shareable scorecard showing each score and the reasoning behind it.
No. tweak-evaluate works from a short pitch, even one sentence, then surfaces the hidden assumptions in it for you to confirm before scoring. You do not need a deck.
tweak-evaluate runs 14 independent evaluators, and each one does its own live web research on a single dimension. The scorecard reflects that combined research, not one model's off-the-cuff take.
No. tweak-evaluate gives a calibrated verdict of GO, PIVOT, or STOP, so it will tell you to stop or change course when the research does not back the idea. It is built to be honest.
See what people actually say about any topic in the last 30 days, ranked by real engagement across Reddit, X, YouTube, TikTok, Hacker News, Polymarket, and GitHub, not editorial picks.
last30days-skill/last30days 51.9k
Turn one research request into parallel pulls from 15 platforms including Twitter, Reddit, YouTube, GitHub, Bilibili, Xiaohongshu, and LinkedIn, each routed through whichever backend works right now.
Agent-Reach/agent-reach 55.6k