noashavit/noas-persona-loop — explained in plain English
Analysis updated 2026-05-18
Build audience personas grounded in real analytics or engagement data instead of guesses.
Run a draft blog post, ad, or email past your personas before publishing it.
Score each persona's prediction against real results and correct its biases over time.
Judge product ideas or feature framing against how your actual audience has behaved.
| noashavit/noas-persona-loop | 0-bingwu-0/live-interpreter | 010zx00x1/faresnipe | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | — | Python | Python |
| Setup difficulty | easy | moderate | easy |
| Complexity | 2/5 | 2/5 | 2/5 |
| Audience | pm founder | general | general |
Figures from each repo's GitHub metadata at analysis time.
Requires Claude Code plus some real audience evidence, such as analytics exports, before personas can be built.
Noas Persona Loop is a set of Claude Code skills that build audience personas from real observed behavior, then use those personas to judge your work before you publish it. The pitch is that most marketing personas are made up from demographics and guesses, like a fictional Jordan who supposedly cares about developer experience with no actual evidence backing that claim. This project instead grounds every persona trait in real evidence you provide, such as analytics exports, spreadsheets, or screenshots of past engagement. The system works as a loop made of three skills. The first, persona build, interviews you and reads whatever evidence you have collected, then writes out persona files as markdown documents. The second, persona judge, runs a draft of whatever you are about to publish, whether that is a blog post, ad copy, email, or product idea, past each persona and produces a verdict along with a specific testable prediction about how it will perform. The third, persona learn, takes the real results once your content goes out into the world and scores each persona's earlier prediction against what actually happened, updating the persona file with any biases it discovers. Everything in the project is plain markdown files: agent files that define each persona, skill files that run the three step process, and data file templates where you paste in your own audience evidence. Nothing here is a hosted service or app, it all runs inside Claude Code on your own machine, and your actual audience data stays local and is excluded from version control by default. To use it you need Claude Code installed and some record of how your real audience behaves, even a small one such as thirty engaged people across five to ten past posts. You clone the repository, drop your evidence into a folder, then run the persona build command to generate your first set of personas.
A Claude Code skill set that builds audience personas from real observed behavior and uses them to judge and predict content performance before you ship it.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly pm founder.
This repo across BitVibe Labs
double-check against the repo, no cap.