Le pedí a claude code que por favor hiciera una auditoría de producto, como si fuera las 5 personas que mas tienen contacto con la app.
Te comparto el prompt que uso y que así es como se hace auditoría de producto.
"Audit this codebase as if you were 5 different people who live with this product. Tell me what each of them would run into.
1. Choose the 5 people. Pick the roles that touch this product the most. A good default:
- the operator who uses it all day
- the customer who receives what it produces
- security and compliance
- the on-call engineer who gets paged when it breaks
- a new developer who must ship a first PR this week
For each one, write who they are, what they care about, the one question they would ask, and which files or areas answer that question.
2. Run one read-only agent per person, in parallel. Give each agent its persona, its scope, and what the other four cover, so it stays in its lane. No edits, installs or commits. Only read, search and run checks that change nothing (typecheck, lint, tests that don't touch real data).
3. Every finding needs: file:line, what that person would actually see or suffer, severity, and a confidence tag:
[V] verified: you opened the code
[A] reported by an agent, not checked yet
[I] inference: needs confirming in production
4. Merge the results. Remove duplicates. If two or more people reach the same file by different paths, rank it first. Open the report with a 5-point summary in plain language, then one table per person.
5. Prove every finding with a screenshot, using the kind that fits:
- Screen: what that person actually sees. Run the app locally, read-only. If it needs a login, ask me to sign in.
- Code: the real snippet with line numbers and the evidence lines highlighted, from the commit you audited.
- Terminal: the output of a read-only command (lint count, failing tests, a missing CI folder).
While you build them, have a second agent reopen every cited line and mark each finding confirmed, partial or not confirmed. Fix wrong line numbers. Say whether it still exists on the latest main.
Keep screens with real customer data out of git. Never show secret values: only file:line and variable names.
Save the report as docs/audits/<date>-5-people.md, with a captures folder next to it and each capture under its finding."
Después corre claude para arreglar cada finding. Happy Hacking.
Paco Chim retweeted
AI isn't killing creativity. it's exposing who never had taste in the first place
prove me wrong
Todos deberíamos ser como mi esposa.
Le dije que quería comprar una GLX 550 Lexus porque la nuestra no es 4x4 y con esa camioneta hacer roadtrip épicos x todo el mundo.
Su respuesta? Hagámoslo!
Ni un puto segundo pensó acerca del dinero o lo que conlleva eso. Solo dijo, si.
Se viene chisme de Polsia, parece que todo es una mentira 🥹
Ok, here’s something I can’t figure out:
Polsia is at $25M run rate, which is an impressive achievement for any company, never mind a one-person company.
I’d just love to see some examples of successful companies that are using it. Why aren’t there examples or case studies on the website?
I tried the product again, and my first impression is that the experience below hasn’t changed much from earlier this year.
Of the info it’s showing me here:
1. The task list and product roadmap feel irrelevant
2. The Twitter feed is for what?
3. Maybe the Meta Ads feature is incredible?
I respect any founder who is out there trying to make it happen. I'd just love to some examples - which customers are letting Polsia run their business and making money?
If anything, seeing some real examples will convince more business owners to try this.
Crees que tienes paciencia hasta que un día a las
4:23 am, tu hijo hace popo, el ya un 90% dormido, decides cambiarlo para que no esté incómodo sabiendo que lo tendrás que dormir de nuevo.
5:25 am , Una vez cambiado, 90% dormido, se hace de nuevo.
Dime, tienes paciencia?









