Prodman — AI skills for product managers

01 — THE ARGUMENT

You are not behind on AI.

You are sitting on ten years of work you are not allowed to show. That is a permission problem wearing a skills problem’s clothes, and no amount of prompt-engineering practice touches it. The fix is a process that strips what you cannot publish and keeps the part that was always the interesting bit — the decision you made, and what it cost.

What you remember

Yours. A decision you made and why is not your employer’s property, and it is the half a recruiter actually reads.

What you can publish

“Cut churn 18% for [named enterprise client]” is not publishable. “Cut monthly churn by roughly a fifth on a B2B subscription product” is — and it is the more interesting sentence, because it is about the decision rather than the logo.

What is missing

The gate between them. Everyone writes the case study and then worries. This runs the de-identification pass first, shows you exactly what it stripped, and refuses to guess on anything it is unsure about.

02 — WHAT IS IN IT

Two skills that work.

Deliberately small. One skill you run beats forty you never open. More get added as they prove themselves on real work — not before, and not to make the README longer.

portfolio-from-work

“Ten years of work and I can’t show any of it.”

Interviews you from memory — you do not need to find the documents. De-identifies everything before it writes a word, and shows you what it stripped. You end with case studies you can send and one artifact that actually opens.

launch-minimum

“I keep not launching and I can’t say why.”

Eight things that must exist before you launch, and nothing else. Audits what you have, names what is missing in funnel order, and refuses to let you build a ninth thing until the eight are real.

NEXTrelease-notes-from-diff

Turn what shipped into something a customer will read, without the changelog voice.

NEXTdemo-from-prd

A clickable artifact on synthetic data, from a spec, in an afternoon.

git clone https://github.com/aragjha/prodman-stack
cd prodman-stack
claude

# then, in Claude Code:
> Read skills/portfolio-from-work/SKILL.md and run it on my last project.

Open the repo↗

03 — THE PART NOBODY ELSE PUBLISHES

Every skill ships with the checks it has to pass.

The two largest PM repos on GitHub have 26,700 and 7,100 stars between them, and neither one tells you whether anything in it actually works. A prompt that produces confident, well-written, strategically useless output looks identical to one that works — until you have a check that can fail.

So every skill in this repo carries a VERIFY block, written before the skill was trusted. Here is the real one from launch-minimum:

VERIFY — launch-minimumthe checks · written before the skill was trusted

◇All eight assessed — each item has a state and evidenceAUTO

◇Evidence, not intention — every ✓ cites a URL, a file, or a transactionAUTO

◇Gaps in funnel order — missing items listed earliest-first, not easiest-firstAUTO

◇No partial marked ready — a ⚠ counts as missingAUTO

◇The ninth-thing list exists — AFTER is written, even when emptyAUTO

◇Smallest version proposed — every gap sized in hours, never weeksADVERSARIAL

◇A real verdict — go, or a dated plan. Never “nearly ready”ADVERSARIAL

Two red flags drop this skill to rung zero on their own: marking the path to pay done without a real card having gone through, and declaring go with a broken capture. Each skill also ships five frozen adversarial inputs built to make it fail, with every trap written down so you can audit the check instead of trusting it. Results publish as runs happen — failures stay on the board. A scoreboard you can only pass is not a scoreboard.

04 — THREE WAYS IN

Free is the capability. Paid is the capability kept.

The line between them is not a feature list, it is a real one: everything free is a one-shot generator — run it, get a good output, close the session. Everything paid needs memory, measurement, or your context to exist at all. If you only ever need the one-shot, take the repo and go. It is genuinely yours.

The Stack

Free · MIT

Every skill, forever, with no account and no catch.

Every skill as it ships
The eval golden sets — the adversarial inputs each one is tested against
Worked examples, start to finish
Fork it, sell what you build with it

Start free

START HERE

The 7-day challenge

₹2,500 · one-off

One prompt a day, a shared room, and I review your artifact personally on day seven.

Finish with three case studies and one working demo
From work you were never allowed to talk about
Every question answered the same day
15 places — the number one person can actually review

See the seven days→

The OS

$97 /month

The same skills, except they know your product and score their own output.

Memory — stop re-explaining your company every session
Scoreboards — every run checked, results published
A new workflow every week
The private room, weekday replies

What it does→

05 — WHAT THE FREE REPO CANNOT DO

Three things, said now rather than on day three.

✕It will not remember you.Every session starts from zero. You re-explain your product, your stage and your stakeholders each time, and the quality of what you get back depends on how patient you were that day.

✕It cannot tell you whether the output is right.It produces. It does not check. The golden sets let you check by hand — nothing runs them for you.

✕It does not improve because you used it.Your corrections die when the session closes, so the thing you fixed last Tuesday comes back on Thursday.

Those three are the whole difference between a folder of skills and a system — and they are exactly what the paid OS is. They are stated here because you will hit them yourself within a week, and finding out from the product is better than finding out from the bill.

Get each new skill the day it ships

One new skill lands in the repo as it is built and tested, each with the eval golden set it has to survive — five frozen adversarial inputs with every trap written down, so you can audit the checks instead of trusting them. Drop your email and you get each one the day it ships, along with the scoreboards as runs happen — failures included.

No newsletter. No drip sequence. A skill, when there is a skill.

06 — THE OBVIOUS QUESTIONS

Asked before you have to.

Is this just a prompt pack with a landing page?

Fair question — most of them are. The difference is the VERIFY block: every skill here states, before it is trusted, the checks it must pass and the red flags that disqualify it outright. A prompt pack cannot fail. These can, and when one does the result stays published.

Do I need to be technical?

You need to be able to install Claude Code and type a sentence into it. If you have ever cloned a repo you are over the bar. The skills interview you in plain English and you answer from memory — there is nothing to configure.

Is it actually safe to run this on work I signed an NDA about?

The de-identification gate runs before anything is written, not after, and it shows you every substitution it made so you can check them yourself. Where it is unsure it flags and stops rather than guessing. That said: it is a tool, you are the one publishing, and you should read what it produces before you send it anywhere. Anything you would not say out loud at a conference, do not ship.

Why would I pay when the free repo is MIT?

Most people should not, at first. Take the repo, run it, build the portfolio. You pay only when the three limits above start costing you more than the subscription — which happens when you are using this weekly rather than once. If it never happens, you have lost nothing and you keep everything.

What happens to my email?

It gets a message when a skill ships. That is the entire use. No sequence, no partners, no “we thought you’d like.” Reply to any of them and a person answers.