A judgment layer for AI product work.
AI made output almost free. More code. More PRDs. More variants. More features. More copy. The scarce thing now is not output, it is judgment. What is worth making. What to cut. What story to tell. What to refuse.
This stack is built for that. Three skills that turn your coding agent into something closer to a thinking partner: one for deciding what should exist, one for raising the standard of what does, and one for making the world want it.
I wanted an AI that thought with me, not for me.
Most agent workflows are powerful but shapeless. Ask for an idea, get ten. Ask for a plan, get a long plan. Ask for product feedback, get a polite list of improvements. The model is doing work but it is not applying judgment. It is not asking what should exist, what should be cut, what the user actually feels in the broken moment, or whether the work has a story strong enough to survive contact with the world.
I wanted structure for that. A path.
Steve Jobs became the lens because his transferable skill was not charisma or cruelty. It was taste as discipline: start with the experience, force the cut, make the product and the story one thing, refuse to hide weak thinking behind more features. He was also wrong plenty of times. But, those failure patterns are part of this, not embarrassments to skip over.
It matters most in AI. AI-powered X is not a product. A model is not an experience. A clever agent loop is not a reason for anyone to care. The job is to decide what actually changes for the user when the system exists, and to be honest when the answer is nothing yet.
The Steve Jobs Stack is my attempt to make that thinking reusable. Not a personality. Not a costume. A discipline for making better things.
I also maintain vikranthreddimasu/steve-jobs-skill, a persona skill: ask Jobs a question, get a Jobs-shaped answer (mental models, decision heuristics, expression DNA). My English translation of alchaincyf/steve-jobs-skill by Huashu (花叔), with full credit to the original author.
This stack is the discipline version: the principles disappear into your PRDs, reviews, and launch copy. The skill is the conversational version: Jobs's voice helps you think out loud about a question. They compose well, you can use the stack to decide what to build, the skill to think about it.
One command (recommended). Installs all three skills into every coding agent you have - via the open agent skills CLI:
npx skills add vikranthreddimasu/steve-jobs-stack --allRe-runnable. Add -g for global install across all projects, drop --all to choose interactively, or pass --skill steve-jobs-make to install just one.
If it's any easier, paste this in your coding agent. Claude Code, Cursor, and Codex all handle it from there.
Install the Steve Jobs Stack from this repo:
https://github.com/vikranthreddimasu/steve-jobs-stack
Run: npx skills add vikranthreddimasu/steve-jobs-stack --all
Then ask me what I am working on.
Once installed, the skills route themselves. Describe what you are working on and the right one takes over. A few prompts to get started:
Help me draft a PRD for X # triggers steve-jobs-make
Review my README, be honest # triggers steve-jobs-refine
Write the launch tweet for Y # triggers steve-jobs-pitch
The stack follows the lifecycle of making anything with AI.
steve-jobs-makefor what does not exist yet.steve-jobs-refinefor what does, and needs judgment.steve-jobs-pitchfor what is ready to be introduced.
These are not templates. They are modes of work. You describe what you are doing and the right one takes over.
For a new product, agent workflow, research direction, PRD, name, or v1 plan. Before producing the artifact, the skill interrogates the idea: what it is really for, who it serves and in what specific moment, what broken status quo it replaces, the one risky bet it depends on, and what happens when the system is wrong. For anything AI-shaped, that last question is the product. The artifact is a byproduct of the thinking, not the goal.
For work that already exists: a draft, a feature, a launch candidate, a strategic direction. The skill reviews the real artifact, not the description of it. It evaluates direction before execution. When asked whether to ship, add a feature, or keep going, it makes a binary call and names the one move, almost always a cut, that takes the work from competent to memorable. When something feels off but you cannot say why, it pattern-matches against documented Apple-era failures so the diagnosis has a name.
For introducing the thing: launch tweets, README heroes, Show HN posts, demo scripts, release notes, conference proposals, elevator pitches. The skill verifies every concrete claim, builds four assets first (the keynote line, the three, the demo moment, the friend message), and treats every launch artifact as a translation of those, not a fresh draft. The job is to make the value obvious in 30 seconds.
I have an idea for an AI coding agent that reviews migrations
before deploy. What should v1 be?
Routes to make. The skill forces the customer moment, the antagonist, the one bet, the failure mode, and the ruthless v1 before producing a single bullet.
Review this agent workflow. Be honest about whether it is ready
to ship.
Routes to refine. The skill looks at the actual workflow, makes the readiness call, names the weak point, and gives the one move that lifts it before launch.
Write the Show HN title and post for this repo.
Routes to pitch. The skill verifies the claims, finds the keynote line, builds the demo moment, and writes the post. Not three options. The post.
Each skill bundles the reference file it actually uses, so the research travels with the install:
skills/steve-jobs-make/references/product-case-studies.md— the build decisions: the 1997 product cull, iMac, iPod, Apple Retail.skills/steve-jobs-refine/references/failure-patterns.md— the traps: Lisa, NeXT cube, Apple Cube, MobileMe, Antennagate, the App Store reversal, options backdating, cancer delay, and the personal-cost pattern.skills/steve-jobs-pitch/references/keynote-case-studies.md— the launches: Macintosh, iMac, iPod, iPhone, MacBook Air, iPad.
A fourth file, references/research-report.md, stays at the repo root as the foundational background — the core principles (customer experience first, focus, integrated stack, liberal arts plus technology, simplicity, invisible craft, talent density, small teams, DRI ownership, self-cannibalization, keynote craft, taste over data, crisis as clarity, real artists ship, the people doing the work decide, and the limits of conviction). Browse it for context; the skills do not load it every time. They reach for what the work calls for.
steve-jobs-stack/
├── README.md
├── LICENSE
├── references/
│ └── research-report.md
└── skills/
├── steve-jobs-make/
│ ├── SKILL.md
│ └── references/
│ └── product-case-studies.md
├── steve-jobs-refine/
│ ├── SKILL.md
│ └── references/
│ └── failure-patterns.md
└── steve-jobs-pitch/
├── SKILL.md
└── references/
└── keynote-case-studies.md
Each skill is a single editable SKILL.md. Fork it, tune it, make the discipline fit how you want to work.
MIT.
Built by Vikranth Reddimasu.