A structured way to model a product before building it. Sits between an idea and its implementation.
Compatible with the open Agent Skills standard. Runs on Claude and ChatGPT.
Product OS is a skill that runs a conversational nine-step flow with a founder or builder, then produces a single artefact: a Product OS document conforming to a schema. It does not generate code, wireframes, screens, or user stories. It generates the model those things get built against.
The premise: AI can generate almost any implementation from a decent prompt. What it cannot reliably infer is what exists in the domain, which interactions matter, which constraints preserve integrity, what compounds over time, and what outcome should emerge. Those are modelling problems. This skill makes the builder answer them first.
- Intended outcome — the state of the world the product brings about
- Compounding hypothesis — what the builder believes will grow every time the primary loop runs (verified later)
- Domain — where the product lives
- Entities — actors and objects, with a state check for two-state entities (user vs non-user of the same class)
- Relationships — directed triples between entities
- Interactions — what happens repeatedly, who trusts what
- Constraints — rules that preserve integrity under real use
- Alignment check — verifies the compounding hypothesis against the model, runs the non-permanence check, applies a capability library
- Produce the Product OS document
The output is a markdown file with parseable YAML blocks. The YAML is canonical; prose between blocks records reasoning.
Skills require Claude Pro, Max, Team, or Enterprise with code execution enabled.
Via Claude.ai: open settings, go to Skills, upload the product-os/ folder as a ZIP.
Via the Claude API: use the Skills API to attach the folder. See Anthropic's Skills quickstart.
Skills are generally available for ChatGPT Business, Enterprise, Healthcare, and Edu.
Via ChatGPT: in a workspace, select Create → Upload, and upload the product-os/ folder as a ZIP.
Via Codex or the API: see OpenAI's Skills documentation.
Invoke the skill by asking to model a product before building:
"I want to build a marketplace for used bikes near universities. Can we run Product OS on it first?"
The skill will not auto-run on plain build requests ("build me an app for X"). It will offer itself and proceed only if the user accepts.
SKILL.md— the skill definition and Product Exploration Mode flow. Read by the AI when the skill is invoked.PRODUCT_OS_SCHEMA.md— the schema the Product OS document must conform to. Referenced by the skill.CHANGELOG.md— version history and design decisions.README.md— this file. Not part of the skill itself.LICENSE— Apache 2.0.
Only SKILL.md and PRODUCT_OS_SCHEMA.md are strictly needed for the skill to run. Include CHANGELOG.md if you want the AI to have context on the skill's evolution.
This is v0.2. Iterations are expected. If you use Product OS on a real product and find something that would strengthen it for other products, please open an issue or PR. Two things worth calibrating over time:
- Hypothesis-vs-verification gap on compounding. Step 2 records the founder's compounding thesis; Step 8 verifies it. How often does the verification return
confirmed,revised, orabsent? A pattern in the gap would tell us something about how founder intuition maps to structural reality. - Capability library coverage. The current library covers measurement independence, external anchoring, handover mechanism, trust accrual, and membership substance. If a capability category recurs as a finding across enough Product OS runs to warrant inclusion, it should be added.
Product OS emerged from a working session modelling HeadsUp, a precinct improvement operator in Johannesburg. What started as a domain-specific tool got generalised into a product-agnostic skill after several patterns surfaced that would apply to any product: the compounding hypothesis, two-state entity check, non-permanence orientation, and the capability library.
The product-emergence predecessor is retained in CHANGELOG.md for provenance.
Apache 2.0. See LICENSE for the full text.