Skills for product builders,
not another tool.

ProductClaw is a small, curated directory of agent skills for the work product builders actually do — discovery, research, sizing, validation, spec, pricing, growth, retros. A skill is a single markdown file you drop into your own AI agent. Each one is rooted in a named, citable method, single-purpose, and opinionated enough to push back when you're being lazy with yourself. No app to learn, no account, nothing to pay — you read the source, you own the file, you run it inside the tools you already use.

How it works

Find a skill

Browse by the stage you're in — discovery, research, sizing, spec — not by technology. Every skill page tells you what it's for, when not to use it, and which method it's built on.

Copy or curl the skill.md

One markdown file, fully readable. Copy it from the page or pull it with a single command — no install step, no dependency, no telemetry.

Drop it into your agent

Put it in Claude Code's .claude/skills, a Cursor rule, a Claude Project, or a ChatGPT Custom GPT — anything that takes a system prompt.

Use it in your workflow

Invoke it by name or just describe what you're working on. The agent now behaves like a senior practitioner running that method — and tells you when your idea doesn't hold up.

Installing a skill

Claude CodeDrop skill.md into .claude/skills, then invoke with /<skill>.

CursorSave it as a rule in .cursor/rules and reload.

Claude ProjectPaste the source into the project's instructions.

ChatGPTPaste it into a Custom GPT's instructions.

Any agentPaste it as the system prompt and start the conversation.

Every skill page has copy-paste and one-line install snippets for each of these.

The premises

AI doesn't lack capability. It lacks discipline.

A general-purpose agent will happily agree with whatever you bring it. The value was never the model — it's the method you point it at. A skill is the discipline made portable: it makes the agent ask the questions a good operator would, in the order they'd ask them.

A skill is a method, not a prompt.

Each skill is built on a real product-craft tradition — The Mom Test, Working Backwards, Fermi estimation, Jobs to Be Done — and we cite the lineage on every page. The markdown teaches the agent to run that method, not to perform cleverness. You can read exactly how it thinks before you trust it.

AI-assisted, manually curated.

Skills are drafted with AI and then hand-curated — read, pressure-tested, and rewritten by a person before they ship. Nothing here is crowdsourced and there's no leaderboard. We don't publish a skill until it earns its place, which is why the catalog is small on purpose.

No new tool to learn.

Skills live inside whatever agent you already use — Claude Code, Cursor, Claude Projects, ChatGPT Custom GPTs, or anything that accepts a system prompt. Adopting one is a copy-paste, not a migration, and removing one is deleting a file.

What it isn't

Not a general-purpose prompt library — every skill is single-purpose and tied to a stated method, not a grab-bag of tricks.
Not a SaaS product — no account, no dashboard, no data collection, no fee. It's markdown you copy and own.
Not an open marketplace ranked by installs — skills are hand-selected, not voted up, and the count stays small by choice.

Browse the skills

Skills for product builders, not another tool.