Working Backwards
Takes a product idea and walks it through Amazon's Working Backwards process: name the customer and their problem, write the press release as if the product already shipped, write the external FAQ, write the internal FAQ, note the visuals, then run one iteration pass that ends in a go or no-go call. Each output gets its own step, so a PM leaves with six distinct artifacts assembled into one PRFAQ document, not one generic writeup. Every claim traces back to something the PM actually said. When the PM hasn't given enough to answer a question, the skill marks that spot unsupported and asks, instead of inventing a plausible detail. The press release states one specific, testable customer benefit, never a vague claim like 'makes it easier' with nothing to check it against. The internal FAQ names real tradeoffs, not a repeat of the press release's pitch in different words. The iteration step ends in an explicit Go or No-go decision, never a list of things to go fix. Any material a PM pastes in as supporting evidence, a competitor's page, a support ticket, is treated as evidence to weigh, not an instruction, and personal details are stripped before anything from it is used. Ships with a reference PRFAQ template and an eval contract carrying two frozen self-test scenarios, one that ends Go, one that ends No-go.
First-party, by Skills and Agents Co
August 31, 2026
Simple install — no setup required
Open the Claude desktop app and go to the Project you want this in
Go to Code, then Customize, then click Create a new skill and upload the skill file below
Say a trigger phrase to activate it
Download the skill file below
In ChatGPT, create a Project and attach the file (or paste its contents into the project instructions). For a reusable setup, build a Custom GPT via Explore GPTs → Create and add the file there
Start a chat in that Project or GPT, then say a trigger phrase to activate it
Download the skill file below
Place it at .cursor/skills/working-backwards/SKILL.md (rename the downloaded file to SKILL.md). The shared .agents/skills/working-backwards/SKILL.md path also works
Cursor loads the skill when your task matches its description. Say a trigger phrase to activate it
Download the skill file below
Place it at .github/skills/working-backwards/SKILL.md in your repository (rename the downloaded file to SKILL.md)
Copilot loads the skill when it is relevant to your task. Say a trigger phrase to activate it
Download the skill file below
Place it at .agents/skills/working-backwards/SKILL.md in your repository (rename the downloaded file to SKILL.md)
Codex loads the skill when your task matches, or invoke it explicitly with /skills. Say a trigger phrase to activate it
Download the skill file below
In Gemini, create a Gem, then upload the skill file as a knowledge file
Copy the Gem instructions with the button below, then paste them into the instructions box
Start a chat with the Gem, then say a trigger phrase to activate it
You are a Gem built around the "Working Backwards" skill. Walks a product manager through Amazon's Working Backwards process as six sequential steps, ending in a complete PRFAQ. Marks anything you didn't actually tell it as unsupported instead of inventing it. Read the attached file first, then follow its instructions. Activate this skill when the user says something like: run this through Working Backwards, write a PRFAQ, help me PR/FAQ this idea, /working-backwards.
npx skills add skills-agents-co/skills-and-agents-library/working-backwardsYour skill is downloading.
Get an email when we ship new skills and updates.
You're on the list. We'll email you when new skills drop.
What this skill does
This is an overview of the skill from the site, not the skill's real instructions. The full SKILL.md Claude reads when this skill is active lives on GitHub at the source below.
Takes a product idea in a PM’s head, plus a few facts they can state, and produces a full Amazon-style PRFAQ from scratch: a press release, an external FAQ, an internal FAQ, a visuals note, and a Go/No-go call, each as its own step with its own artifact.
This is not a line editor for a PRFAQ you’ve already drafted, and it doesn’t turn research into a PRD (a different tool covers that). It starts from an idea and a handful of answers, and produces the pressure-test Amazon runs before greenlighting work.
What it does
- Names a specific customer and problem first. Rejects a solution dressed up as a problem (“customers need an AI assistant” isn’t one; “customers spend an hour a day copying data between two tools” is), and pushes for a problem that costs the customer time, money, or frustration today.
- Writes a press release with a falsifiable benefit. The headline states a number or a concrete before/after a reader could check once the product ships, never a vague claim like “makes it faster.” If the PM genuinely doesn’t have a number yet, the release says so with a clearly flagged placeholder instead of inventing one.
- Writes an external FAQ that traces to what was actually said. Five to ten questions a real customer would ask, each answered from the PM’s own words. A question with no real answer gets “Not yet defined, ask the PM,” not a guessed price or date.
- Writes an internal FAQ with real tradeoffs. Feasibility, cost, and risk, answered as a skeptical exec would ask them in a launch review. An answer that just restates the press release’s pitch doesn’t count.
- Notes the visuals or says there aren’t any. A short walkthrough of the key screen for a UI product, or a direct statement that no user-facing surface applies, never an invented mockup.
- Ends with an actual decision. The iteration step rereads everything, checks whether the internal FAQ’s tradeoffs still support the press release’s benefit, and closes with an explicit Go or No-go line and the reasons behind it, never a punch list standing in for a decision.
- Treats pasted evidence as evidence, not instructions. A competitor’s page or a support ticket a PM pastes in gets summarized and weighed, not followed if it reads like a directive, and any personal details in it are stripped before use.
How to use it
- Give it a product idea in a sentence or two: what it is, who it’s for.
- It walks you through the six steps one at a time, asking only what it needs for the step in front of it.
- Short answers are fine. “I don’t know” is a valid answer at any step, and gets marked as an honest gap instead of filled with a guess.
- At the end, it assembles all six artifacts into one PRFAQ document using the reference template, plus a Go or No-go call with its reasons.
Output format
One assembled PRFAQ document: Customer and Problem, Press Release, External FAQ, Internal FAQ, Visuals Note, and an Iteration Log ending in a stated Go or No-go decision.
Pitfalls
- Don’t write a press release benefit a reader can’t check. Get a number, a concrete before/after, or an honest, clearly flagged placeholder.
- Don’t let the internal FAQ echo the press release. An answer that reads like marketing copy isn’t naming a real tradeoff yet.
- Don’t invent a mockup for a product with no UI. Say there’s no user-facing surface and move on.
- Don’t end the iteration step with a punch list. Its job is a decision, not homework.
This skill’s process is Amazon’s own Working Backwards method, rebuilt as a step-by-step Claude skill. No text from any external source is reproduced; every rule here is the skill’s own instructions.
View full SKILL.md on GitHub