Requirements Wayfinder
Plan what a feature needs with an AI agent, one decision at a time. You end up with a requirements document the delivery team can pick up. It produces no technical spec and no code.
For AI agents
Start at /llms.txt. It lists every file on this site. The skill itself is /requirements-wayfinder/SKILL.md.
To use it in any AI chat that can read web pages, paste this:
Fetch https://requirements-wayfinder.pages.dev/requirements-wayfinder/SKILL.md and follow it.
My feature idea: <two or three sentences>What it does
A new feature idea is mostly open questions. Who is it for? What counts as success? What's out? The skill lists those questions on a map and answers one per session.
Most questions are a short interview. The agent asks numbered questions and suggests an answer for each, and you can reply "yes" to accept. Some are research it does on its own. Some are a rough written draft, such as a sample one-pager, for you to react to.
When no questions are left, the agent compiles the answers into the requirements document. That document covers:
- Problem
- Who it's for
- Goals and success measures
- In scope and out of scope
- Requirements, each marked must, should or could
- Constraints
- Risks
- Open questions for the delivery team
How to use it
Claude Code or Claude desktop. Unzip requirements-wayfinder.zip into ~/.claude/skills/. Open a folder for your planning work and type /requirements-wayfinder followed by the idea, or just ask Claude to plan the feature's requirements. The map is saved as files in that folder.
claude.ai. Turn on "Create and edit files" under Settings > Capabilities. Then go to Customize > Skills, click "+", choose "Upload a skill" and pick the same ZIP. Then ask Claude to "use Requirements Wayfinder" with your idea. Custom skills need a Pro, Max, Team or Enterprise plan. If a connector Claude can edit is set up, such as Claude Docs or Google Docs, the map is saved there. Otherwise Claude prints the map at the end of each chat for you to paste into the next one.
Any other AI chat. Paste the prompt from the box above. It works in any chat or agent that can open a URL.
The skill answers one question per session. To pick up where you left off, start a new session and point the agent at the map.
How it differs from Wayfinder
Matt Pocock wrote Wayfinder for engineers. It charts a big effort as tickets on a repo's issue tracker and resolves them until the way to a spec or a code change is clear. The method works just as well for business decisions, but the defaults assume a developer. This version keeps the method and changes the rest.
| Wayfinder | Requirements Wayfinder |
|---|---|
| Can end at a spec, a decision or a code change | Always ends at a requirements document. Technical questions go to the delivery team as open questions. |
| Map and tickets live on GitHub, GitLab or a repo folder | A plain folder, a living document, or a markdown block pasted between chats. No repo or git. |
| Research results go on a git branch | Research results go in a file or section next to the question, with sources. |
| Prototypes can be UI or logic code | Drafts are written only: a one-pager, a customer journey, a mock announcement. |
Domain modeling writes CONTEXT.md and ADRs | A plain glossary of agreed terms. Decisions stay with their questions. |
| Stops when the way is clear and hands off | Adds a final session that writes the requirements document and walks through it with you. |
| Uses terms like frontier, fog of war, HITL and AFK | Uses plain words: open, answered, parked, ruled out. |
Skills on this site
The adapted skill:
- requirements-wayfinder is the skill to run. Its folder also has an unchanged copy of wayfinder.md, which it reads for the method.
The skills it builds on, copied unchanged from mattpocock/skills:
- wayfinder charts a big effort as a map of decision tickets and resolves them one at a time.
- grilling interviews you in numbered rounds, each question with a recommended answer. It's the default question type.
- grill-with-docs runs grilling and domain-modeling together and writes the glossary as it goes.
- domain-modeling keeps terms precise and records them in a glossary. Formats: glossary, ADR.
- research answers a factual question from primary sources and cites them.
- prototype builds a throwaway artifact to react to. The adaptation uses written drafts only. Branches: logic, UI.
Source and license
The original skills come from github.com/mattpocock/skills by Matt Pocock, copied at commit 5beb35b. They're MIT licensed (license text). The adaptation is MIT licensed too. For updates to the originals, go to the GitHub repo. The copies here don't update.