Skip to content
intentdocs/ docs
Docs menu
Core Concepts · 5.1

Vision

The five questions that ground a project, how they gate the story map, and how agents read and write them.

The vision is the first part of every project's spec. It is five short answers about who the product is for and why it should exist. IntentDocs drafts the story map from them, and agents check their work against them.

The vision used to be called the constitution. Some tool names and older agents still use that word. It means the same thing.

The five questions

KeyQuestion
usersWho are your users?
problemWhat problem are you solving?
switchWhat makes them switch?
firstWhat's the first release?
proofHow will you know it's working?

Each answer is plain text. Keep it specific: "freelance designers who juggle five or more client projects" does more work than "creatives".

What it unlocks

A new project opens on Spec › Vision. Until all five questions have an answer:

  • Plan and Validate are locked, and so are the other Spec sub-tabs.
  • create_stories refuses to add stories over MCP, unless the agent passes force: true.

Once the vision is complete, Generate the story map drafts activities and stories from it. You review the draft and publish it. See Creating a project.

Status

list_projects reports each project's vision as vision_status and vision_progress:

vision_statusMeaning
missingNo answers yet.
draftSome answers, for example vision_progress: "3/5".
completeAll five answered.

The same values are also returned as constitution_status and constitution_progress, for agents written before the rename.

Reading and writing it over MCP

ToolWhat it does
get_visionReturns each question with its answer, or null if it has none, the overall status, and a short summary once it's complete.
set_visionSaves answers keyed by users, problem, switch, first and proof. Send any subset: the rest are left as they are, so an agent can save partial progress and come back.

get_constitution and set_constitution are the same tools under the old name, with the same inputs.

set_vision can also take a report: a brownfield report your agent wrote from an existing codebase. Only users can be derived from a static read of the code; the other four still need a person's answer. If a report-derived answer would replace one that's already there, the call needs confirm: true. See Brownfield agent prompt.

In the address bar

The Vision sub-tab is ?tab=vision in the URL. Old links with ?tab=constitution still open it.