agentleFS
mariomilemariomile/builder-os/skills/ux-architecture/SKILL.md

ux-architecture

Skill2 starsChanged 50 days agoMITView on GitHub

Designs how a feature works and looks: flows, every state of each step, components, accessibility, then rendered versions the person picks and a design kit. Use when a feature needs flows, screens or states, a look chosen, or design options.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

Files
SKILL.md

UX Architecture

The structural half of phase 4: placement, flows, states, components and the accessibility floor, decided before a pixel is chosen. The look comes next, in the direction mode: rendered versions that differ in structure, chosen by the person in rounds, then the kit and tokens; follow design direction, with its own marker, inside phase 4 or alone.

Mode. Standalone designs the supplied feature or flow from the available product context and the existing design system. In the lifecycle it runs inside phase 4 for spec-writing and writes the initiative's DESIGN.md. Both per operating modes. Read the section of UX methods behind steps 3 to 7, and Drawing the Flow for step 4.

Procedure

  1. Read the feature inputs. Standalone uses the supplied feature, goal and constraints. The lifecycle reads 03-solution-bet.md: its primary user action is the flow this skill designs. Done when the primary action is stated in one line.
  2. Map the current structure: navigation, the area this touches, existing components; without repo.read, ask and mark the inventory unverified. Done when the touched area and its components are listed.
  3. Place the feature where the user expects it, or name the cost of the easy spot, with current and proposed structure side by side. Done when what moves is named.
  4. Design the flows. Every entry point, the steps (user action, system response), decision points, both exits, reversibility; a multi-step flow states its abandonment exit (discard, draft or resume). On a new product (product track) one flow is the first day: from the first visit to the first useful outcome, with the empty screens a new account meets filled by sample data, a guided first action or an import, and the step that counts as activated named for the spec's **Activation:** (gate 4.17). Then draw each flow under its table as a Mermaid flowchart (entries, steps, decisions with labeled exits, success and abandonment), in the same words, per drawing the flow; the person reads and edits the drawing in the page bos.mjs flows writes, the build reads the table (gate 4.20). Done when every flow has its entries, steps and exits, drawn and tabled, and a new product has its first day.
  5. Cover the states: empty (with the action that fills it), loading, partial, error, success, permission. The permission state follows the existing auth, tenancy and permission contract, inspected first; where it cannot be read, the assumption is labeled unverified. Error copy names what failed and the next action, where disclosure is permitted. Strings live where the product's languages can be added (PRODUCT.md → Markets): each language's copy, dates, numbers and currency, and the longest translation tested on the narrowest screen. Done when every step of every flow has its six states. 5b. Design the messages (gate 4.21). Every email, notification, chat or text message the product sends outside its screens, found from the flows (a sign-up sends a welcome, an export emails a file, a failed payment warns): what triggers it, who gets it, the channel, its real text in the voice, how to stop it, and a preview a person opens, an HTML file per email or a mock of the chat bubble in design/messages/, per UX methods. For an agent that talks on a channel, the messages are the product: each turn type the spec's ## Agent behavior names gets its row. Done when every message has its row and its preview, or **Messages:** none with the reason.
  6. Inventory the components: exists, extend (naming what else it affects), new (with its justification). Done when every component in the flows has one of the three.
  7. State the accessibility floor: the keyboard path with tab order for the primary flow, the contrast target as a ratio, focus behavior on open, close, error and completion. Done when all three are written.
  8. Deliver the design to the requested destination. In the lifecycle write DESIGN.md per the design template, headings and labels exact, then hand the primary flow to ux-architecture (direction) and the flow list to the spec; gate 4.9 also needs the visual direction. Done when DESIGN.md holds the flows, states, messages, inventory and floor.

Watch for

DoInstead of
Keep the actual auth contract, including a deliberate concealed-resource 404Replacing it with an access message that reveals the resource
The existing pattern firstFour new components for one feature
Supplied context and the requested destination for a standalone requestCreating initiative state for it

More in UX methods.

Capabilities

CapabilityUsed forFloor if absent
repo.readExisting components, navigation and flowsAsk the person to describe them; mark the inventory as unverified
docs.searchDesign system documentation, prior flow decisionsAsk
analytics.replayWhere users actually stall in the current flowDesign from structure rather than observation, and say so
diagram.renderLaying a flow out on a canvas the person edits in a workshopThe page bos.mjs flows writes; without commands, the Mermaid block in DESIGN.md, which renders wherever Markdown does
files.read / files.writeSupplied context, requested design and lifecycle artifactsInline design needs no file write; lifecycle writes require it

No data capability is required. The visual layer is always covered: ux-architecture (direction) renders versions as files on any host, and uses a design toolchain where one resolved.

Output Contract

## DESIGN COMPLETE with the flow list and a diagram per flow (the first day on a new product), the state matrix, the messages with their previews, the component inventory and the accessibility floor. Lifecycle output is the initiative's DESIGN.md per the design template.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.