authoring-wireframes — SKILL.md

Variant: standard · When to use: producing (or amending) a structural wireframes document from the upstream user-flows, to a bar an engineer can build the screen structure from.

Overview

This skill is the how-to of writing a strong, comprehensive wireframes document — the judgment a producer applies, not the section list. A wireframe here is low-to-mid fidelity and structural: the layout regions (on a grid, in a shared app-shell), content hierarchy, components, affordances, content/microcopy intent, per-screen states, and screen-composition accessibility of each key screen — expressed in text as a structured layout description + an ASCII/markdown box sketch + per-element annotations, not a pixel mockup. It assumes two collaborators: a template tool that supplies the section structure, and a deep-research capability to ground each screen in established UI patterns. The producer is handed the upstream user-flows (the screens and state-transitions each flow traverses) plus a design-system if one exists — and lays out every flow-named screen, never a blank page. The bar to clear: the doc is buildable (an engineer can build the screen structure), covered (every flow-named screen + state has a wireframe), and well-composed (an objective layout-quality bar).

Fidelity is a spectrum across three axes — interactivity, visuals, content (NN/g). A wireframe is low-to-mid on visuals, mid on structure, low on interactivity. Final color/type/pixels are downstream hi-fi; the token system + per-component a11y contract are the design-system's — this skill stays structural. The method is medium-independent: the artifact today is markdown; a future design-tool backend changes only the medium, not the thinking or the bar.

When to activate

• Authoring a new wireframes document from a project idea + the upstream user-flows.
• Laying out the key screens (and their states) that a set of user-flows names.
• Amending an existing wireframes doc when a flow, screen, region, or referenced component changes (see Step 8).
• Filling a wireframes template with researched, decision-complete structural content.

Do NOT activate when:

• Reviewing or grading a finished wireframes doc → use reviewing-wireframes.
• Producing high-fidelity visual design / final pixels / color + type / motion → that is the downstream hi-fi pass; the design-system owns the visual tokens.
• Re-deciding which screens and transitions exist → that is the upstream user-flows; this lays out each screen.
• Authoring or grading the token system / component catalog → that is the design-system.
• Authoring a different document type → use that type's skill.

Inputs

Read every document the plan hands you — your depends_on set (the typical upstreams: the user-flows that name the screens, and a design-system where one exists). Trace this document's content back to them. Do not assume a fixed input: the named upstreams are guidance, not a precondition. Be self-contained — produce from whatever context you receive; when an expected upstream is absent, proceed on what you have and surface the gap as an explicit assumption (Step 7), never fabricate. Use a research capability (deep-research) where available to ground each screen, not merely to fill the template.

Capability context (when provided): If a capability_record (a record from capability-map.yaml product_capabilities) is injected by the caller, read it before Step 1. It defines your scope boundary: owns = entities you cover; refs = entities you reference but do not own; publishes/consumes = events you surface; entry_points/exit_points = how users arrive and leave; has_ui/has_api/has_persistence = which surfaces apply. When present, treat it as a hard constraint — do not stray outside the boundary it defines.

Workflow

Step 1: Take the structure from the template tool — don't invent an outline

Get the section structure from your wireframes template tool (comprehensive variant). Do not restate a section list here; this skill supplies the method that fills those sections well. If no template is available, obtain/forge a comprehensive wireframes structure, then proceed.

Step 2: Derive the screen list from the upstream user-flows

The wireframes doc depends_on the user-flows. Walk every flow; every screen a flow names, and every state-transition it implies, gets a wireframe. Build a screen-inventory table (each flow-named screen/state → its wireframe section + the states it visits) so coverage is auditable — this table is also the amend-ripple surface (Step 8). A flow naming a screen with no defined content is a gap — record it (Step 7), never silently invent it.

If a capability_record is present: constrain the screen list to the capability's entry_points/exit_points boundary. Mark exit_points as cross-capability transitions — do NOT design what happens after. The system-wireframes doc owns the shell and cross-capability navigation.

Step 3: Research to ground each screen — don't invent patterns; hold the fidelity line

Use deep-research to ground each screen in established UI patterns (conventional layouts for its type — list/detail, master-detail, dashboard, settings, wizard, auth, search-results — and standard empty/loading/error treatments) rather than inventing structure. Where a design-system exists, reference its real components and tokens — never invent them. Hold the fidelity line: lay out structure (regions, grid intent, component selection, affordances, states), never exact hex/px/type — that is hi-fi overshoot. If no research capability is available, don't fabricate a pattern or a component name — lay out the conventional structure and flag uncertainty as an assumption.

Step 4: Establish the global frame — grid, app-shell, layout-quality bar

Before the per-screen work, fix the conventions every screen inherits:

• Layout grid + spacing cadence (structure, not pixels): the column structure (e.g. 12-col) + an 8pt (or 4pt dense) rhythm intent. State the cadence; the design-system owns exact px. A raw 13px is off-cadence and a hi-fi overshoot.
• App-shell / page-frame: ONE shared header/nav/container/max-width, defined once and reused on every screen — no per-screen full-bleed dumps. This shared region is what an amend ripples through.
• Objective layout-quality bar (every screen composes to it; subjective taste is NOT required): primary action on the scan path (F/Z), related elements grouped (Gestalt proximity), no gratuitous region/element, consistency across sibling screens.

If scope=system: your input is ALL capability records with has_ui=true — wire all entry_points into the nav and produce transition specs for all exit→entry seams.

Step 5: Apply the per-screen method

Fill the template's per-screen sections to this method. For each key screen:

• Purpose & flow context — one sentence on what the screen lets the user do, the flow step it serves, entry points + where each affordance exits.
• Layout regions + content hierarchy — name the regions (Step-4 vocabulary) and their arrangement on the grid; order content by visual priority driven by the primary task; group related items (Gestalt) and place the primary action on the scan path. Structure before any visual concern.
• Box sketch — an ASCII/markdown sketch of the default/populated layout: regions labelled, key elements placed, the legend defined once and reused. Keep the sketch and its annotations in sync.
• Components & affordances (annotations) — per notable element: the design-system component (or a generic type), full anatomy where it applies, its behavior on interaction (tap/hover/validation/conditional visibility), where it leads, and edge cases (truncation/overflow/sort/paginate/dropdown). Annotations turn a static layout into a buildable spec.
• Data-display (for data-dense screens) — table vs list vs cards; for a table: columns + sort/filter/paginate/row-density/sticky-header/row-expansion; progressive disclosure (what's primary vs deferred behind accordion/disclosure/tooltip/show-more); forms with grouped fields + structured inputs (no hand-authored serialized strings).
• Content & microcopy (intent, not final copy) — the intent of each text slot (label/heading/button/empty/error/placeholder); speak the user's language (no raw data-model identifier as a label — human label + one-line description); placeholder ≠ label; terminology consistency (one name per concept across screens); i18n text-expansion room + RTL where localized (scope out single-locale with a one-line reason).
• Per-screen states — empty (why + guide-to-action CTA, plain language), loading (a skeleton mirroring the populated layout over a bare spinner where the layout is known), populated/default, error (plain-language cause + recovery + placement, inline vs blocking, no raw codes), success/confirmation (what changed + next step) and partial where the path visits them. Trace the states to the ones the flow's path actually visits. Happy-path-only is the #1 gap.
• Responsive notes (where form-factor matters) — what stacks/collapses (nav → hamburger/off-canvas)/hides/reorders and the mechanism; content-priority change on small screens (promote/demote/hide, not just stack); touch-target reservation.
• Accessibility — screen-composition — landmarks per region, one-h1 heading order, reading/focus order (incl. on conditional reveal), accessible names for icon-only controls, keyboard operability (no mouse-only step), target-size reservation (≥24×24 CSS px; SC 2.5.8), focus-not-obscured by sticky regions (SC 2.4.11), non-color-only intent (SC 1.4.1). Defer pixel contrast + focus-appearance to the design-system (the per-component contract); if no DS/hi-fi exists downstream, record that deferral as an assumption (Step 7), don't drop it.

Step 6: Self-check against the bar before handing off

Confirm all hold (this is the bar reviewing-wireframes asserts — same list, single-sourced, each proportional to the screen's archetype):

1. Full coverage — every flow-named screen + state-transition has a wireframe (no orphans/gaps vs the flows).
2. All applicable states — empty/loading/populated/error + success where a state-change occurs + partial where relevant; quality, not just presence.
3. Layout & composition — regions + hierarchy unambiguous; on a grid; in the shared app-shell; the objective layout-quality bar met (scan-path/grouping/no-gratuitous/sibling-consistency).
4. Components — DS components named + consistent + composed; none invented.
5. Affordances + data-display — behaviors/destinations annotated; data-dense screens specify their table/disclosure/form structure.
6. Content & microcopy — text-slot intent stated; no data-model leak in labels; persistent labels; consistent terminology.
7. Responsive — reflow mechanism + mobile priority where form-factor matters.
8. Screen-composition a11y — landmarks/one-h1/reading order/names/keyboard/target-reservation/focus-not-obscured/non-color (pixel-contrast deferred to the DS).
9. Annotation & handoff — legend defined; sketch ⇄ annotations in sync; elements map to DS components (the handoff translation).
10. Gaps surfaced — undefined screens/content, missing components, deferred component a11y = explicit assumptions/open-questions.
11. Structural, not hi-fi — lo-to-mid fidelity; no final pixels/color/type.
12. Versioned (on amend) — the doc's own version + changelog reflect the change (Step 8).

Step 7: Surface gaps explicitly

A flow naming a screen with undefined content, an element needing a component the design-system doesn't define, or component-contract a11y with no DS/hi-fi downstream → an open question or assumption. List it; flag missing components for the design-system owner. Never paper a gap with an invented screen/component.

Step 8: Amend mode — edit, don't redraw (when iterating)

When handed an existing wireframes doc + a change, treat it as a scoped, versioned diff, never a regenerate:

1. Scope + ripple — identify which screen(s)/state(s)/region(s) change, then trace the ripple: every screen reusing a changed shared region (app-shell/nav/header), every inventory row / flow hand-off referencing an edited/removed screen, every screen naming a changed/deprecated DS component, every screen added/removed by an upstream flow change.
2. Edit, don't redraw — minimal in-place edit; untouched screens stay byte-for-byte unchanged; a shared-region change is applied to ALL reusing screens in one pass (the one legitimate wide ripple), not screen-by-screen drift; stable screen + component names preserved.
3. Version + changelog — bump the doc's own version: MAJOR (removed/renamed screen or removed state — breaks the build/hi-fi contract), MINOR (added screen/state/region/component), PATCH (annotation/notation/copy-intent fix); add a Keep-a-Changelog entry at the screen/state grain.
4. Deprecate before removing — mark a screen/state deprecated in a MINOR (note the replacement) before removing in a MAJOR; on removal confirm no inventory row / flow hand-off still references it and prune the inventory.

Rules

Hard rules (never violate):

• One wireframe per flow-named screen/state. Coverage is keyed to the upstream user-flows; an orphan screen or missing state is incomplete.
• All applicable states. empty/loading/populated/error + success on a state-change. Happy-path-only is not done. A truly static screen collapses inapplicable states (proportional).
• Reference real components; never invent. Name the design-system's real components/tokens; a custom need is a flagged gap, not a silent invention.
• Structural, not hi-fi. Layout + hierarchy + annotation; no final pixels, color, or type — that is the hi-fi pass; the design-system owns tokens.
• Own composition a11y; defer the component contract. The wireframe owns landmarks/heading/focus-order/names/keyboard/target-reservation/focus-not-obscured/non-color; the design-system owns pixel contrast + focus-appearance + the component's own target-size/keyboard. Don't re-judge what the DS certifies; don't drop it when no DS exists (assumption instead).
• Speak the user's language. No raw data-model identifier as a visible label; a placeholder is not a label.
• One app-shell, on a grid. Every screen sits in the one shared shell on the grid cadence — no per-screen re-invention.
• Compose, don't duplicate. Take the section structure from the template tool; don't paste a competing outline.
• Lay out the given flows. The screen set comes from the user-flows, not imagination; don't re-decide navigation.
• Surface gaps, don't invent. Undefined screen/content, missing component, deferred a11y = explicit assumption/open-question.
• Amend, don't regenerate. On iteration, edit in place + version + changelog; never re-draw untouched screens.
• Buildable or not done. Don't hand off a doc an engineer cannot build the screen structure from.
• Capability boundary (when capability_record provided). Scope the document output to that record's boundary. Producing content outside the boundary (covering a refs entity as if it were owned; designing flows past an exit_point) is a scope violation — equivalent to inventing content that isn't in the spec.

Preferences (override-able):

• "Comprehensive" sets ambition, but stay proportional — completeness-of-decisions for the screen's archetype, not word count. A trivial screen collapses sections it doesn't need.
• Define the ASCII notation legend once per doc and reuse it.
• Prefer a skeleton-that-mirrors-the-populated-layout over a bare spinner where the layout is known.
• The objective layout-quality bar is objective only — don't impose subjective layout taste.

Gotchas

• Happy-path-only screens. Omitting empty/loading/error/success is the single most common gap — the states are where real UX lives.
• Invented components. Naming a component/token the design-system doesn't define drifts from the system and breaks handoff. Reference real components or flag the gap.
• Coverage gap vs the flows. A flow names a screen that never gets a wireframe — the inventory table exists to catch exactly this.
• Drifting into hi-fi. Exact colors, type, pixel spacing overshoot the structural scope and step on the hi-fi/design-system scope. Stay structural.
• Double-judging a11y. Re-specifying pixel contrast/focus-appearance the design-system owns is drift, not rigor — own composition a11y, defer the component contract.
• Data-model leaking into labels. A raw identifier (manifest_docs) as a visible label is the schema leaking through — translate to a human label + description.
• Generic fill. A layout that fits any product means you laid out the template, not this product's flow. Tie every screen to its flow step + real content.
• Static layout, no annotations. A box sketch with no per-element annotation is a picture, not a spec. Annotate components, affordances, states — and keep them in sync with the sketch.
• Shared-region drift on amend. Editing the app-shell on one screen but not the others that reuse it is the most common amend regression — apply it to all reusing screens in one pass.

Anti-patterns

• "The populated view is enough." Omits empty/loading/error/success — document all applicable states.
• "I'll make up a component for this." Reference the design-system's real ones or flag the gap.
• "I'll redesign which screens exist." That re-decides navigation (user-flows' job) — lay out the given screens.
• "I'll write the outline myself." Duplicates the template tool — take the structure from it.
• "Let me add colors and exact spacing to make it concrete." Overshoots into hi-fi — stay structural.
• "I'll also pin the exact contrast ratio." That is the design-system's component contract — reserve non-color intent, defer the pixel value.
• "The flow is vague here, so I'll invent a plausible screen." Surface it as an assumption instead.
• "It's a small change, I'll just regenerate the doc." Amend in place + version + changelog; never re-draw untouched screens.

Output

A comprehensive, structural wireframes document that meets the Step 6 bar (full coverage vs the flows; all applicable states; unambiguous layout/hierarchy on a grid + shared app-shell to the objective layout-quality bar; identified + consistent components; annotated affordances + data-display; content/microcopy intent; responsive where it matters; screen-composition a11y; legend + sketch⇄annotation sync + design-to-code mapping; gaps surfaced; structural-not-hi-fi; versioned on amend). Each screen is a structured layout description + an ASCII/markdown box sketch + per-element annotations. The abstract consumer is the downstream hi-fi pass + UI engineering (which build the screen structure) and a reviewer (which asserts the same bar). The doc's structure comes from the template tool; this skill supplies the content quality.

Related

• A wireframes template tool (e.g. a content/template gateway) — supplies the comprehensive per-screen section structure this skill fills.
• A deep-research capability — grounds each screen in established UI patterns and (where present) the design-system's real components.
• reviewing-wireframes — asserts the same buildability + coverage + composition + layout-quality bar; author and reviewer share one bar so they don't drift.
• The upstream user-flows (a depends_on document) — names which screens and state-transitions need wireframes.
• A design-system (where one exists) — supplies the real components/tokens to reference and owns the per-component a11y contract (contrast, focus-appearance); the wireframe references + reserves, never re-defines.
• The downstream hi-fi pass — owns final visuals/pixels; this skill stops at structure.

Progressive disclosure

• references/composition-and-states.md — depth on the grid/8pt cadence, app-shell, Gestalt/scanning hierarchy, the data-display (table/disclosure/form) catalog, and the per-screen state-quality rules. Load when composing a data-dense screen.
• references/a11y-and-amend.md — depth on screen-composition WCAG 2.2 (SC 2.5.8 / 2.4.11 / 2.4.3 / 4.1.2 / 1.4.1, the component-contract boundary) + the amend ripple/versioning procedure. Load for an a11y pass or an amend.
• references/sources.md — research provenance for the method + the bar (load only to audit where the guidance came from).

Body budget

• description ≤ 1,024 chars (agentskills.io cap).
• Body ~500 lines / 5,000 tokens (soft target — quality takes precedence; flag if consistently over 700 lines / 7,000 tokens).
• Heavy content lives in references/, loaded on demand.