Matlab Build App
MathWorksOfficialCreated by MathWorks
- Skill
- Workflow
- Software development
- BSD-2-Clause
- No issues found
AutoGineer summary
Written by AutoGineer from the skill’s own files, not by the author
This skill provides a structured approach to developing MATLAB applications, guiding users from initial requirements to a fully implemented codebase. It helps determine the optimal architecture (UIFigure or UIHTML) and layout archetype (e.g., Dashboard, Explorer, Wizard) based on the user's specific needs, team expertise, and maintenance requirements. The skill enforces a rigorous planning phase, requiring an implementation plan before any code is written, and supports multiple serialization formats for UIFigure apps, including App Designer (.mlapp) and plain-text (.m + .xml) formats.
What it does
- Conducts discovery to identify app purpose, maintenance needs, and visual requirements.
- Recommends the best architecture (UIFigure or UIHTML) and layout archetype.
- Produces a detailed implementation plan and writes it to a markdown file.
- Supports building UIFigure apps with optional App Designer serialization.
- Ensures structural integrity by enforcing grid-based layouts and proper background task handling.
- Provides a clear, step-by-step build sequence based on internal documentation.
When to use it
- When you need to build a new MATLAB GUI or interactive tool.
- When you are unsure whether to use UIFigure or UIHTML for a specific project.
- When you need to define a clear development plan for a complex MATLAB application.
- When you want to create an App Designer app using either binary or plain-text formats.
Advantages
- Ensures consistent, maintainable application architecture.
- Prevents common pitfalls by enforcing proven layout and background task patterns.
- Provides clear guidance on choosing between App Designer and programmatic approaches.
- Facilitates collaboration by requiring a documented implementation plan before coding.
Usage examples
- “I need to build a MATLAB dashboard to monitor sensor data. What approach should I take?”
- The agent will ask discovery questions about the app's lifespan, team skills, and visual needs, then recommend an architecture and layout archetype before creating a plan.
- “Build me a UIHTML app for visualizing experimental results.”
- The agent will skip the architecture recommendation, ask for any missing requirements, and proceed to generate an implementation plan for a UIHTML-based app.
Install
5 of its files are not mirrored here, so install it from its source:
https://github.com/matlab/matlab-agentic-toolkit/tree/2a7f215e965597acf6ed495beb9e7b8aff45a3be/skills-catalog/matlab-app-building/matlab-build-app
Documentation by MathWorks
MATLAB Build App
Determine the best architecture for a MATLAB application, produce an implementation plan grounded in internal references, and execute the build. For UIFigure apps, optionally serialize into an App Designer format.
When to Use This Skill
Use this skill when:
User wants to build a MATLAB app, GUI, or interactive tool
User asks "how should I build this app?" or "which approach should I use?"
User describes an application — with or without specifying an implementation path
User mentions: MATLAB app, GUI, uifigure, uihtml, interactive tool, dashboard, visualization app, App Designer, .mlapp, plain-text app
User describes spatial layout needs: dashboard, control panel, sidebar, tabs, wizard, stepper, canvas, workspace
User has already chosen a path (e.g., "build me a uihtml app") — handle directly
When Not to Use
The request is purely about MATLAB computation with no UI component
The user is asking about an existing app they want to modify → read
references/editing-guide.mdThe user wants to convert between App Designer formats (e.g., .mlapp to plain-text or vice versa) → tell them to use File > Save As in App Designer. Do NOT attempt the conversion programmatically.
Critical Rules
MUST ask discovery questions before recommending a path — unless the user already specified one
MUST confirm the path choice with the user before producing the implementation plan
MUST produce an implementation plan grounded in internal references before writing any code
MUST write the plan to a file (
<app-name>-plan.md) in the working directoryNEVER recommend a path without understanding the user's constraints (unless path was pre-specified)
NEVER apply dark mode, custom colors, or visual themes unless the user explicitly requests them
ALWAYS present the recommendation as guidance, not a mandate — the user decides
MUST choose archetype based on the user's primary task, not aesthetics
NEVER treat two archetypes as equals within one app — one is always the primary container
UIFigure app: MUST use
uigridlayoutfor all structural layout — neverPosition-based sizingUIHTML/web app: MUST use CSS Grid or Flexbox for chrome — no absolute positioning for structural panels
The chrome (header, sidebar, tabs, step indicator) MUST remain spatially stable
Background execution (any operation taking more than ~1 second) requires a class-structured app and MUST be read up first in
references/background-tasks.md— do NOT hand-writeparfeval/backgroundPool/DataQueueplumbing from memory. For an App Designer app (.mlappor plain text) you MUST use theaddBackgroundTaskverb and NEVER hand-roll the infrastructure or reuse the standalone-programmatic manual wiring. For a standalone programmatic app, use the MVVM structure per that doc. A flat nested-function app that needs background work MUST escalate to one of these.App Designer serialization: MUST read
references/app-designer/agent-guide-shared.md(covers ownership models, editing discipline, property rules, quoting) plus the format-specific guide (agent-guide-mlapp.mdoragent-guide-plaintext.md) before building. Those docs are the single source of truth; do not attempt to edit app files without reading them first.
Workflow
User request arrives
│
├── Path NOT specified → Full Discovery
│ │
│ ▼
│ Ask Q1-Q4 (purpose, lifespan, polish, team skills)
│ │
│ ▼
│ Identify layout archetype
│ │
│ ▼
│ Recommend UIFigure vs UIHTML based on Q1-Q4
│ │
│ ▼
│ *** STOP: Present recommendation and WAIT for user confirmation ***
│ (Do NOT produce a plan or read references until user says yes)
│ │
│ ▼
│ Ask Q5 (FINAL): Serialization format? (UIFigure apps only)
│ │
│ ├── App Designer → release check, present format choice,
│ │ continue with UIFigure references + add serialization step
│ │
│ └── Standalone programmatic → no extra step
│ │
│ ▼
│ Produce Implementation Plan → write to <app-name>-plan.md
│ │
│ ▼
│ User reviews plan → confirms or adjusts
│ │
│ ▼
│ Begin building (read references per the plan)
│
└── Path IS specified
│
├── Serialization explicit (e.g., ".mlapp app", "plain-text app")
│ → Skip Q5, proceed with stated format
│ → Ask only missing requirements + archetype if unclear
│ → Produce Implementation Plan
│
├── Architecture explicit but serialization not stated
│ (e.g., "build me a uifigure app")
│ → Still ask Q5 (serialization) before planning
│ → Ask missing requirements + archetype if unclear
│ → Produce Implementation Plan
│
└── UIHTML or standalone programmatic explicit
→ Skip Q5 entirely (not applicable)
→ Ask missing requirements + archetype if unclear
→ Produce Implementation PlanDiscovery Questions
Ask these conversationally — not as a rigid checklist. Gather requirements and layout intent first; ask the serialization question last, once you understand what the app needs to do.
Core Questions
1. What does the app do and who is it for?
What will this app do? Who will use it?
Listen for archetype signals: "dashboard", "control panel", "step-by-step", "workspace".
2. Is this a quick tool or something you'll maintain over time?
Is this meant to be maintained and evolved, or is it more of a quick, proof-of-concept tool?
Maintained → proceed to remaining questions; team skills matter
Ephemeral → favor UIHTML app path; can skip team question
3. How polished does the UI need to look?
Are standard MATLAB buttons, sliders, tables, and plots enough? Or do you need custom visuals — branded, animated, or visually richer?
Standard controls sufficient → UIFigure app signal
Custom visuals needed → UIHTML app signal
4. Who will work on this app going forward?
Will this be maintained by people comfortable only with MATLAB, or by people also comfortable with web technologies?
Only ask for maintained apps.
MATLAB-only team → UIFigure app
Web-comfortable team → UIHTML app
5. (FINAL) How should this app be saved?
Do you want this app to open and edit inside App Designer, or do you prefer a standalone programmatic file (no App Designer dependency)?
This is a serialization question, not an architecture question. The app's structure (components, layout, callbacks) is designed using UIFigure knowledge regardless of the answer. Q5 only determines how that design is persisted to disk.
App Designer →
.mlappor plain-text.m+.xml. Check release (plain-text needs R2026b+). If Q3 indicated custom visuals, note that App Designer constrains visual customization vs UIHTML and confirm.Standalone programmatic → programmatic
.mcode, no App Designer dependency. MVVM may warrant multiple files; that is decided at build time.
Follow-up Probes (only if path isn't clear)
Interaction style: Real-time feedback needed? → slight UIFigure lean
Existing work: Existing MATLAB UI code? → UIFigure. Existing web assets? → UIHTML
Distribution: May need to work outside MATLAB? → UIHTML
Long computations: Will any operation take more than a second? → use a background task (read
references/background-tasks.md). Note this requires a class-structured app: App Designer serialization, or MVVM if standalone programmatic. Heavy, long, or one-shot work is ALWAYS a background task, never a timer; a timer fits only scheduled, cheap-per-tick work that must update the UI on the main thread (readreferences/timers.md).
Path Decision Logic
Two stages: first determine architecture (Q1-Q4), then determine serialization format (Q5).
Stage A — Architecture (from Q1-Q4):
Ephemeral app? → Strong default to UIHTML app
Maintained app? → Weigh signals:
Strong UIFigure signals:
✓ Maintained + MATLAB-only team
✓ Standard UI components sufficient
✓ Tight real-time interaction
✓ Existing MATLAB UI code
Strong UIHTML signals:
✓ Ephemeral (regardless of team)
✓ Maintained + web-skilled team
✓ Custom visuals, branded look, animations
✓ Existing web assets
✓ Future portability outside MATLAB
UIFigure + uihtml accent:
Mostly standard app but needs one custom visualization panel.
Stage B — Serialization format (Q5, UIFigure apps only):
Q5 = Standalone → programmatic code (no App Designer). Done.
Q5 = App Designer → adds a serialization layer ON TOP of Stage A.
UIFigure references still apply in full. Serialization is IN ADDITION TO,
not instead of, the normal UIFigure build.
→ Check release, present format choice (plain text vs .mlapp).
→ If Stage A indicated UIHTML for custom visuals, surface the trade-off:
App Designer constrains custom visuals — confirm before committing.Weighting: 1. Architecture (Q1-Q4) 2. Serialization (Q5, applies only to UIFigure)
Paths at a Glance
| Architecture | Key strength |
|---|---|
| UIFigure app | Single language, any MATLAB dev can maintain |
| UIHTML app | Full visual control, rich interactivity |
| UIFigure + accent | Best of both when you need one custom visual |
For UIFigure apps, three serialization formats:
| Format | Trade-off |
|---|---|
Programmatic .m | Standalone, no App Designer dependency |
App Designer .mlapp | Binary, works on any release, opens in App Designer |
App Designer plain text .m + .xml | Source-controllable, AI-editable, R2026b+ |
Layout Archetype
Pick based on the user's primary task:
| Archetype | Primary user task | Structure |
|---|---|---|
| Dashboard | Monitor and compare at a glance | KPI cards + charts + table; read-only |
| Explorer | Adjust parameters, observe live results | Sidebar controls + live display |
| Tabbed | Navigate between independent sections | Tab bar + content panels |
| Wizard | Complete a sequential workflow | Step indicator + one-step-at-a-time |
| Canvas | Create, inspect, or edit spatial artifacts | Central workspace + tools |
If unclear, ask: "Will users mostly be watching results (Dashboard), tweaking controls (Explorer), switching sections (Tabbed), going through steps (Wizard), or working on a central figure (Canvas)?"
Implementation Plan
After confirming path and archetype, produce a plan and write it to <app-name>-plan.md in the working directory. Present a concise summary inline with the approval request.
Plan Template
**Implementation Plan: [App Name]**
**Architecture:** [UIFigure app / UIHTML app / UIFigure + accent]
**Serialization:** [Programmatic .m / App Designer .mlapp / App Designer plain text]
**Layout:** [Archetype] — [one sentence describing spatial structure]
**Structure:**
- [Major panel/area — spatial description]
- [e.g., "Left sidebar (220px) with date range picker, category filter, refresh button"]
- [e.g., "Main area with line chart showing filtered results"]
**Key behaviors:**
- [User-visible interaction effects]
**Internal references that will be used:**
| Reference | Role in this app |
|-----------|-----------------|
| `references/path/file.md` | [what it provides] |
| ... | ... |
**External skills:**
- `matlab-build-chart` — [role, if applicable]
- `matlab-apply-theme` — [role, if applicable]
**File organization:**
[app directory tree]
**Implementation sequence:**
1. [First reference read] — [what it establishes]
2. [Next reference] — [what it adds]
3. [...]Each reference pointer carries provenance: [src: references/uifigure/grid-layout.md §Spanning]
Plan Quality Checks
Before presenting:
[ ] Every reference listed has a concrete role
[ ] Structure uses spatial terms the user can visualize
[ ] Behaviors describe user-visible effects
[ ] Implementation sequence shows a logical build order
[ ] Plan is specific enough for "yes, that's what I want" or "change X"
Reference Router
After plan approval, read the relevant internal references to execute the build.
UIFigure Path
| When building... | Read |
|---|---|
| Grid layout, sizing, components | references/uifigure/guide.md |
| Grid details (spanning, collapse) | references/uifigure/grid-layout.md |
| Panels, tab groups, nested grids | references/uifigure/containers.md |
| Component reference (controls, display) | references/uifigure/components.md |
| Callback patterns, data sharing | references/uifigure/callbacks.md |
| Layout recipes (sidebar, form, split) | references/uifigure/layout-patterns.md |
| MVVM architecture (complex apps) | references/uifigure/mvvm-guide.md |
| View binding patterns | references/uifigure/mvvm-view-binding.md |
| ViewModel testing | references/uifigure/mvvm-testing.md |
| Background execution — concepts + routing (read first) | references/background-tasks.md |
| Background execution — standalone programmatic (MVVM) wiring | references/uifigure/mvvm-background-tasks.md |
| Timers — periodic main-thread work (NOT for offloading heavy work) | references/timers.md |
UIHTML Path
| When building... | Read |
|---|---|
| MATLAB-JS bridge, setup(), events | references/uihtml/bridge-guide.md |
| Communication patterns (4 patterns) | references/uihtml/communication-patterns.md |
| Data type conversion | references/uihtml/data-types.md |
| Platform constraints | references/uihtml/platform-limitations.md |
| uihtml creation, hybrid layouts | references/uihtml/setup.md |
| Error handling (both sides) | references/uihtml/error-handling.md |
| JS MVVM architecture | references/uihtml/mvvm-guide.md |
| Observable/Computed classes | references/uihtml/mvvm-observable-classes.md |
| JS View binding | references/uihtml/mvvm-view-binding.md |
| JS coding patterns, modules | references/uihtml/js-coding-guide.md |
| JS testing & debugging | references/uihtml/js-testing-debugging.md |
| Chart.js setup & patterns | references/uihtml/charting-guide.md |
| Chart type selection | references/uihtml/chart-type-selection.md |
| Chart.js initialization | references/uihtml/chartjs-setup.md |
| Line & bar charts | references/uihtml/line-bar-charts.md |
| Doughnut & scatter charts | references/uihtml/doughnut-scatter-charts.md |
| Chart data updates | references/uihtml/chart-updates.md |
| Chart ↔ bridge integration | references/uihtml/chart-bridge-integration.md |
| Chart performance | references/uihtml/chart-performance.md |
| CSS styling & tokens | references/uihtml/styling-guide.md |
| Component CSS (buttons, inputs) | references/uihtml/component-styles.md |
| CSS layout patterns | references/uihtml/css-layout-patterns.md |
| Brand design tokens | references/uihtml/brand-design-tokens.md |
| Dark mode | references/uihtml/dark-mode.md |
| External color schemes | references/uihtml/external-color-schemes.md |
App Designer Serialization (UIFigure apps only)
Important: App Designer serialization is an additional step AFTER the normal UIFigure build. You MUST also read the UIFigure path references above (components, grid-layout, callbacks, containers, layout-patterns) to design the app's structure. The serialization references below cover only how to persist that design to disk.
Both formats use the same AppDesignerAgentInterface API via the bundled scripts/ tool; only the file extension passed to create()/open() differs.
Read first (both formats):
| When building... | Read |
|---|---|
Verbs, build sequence, editing, save()/validate()/finalize(), inspect() | references/app-designer/agent-guide-shared.md |
| Background execution — concepts + routing (read first) | references/background-tasks.md |
Background execution — addBackgroundTask verb reference | references/app-designer/background-tasks.md |
Plain-text (.m + .xml):
| When building... | Read |
|---|---|
Ownership model, .xml authoring, flow, gotchas | references/app-designer/agent-guide-plaintext.md |
| classdef structure & block ordering rules | references/app-designer/plaintext/classdef-template.md |
| XML structure and which properties to set | references/app-designer/plaintext/xml-template.md |
| Per-component property names, types, defaults | references/app-designer/plaintext/property-reference.md |
| Authoritative XML schema | references/app-designer/plaintext/matlabAppSchema.xsd |
Binary .mlapp:
| When building... | Read |
|---|---|
Build flow, native property values, save()/finalize() | references/app-designer/agent-guide-mlapp.md |
Editing an Existing App
| When editing... | Read |
|---|---|
| Any existing app (determines format, routes to approach) | references/editing-guide.md |
Archetype References
| Archetype | File |
|---|---|
| Dashboard | references/archetypes/dashboard.md |
| Explorer | references/archetypes/explorer.md |
| Tabbed | references/archetypes/tabbed.md |
| Wizard | references/archetypes/wizard.md |
| Canvas | references/archetypes/canvas.md |
External Skills (invoked, not read)
matlab-build-chart— when the app includes MATLAB plots (UIFigure path or UIFigure+accent)matlab-apply-theme— when the app needs dark mode, brand colors, or custom palettes (UIFigure path)
After Plan Approval
All UIFigure apps (regardless of serialization format)
Read the archetype reference for the spatial skeleton
If MVVM warranted (Explorer, Wizard, Canvas, or complex apps): read the architecture reference
Read the UIFigure guide to establish layout (grid, components, containers)
Layer in path-specific references per the implementation sequence
For visual polish: invoke
matlab-apply-theme(UIFigure) or read styling references (UIHTML accent)For charts: invoke
matlab-build-chart(UIFigure) or read charting references (UIHTML accent)If App Designer serialization: read the app-designer reference docs (shared guide + format-specific guide), then use the
AppDesignerAgentInterfaceAPI to persist the app. The docs contain the complete build flow, property rules, and editing discipline.
UIHTML path
Read the archetype reference for the spatial skeleton
If MVVM warranted: read the JS MVVM architecture reference
Read the bridge guide to establish the MATLAB-JS connection
Layer in path-specific references per the implementation sequence
For visual polish: read styling/CSS references
For charts: read charting references
Format Fallback
If the plain text path produces errors that cannot be resolved (e.g., version mismatch, unsupported component), fall back to the .mlapp path — the same verbs drive both formats; re-create the handle with a .mlapp extension. Only fall back to fully programmatic UIFigure code as a last resort.
Recommendation Template
Recommended architecture: [UIFigure app / UIHTML app / UIFigure + accent]
Based on what you've described:
[Key signal 1]
[Key signal 2]
[Key signal 3, if applicable]
What this means: [1-2 sentences on development experience]
Trade-off: [Main thing they'd give up vs the other path]
Serialization (if App Designer): I'd recommend [.mlapp / plain text] because [reason]. This doesn't change the app's structure, just how it's saved.
Want to proceed with this approach, or would you prefer the other path?
Copyright 2026 The MathWorks, Inc.
Files
- ATTRIBUTION.mdAdded by AutoGineer1.0 KiB
- LICENSELicence file added by AutoGineer1.4 KiB
- SKILL.md20.9 KiB
- assets/colors_and_type.css12.8 KiB
- assets/minimal-bridge.html1.3 KiB
- assets/minimal-bridge.m1.2 KiB
- assets/standalone-chart-template.html11.5 KiB
- manifest.yaml672 B
- references/app-designer/agent-guide-mlapp.md15.7 KiB
- references/app-designer/agent-guide-plaintext.md14.7 KiB
- references/app-designer/agent-guide-shared.md27.1 KiB
- references/app-designer/background-tasks.md4.8 KiB
- references/app-designer/plaintext/classdef-template.md11.3 KiB
- references/app-designer/plaintext/matlabAppSchema.xsd16.8 KiB
- references/app-designer/plaintext/property-reference.md45.1 KiB
- references/app-designer/plaintext/xml-template.md12.3 KiB
- references/archetypes/canvas.md8.0 KiB
- references/archetypes/dashboard.md6.4 KiB
- references/archetypes/explorer.md5.7 KiB
- references/archetypes/tabbed.md6.6 KiB
- references/archetypes/wizard.md9.7 KiB
- references/background-tasks.md6.9 KiB
- references/editing-guide.md3.3 KiB
- references/timers.md2.1 KiB
- references/uifigure/callbacks.md3.0 KiB
- references/uifigure/components.md1.7 KiB
- references/uifigure/containers.md1.3 KiB
- references/uifigure/grid-layout.md4.9 KiB
- references/uifigure/guide.md5.3 KiB
- references/uifigure/layout-patterns.md1.5 KiB
- references/uifigure/mvvm-background-tasks.md3.9 KiB
- references/uifigure/mvvm-guide.md11.1 KiB
- references/uifigure/mvvm-testing.md2.8 KiB
- references/uifigure/mvvm-view-binding.md5.4 KiB
- references/uihtml/brand-design-tokens.md5.6 KiB
- references/uihtml/bridge-guide.md7.0 KiB
- references/uihtml/chart-bridge-integration.md7.5 KiB
- references/uihtml/chart-performance.md5.0 KiB
- references/uihtml/chart-type-selection.md3.1 KiB
- references/uihtml/chart-updates.md4.7 KiB
- references/uihtml/charting-guide.md3.9 KiB
- references/uihtml/chartjs-setup.md4.0 KiB
- references/uihtml/communication-patterns.md4.2 KiB
- references/uihtml/component-styles.md4.3 KiB
- references/uihtml/css-layout-patterns.md5.0 KiB
- references/uihtml/dark-mode.md3.8 KiB
- references/uihtml/data-types.md2.7 KiB
- references/uihtml/doughnut-scatter-charts.md6.0 KiB
- references/uihtml/error-handling.md4.6 KiB
- references/uihtml/external-color-schemes.md2.6 KiB
- references/uihtml/js-coding-guide.md11.2 KiB
- references/uihtml/js-testing-debugging.md2.9 KiB
- references/uihtml/line-bar-charts.md5.2 KiB
- references/uihtml/mvvm-guide.md11.0 KiB
- references/uihtml/mvvm-observable-classes.md2.9 KiB
- references/uihtml/mvvm-view-binding.md5.3 KiB
- references/uihtml/platform-limitations.md2.5 KiB
- references/uihtml/setup.md3.9 KiB
- references/uihtml/styling-guide.md8.1 KiB
- scripts/+module/+backgroundtask/BackgroundTaskModule.p(opens on GitHub)Not mirrored: opens on GitHub2.5 KiB
- scripts/+module/+backgroundtask/templates/cancelBackground.m357 B
- scripts/+module/+backgroundtask/templates/handleBackgroundComplete.m1.4 KiB
- scripts/+module/+backgroundtask/templates/safeProgress.m734 B
- scripts/+module/+backgroundtask/templates/startBackground.m1.2 KiB
- scripts/+module/AppBuilderModule.p(opens on GitHub)Not mirrored: opens on GitHub242 B
- scripts/AppDesignerAgentInterface.p(opens on GitHub)Not mirrored: opens on GitHub12.6 KiB
- scripts/MlappBackend.p(opens on GitHub)Not mirrored: opens on GitHub9.4 KiB
- scripts/PlainTextBackend.p(opens on GitHub)Not mirrored: opens on GitHub10.2 KiB
references/background-tasks.md
Background Tasks (Background Execution in Apps)
Run work off the main thread so the UI stays responsive, using the parfeval +
backgroundPool pattern (with DataQueue for progress streaming and cancellation).
This hub covers the concepts and routing shared by both serialization paths; the
per-format API detail lives in the two docs linked below.
STOP — pick your path before writing any code
The two paths are NOT interchangeable. Applying the wrong one produces a broken or unmaintainable app.
App Designer app — a
.mlappor plain-text.m+.xmlbuilt throughAppDesignerAgentInterface(create()/open())? → You MUST use theaddBackgroundTaskverb. Readreferences/app-designer/background-tasks.md. Do NOT hand-wire the pattern.Standalone programmatic app (plain
.m, no App Designer)? → Wire the templates in by hand perreferences/uifigure/mvvm-background-tasks.md(requires the MVVM structure).If you are unsure which you are building, you do not yet know enough to add a background task — resolve the serialization path first (see the main SKILL.md).
Background execution requires a class-structured app. A background task needs
durable per-task state (Future, Queue, Running, StopRequested), shared
methods reachable from multiple callbacks, and a CloseRequestFcn that cancels
in-flight work on close. A flat nested-function UIFigure app that grows a background
task is your signal to escalate: build it as an App Designer app (the verb does
everything) or, if staying standalone programmatic, structure it as MVVM.
Editing an existing programmatic app? Restructuring it to MVVM requires explicit
user consent first — see the gate in references/editing-guide.md.
When to use
Best for a single long-running task that keeps the UI responsive while it runs. Use when any of these are present:
Long computation, training, export, download, processing, number crunching
Progress indicators, streaming results, or partial updates
A cancel button for in-progress work
"Keep the app responsive" / "don't freeze the UI"
Any callback body taking more than ~1 second
Do NOT use for:
Work that completes in under a second (run it inline)
Work that must block interaction (use a progress dialog)
The shared infrastructure (single source of truth)
Both paths run identical background-execution code, maintained as template files under
scripts/+module/+backgroundtask/templates/:
| Method | Role |
|---|---|
startBackground(app, taskName, varargin) | Launches the named task (args after the name). Agent/user callable. |
cancelBackground(app, taskName) | Sets StopRequested, cancels a running task (no-op if idle). Agent/user callable. |
handleBackgroundComplete(app, taskName, future) | Routes completion to the task's CompleteFcn; try/catch surfaces user errors via uialert + stderr. Internal. |
safeProgress(app, taskName, msg) | Dispatches DataQueue messages to the ProgressFcn with error protection; routes Finished to handleBackgroundComplete. All progress callbacks complete before completion fires. Internal. |
cleanupBackground(app) | Cancels all task futures (called on close). Internal. |
These are written against a single host object named app that owns the figure
(app.UIFigure), the per-task state (app.(taskName)), and the other infra methods.
The App Designer verb inserts them verbatim; the programmatic path drops them into the
View verbatim. Because both consumers use the files unchanged, a fix lands in one place.
The one piece you always author is the compute function (Fcn) that runs on the
worker. Everything else is infrastructure.
Callback signatures
| Callback | Signature | Notes |
|---|---|---|
| Compute (no progress) | function [result] = fcn(data, ...) | Any number of inputs. Output count auto-detected via nargout. |
| Compute (with progress) | function [result] = fcn(data, ..., sendProgress) | sendProgress injected as the last arg. Call sendProgress(value) to stream. |
| Completion | function completeFcn(app, result, error, wasCancelled) | result is [] when cancelled/errored; error is [] on success. |
| Progress | function progressFcn(app, data) | data is whatever was passed to sendProgress. |
Output handling (nargout of the compute function): 0 → result is [];
1 → the value directly; >1 → a cell array.
Limitations
Assume a single worker; don't depend on more. Worker count depends on the MATLAB license, machine, and (for shared apps) where it runs. Some environments run tasks in parallel, but design as if there is only one. You cannot launch a parallel pool inside a
backgroundPoolworker (parfor,UseParallel=true, etc. are unavailable there).Tasks do not yield. A long-lived task occupies its worker for its whole duration. With one worker, a second task cannot start until the first finishes or is cancelled. Prefer breaking long work into shorter discrete tasks if other work must progress.
Thread-based workers. Only thread-worker-compatible functions/classes run in the background. Unsupported calls error clearly through the CompleteFcn error path. If no thread-supported equivalent fits, the work is not a background-task candidate.
Timing is not guaranteed.
startBackgroundonly queues the work. Do not write UI/logic that depends on the task having started or on how long it takes to begin. Reflect state changes in the CompleteFcn/ProgressFcn callbacks, not by assuming timing.No bidirectional communication. Args flow in at launch; progress streams back out. There is no supported way to send commands to a running worker. For interactive control, restructure as discrete short-lived tasks or wire
PollableDataQueuemanually.Error handling is built in. CompleteFcn errors surface via
uialert+ stderr; ProgressFcn errors log to stderr and disable further progress (task keeps running). Do NOT add try/catch to callback bodies — the infrastructure already protects them.High-throughput progress:
drawnow limitrateis not auto-inserted. If streaming30 updates/sec, add
drawnow limitrateat the end of the progress callback.
Multiple independent tasks
An app can have multiple tasks, each with its own Future and state. Add one per distinct
workflow (App Designer: one addBackgroundTask call each; programmatic: one struct
property + one cleanupBackground line each). Starting a task that is already running
errors: "Background work is already running. Cancel it first." See per-format docs.
Implementation note
AppBuilderModule.p and BackgroundTaskModule.p are internal implementation
details and should never be called directly. Always go through the
AppDesignerAgentInterface API (create, open, addBackgroundTask, finalize).
Copyright 2026 The MathWorks, Inc.
Requirements and compatibility
Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta
“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.
Needs: Declares allowed tools
Security
No issues found by our static scan· No issues found
A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.
Attribution and licence
- Created by
- MathWorks
- Publisher
- MathWorks (Official)
- Source
- matlab/matlab-agentic-toolkit · skills-catalog/matlab-app-building/matlab-build-app · commit 2a7f215
- Licence
- BSD 2-Clause License (redistribution allowed)
- LICENSE (added by AutoGineer from LICENSE.md)
- ATTRIBUTION.md (written by AutoGineer)
- Synced
AutoGineer is not affiliated with or endorsed by MathWorks.
Get it from the source
5 of this skill’s files are not mirrored here, so there is no ZIP. Get the complete skill from its source: matlab/matlab-agentic-toolkit