Matlab Use Experiment Manager — Agent Skill by MathWorks

From the author

Create, modify, or delete experiments in Experiment Manager with live UI sync. TRIGGER when: user asks to create an experiment from code/script, wants to sweep parameters, asks to add/delete/edit experiments, or wants parameter changes reflected in the UI. Combines experiment creation (from code analysis) with frontend UI synchronization.

— the description by MathWorks

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The Experiment Manager Assistant enables users to create, modify, and manage experiments within the MATLAB Experiment Manager app through code-driven workflows. It bridges the gap between script-based experiment definition and the graphical user interface, ensuring that changes to experiment parameters, strategies, or project structures are reflected in the Experiment Manager UI in real-time without requiring application restarts.

What it does

Generates experiment functions from user code or problem descriptions.
Synchronizes experiment modifications directly with the Experiment Manager UI.
Supports creation of general-purpose, built-in training, and custom training experiments.
Facilitates parameter sweeps and strategy configuration.
Provides lifecycle management including adding, deleting, renaming, and duplicating experiments.
Integrates with MATLAB's internal API to maintain state consistency.

When to use it

When you need to create an experiment from existing MATLAB code or a specific problem description.
When you want to perform parameter sweeps or change experiment strategies programmatically.
When you need to add, remove, or rename experiments within an active Experiment Manager session.
When you want to ensure your code-based experiment definitions remain in sync with the visual interface.

Advantages

Eliminates manual UI configuration for repetitive experiment setup tasks.
Maintains live synchronization between code definitions and the Experiment Manager browser.
Supports complex experiment types including Bayesian Optimization and custom training loops.
Reduces errors by programmatically validating experiment structures and function signatures.

Usage examples

“Create a new experiment for my parameter sweep using these ranges.”: The agent analyzes the provided parameters, generates the necessary experiment function, and updates the Experiment Manager UI to reflect the new experiment configuration.
“Rename my current experiment to 'Optimization_Run_01' and update the learning rate parameter.”: The agent renames the experiment in the file system and the UI tree, while simultaneously updating the parameter value within the experiment definition.

Install

3 of its files are not mirrored here, so install it from its source:
https://github.com/matlab/matlab-agentic-toolkit/tree/2a7f215e965597acf6ed495beb9e7b8aff45a3be/skills-catalog/ai-and-statistics/matlab-use-experiment-manager

Documentation by MathWorks

Experiment Manager Assistant

Create experiments from user code or problem descriptions AND keep the Experiment Manager (EM) UI in sync — adding/deleting/editing experiments updates the browser tree and editor live, without restarting EM.

When to Use

Use this skill when the user wants to manage an experiment in MATLAB Experiment Manager: create an experiment from a script or description, sweep parameters, add/delete/edit experiments, rename or duplicate an experiment, or ask about an experiment's strategy, parameters, or values — and have any change reflected live in the EM UI.

When NOT to Use

Do not use this skill for requests that are not about managing an Experiment Manager experiment. Hand off instead:

Train, design, or build a neural network → matlab-train-network.
Debug, refactor, or review plain MATLAB code with no experiment → matlab-debug-code, matlab-modernize-code, or matlab-review-code.
Build an app, dashboard, or GUI from scratch → matlab-build-app.
Make a live script or report → matlab-create-live-script.
Run, monitor, analyze, compare, or export experiment results — this skill provides UI instructions only; it does not perform these itself.

Setup

When this skill is invoked, add the skill's scripts folder to the MATLAB path so that manageExperimentHelpers is available. Run this via the MATLAB MCP evaluate_matlab_code tool:

addpath('<skill-base-directory>/scripts');

This skill manages experiments inside the existing MATLAB Experiment Manager app. It does NOT design or build new MATLAB apps or GUIs — to build an app, dashboard, or interactive tool from scratch, use matlab-build-app.

How to use this skill

Always read common.md first. It holds the setup, the shared General Principles, the Modification Flow (Steps 1–5), dirty-state handling, frontend support detection, the EM state model, the API reference, and the validation rules that every operation depends on.
Then open the reference for the specific task the user is asking for.

Tasks

Pick the one operation that matches the request:

#	Task	Reference
1	Create — from code or description, generate function, open in EM	create.md
2	Load Project — open project in EM, list experiments, offer actions	load-project.md
3	Update — modify parameters, strategy, functions, supporting files	update.md
4	Delete — remove experiment tab, results, tree entry, and file	delete.md
5	Rename — update name, tree label, and tab title	rename.md
6	Duplicate — copy with a custom name	duplicate.md

Two buckets:

Project Management — Load Project, Create
Experiment Management — Update, Rename, Duplicate, Delete

For querying Experiment Manager UI state (answering questions about strategy, parameters, or values), see the Query Experiment State section in common.md.

Files

ATTRIBUTION.mdAdded by AutoGineer1.0 KiB
LICENSELicence file added by AutoGineer1.4 KiB
SKILL.md3.7 KiB
manifest.yaml521 B
references/common.md22.6 KiB
references/create.md15.4 KiB
references/delete.md2.3 KiB
references/duplicate.md2.6 KiB
references/load-project.md700 B
references/rename.md2.6 KiB
references/update.md14.4 KiB
scripts/+experiments/+internal/AgentFrontendBridge.p(opens on GitHub)Not mirrored: opens on GitHub963 B
scripts/+experiments/+internal/AppController.p(opens on GitHub)Not mirrored: opens on GitHub299 B
scripts/manageExperimentHelpers.p(opens on GitHub)Not mirrored: opens on GitHub2.9 KiB

references/delete.md

Operation: Delete Experiment

Shared rules, setup, the Modification Flow, validation, and API are in common.md. Read it first.

Remove an experiment: its tab, results, tree entry, and .mat file.

Before deleting: Check file exists. If not, list available experiments. Never batch a delete with other operations.

Confirm (Step 4)

Always confirm: "Are you sure you want to delete [name]? This will remove the experiment and all its results."

Always state the irreversibility warning in your reply — even when you proceed without waiting. If the request is explicit and you are proceeding directly (e.g. non-interactive), you MUST still name the experiment and warn that the deletion is permanent before performing it — e.g. "Deleting trainNet — this permanently removes the experiment, all its results, and its file, and cannot be undone." Never delete silently or skip the warning just because you are not pausing for a click.

Execute (Step 5)

clear classes
result = experiments.internal.AppController.deleteExperiment(fullfile('<projectPath>', '<experimentName>.mat')); disp(result);
% Force the tree/tabs to refresh so the deleted experiment disappears from the UI.
result = experiments.internal.AppController.syncTree(); disp(result);

This deletes results, closes tabs, removes from tree, and deletes the .mat file.

Confirm the UI is clean before reporting success. After the calls above, the tab, tree entry, results, and .mat file must all be gone. If the tree still shows the experiment, call experiments.internal.AppController.syncTree() again. When the calls succeed, report the delete as complete only once the experiment is removed from the UI, not merely from disk — do NOT tell the user to close and reopen the tab or reopen the project manually.

If deleteExperiment returns status: 'timeout' or an error (the frontend did not respond), follow the timeout/fallback rule in common.md (General Principle 11): delete the .mat file directly from disk so the experiment is gone, then in your reply state plainly that the file and tree entry are removed but the experiment's open tab may still be showing, and tell the user to close that tab. Do not claim the tab was closed when the UI never confirmed it.

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.

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.

Wizard
Matt Pocock· Community
Generate interactive bash wizards to guide humans through complex manual setup, migration, or configuration tasks.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITLow riskHas scriptsGitHub stars of the source repository317.5k
Emails
Corey Haines· Community
Design and optimize automated email sequences for onboarding, nurturing, and lifecycle management.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository59k
Conversion Tracking
Corey Haines· Community
A systematic approach to setting up, auditing, and verifying website conversion tracking across major ad platforms and analytics tools.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITLow riskHas scriptsGitHub stars of the source repository59k
Directory Submissions
Corey Haines· Community
A strategic framework for submitting software products to directories to build backlinks, improve domain rating, and drive discovery.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITMedium riskGitHub stars of the source repository59k
Sms
Corey Haines· Community
Plan, build, and optimize compliant SMS, MMS, and WhatsApp marketing programs for high-engagement customer communication.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository59k
Marketing Loops
Corey Haines· Community
Orchestrate and schedule repeatable, self-running marketing workflows that operate on a defined cadence.
Automation and integrationsClaude CodeClaude appsOpenAI Codex+14 more
MITGitHub stars of the source repository59k

Get it from the source

3 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