#Skills

Module Federation now exposes one unified mf skill.

You do not need to pick from a long list of separate skills, and you do not need to read the docs first and then translate them back to your agent. The simpler approach is to install this one skill and let the agent decide whether it should read docs, inspect config, or troubleshoot a problem.

When a problem exists only in the live page runtime, the skill uses the Divebell MF Extension for structured Remote, Shared, Bridge, and module performance evidence. The browser is only the entry point and loading trigger. Divebell is installed on the agent machine, not in the application.

If you have not read it yet, start with AI Quick Start.

#Installation

npx skills add module-federation/agent-skills --skill mf -y

After that, the main entry is:

/mf <sub-command or question>

If you cannot use the CLI, you can also copy the directory manually from GitHub:

https://github.com/module-federation/agent-skills/tree/main/skills/mf

#How to use it

mf has one entry point, but it contains multiple capabilities. You can either provide an explicit sub-command or just ask in plain language.

With explicit sub-commands:

/mf docs How do I use Module Federation in Modern.js?
/mf integrate
/mf type-check
/mf runtime-error RUNTIME-008
/mf observability

With natural language:

/mf What's the difference between singleton and requiredVersion in shared?
/mf Help me figure out why this project is not pulling remote types
/mf Add Module Federation to the current project
/mf I saw a Module Federation console error with traceId mf-...

#What mf supports

#The most common workflows

#1. Read the docs before answering

This is the most common use case, and the best place to start:

/mf docs What's the difference between singleton and requiredVersion in shared?

That makes the agent read the latest docs first instead of guessing from outdated memory.

#2. Let the agent integrate it for you

If you want to add Module Federation to an existing project:

/mf integrate

The agent will inspect your current project and decide what to do next.

#3. Let the agent troubleshoot directly

If you already have a concrete problem:

/mf type-check
/mf shared-deps
/mf config-check
/mf runtime-error RUNTIME-008
/mf observability

At that point, the goal is not for you to manually investigate first. The goal is to let the agent start diagnosing immediately.

#4. Let the agent debug a live MF page

For a one-off investigation, give the target URL and user path to the skill:

/mf observability
Open https://example.com with Divebell, reproduce the failing checkout flow,
and inspect the Remote and Shared loading chain.

The skill discovers the installed Divebell CLI and MF Extension Skills, opens the page with divebell open <url> --mf, and selects the smallest relevant divebell mf command. Page interaction is used only to trigger the relevant lazy load; conclusions should come from structured Remote, Shared, Bridge, or module performance evidence.

See Troubleshoot Runtime Issues for the installation commands, command map, Remote proxy workflow, and evidence boundaries.

#5. Let the agent read retained observability

Install the Observability Plugin in your app when reports must be collected continuously, uploaded from production, or written by Node/SSR. When a browser console error prints a traceId and read: command, pass the accessible report source to the skill:

/mf observability
I saw this Module Federation console error:

[Module Federation] Observability report generated
traceId: mf-...
read: window.__FEDERATION__.__OBSERVABILITY__["runtime_host"].getReport("mf-...")

Please read the report and fix the issue.

For Node, SSR, or production uploads:

/mf observability
Read .mf/observability/latest.json and explain the likely owner and fix.

#The one thing you really need to remember

You do not need to read the docs first and then tell the agent what to do.

Install mf, then let the agent read, inspect, decide, and act.