AnythingLLM Community Hub: Import & Share Agent Skills

The AnythingLLM Community Hub is a public marketplace where teams share and discover pre-built agent skills, system prompts, and slash commands. Instead of building every capability from scratch, you can import vetted community work directly into your instance. This matters because it cuts your setup time in half and lets distributed teams collaborate on shared automation.

What is the AnythingLLM Community Hub?

The AnythingLLM Community Hub is a central repository hosted at hub.anythingllm.com where users publish reusable components for the AnythingLLM platform. Currently it supports three item types: agent skills (custom actions and workflows), system prompts (pre-written instructions for consistent LLM behavior), and slash commands (shortcuts that trigger specific functions). Think of it as GitHub for AnythingLLM templates.

The Hub serves both self-hosted and managed AnythingLLM instances. With 68.1K GitHub stars and a 7.7K-fork community, AnythingLLM has built enough momentum that published skills cover common use cases: customer service workflows, content generation, data retrieval, and team-specific processes. The MIT license ensures everything shared remains open and modifiable.

Why does this matter for your ops? Because reinventing the wheel on every skill is expensive. If another team has already solved "extract data from Slack and summarize daily trends", importing their work saves you days of development. At a 10-50 person company, that time difference is material.

How to Access & Browse the Community Hub

Accessing the Hub is straightforward: navigate to hub.anythingllm.com in your browser. No login required to browse public items. Once there, you can search by keyword, filter by item type (skills, prompts, commands), and read detailed descriptions, documentation, and author notes.

If you are running AnythingLLM desktop, you can also access the Hub directly from the app interface without leaving your workspace. Look for the "Community Hub" or "Skills" section depending on your version. Desktop users get a integrated import flow: search, click "Import", and the platform handles the rest.

Public items show their creator, publication date, update history, and download count. Some items earn a blue "Verified" badge from the AnythingLLM team, which signals basic review and quality checks. This does not mean the item is guaranteed bug-free, but it indicates the team has vetted the code and description.

You can filter by multiple criteria: popularity (most downloaded), recency, or language/category tags that publishers add. Private items do not appear in public search; you only see those if you are invited and authenticated with the right credentials.

Importing Skills & Prompts: A Step-by-Step Guide

Importing from the Community Hub takes five minutes. Here is the process: first, locate the item you want on hub.anythingllm.com and open its detail page. You will see a blue "Import to AnythingLLM" button or an "Import String" you can copy directly.

Click the button (or copy the string), then navigate to your AnythingLLM instance. Go to Settings > Community Hub or your instance's equivalent import section. Paste the import string or authenticate and select the item. The platform will ask where to save it: choose your workspace and confirm.

Once imported, system prompts immediately apply to new conversations in that workspace. Agent skills appear in your agent-builder interface, ready to attach to a workflow or action block. Slash commands activate automatically and are available via "/command-name" in chat.

After import, AnythingLLM downloads and validates the item in your instance. If validation fails, you will see an error message. For self-hosted deployments, a common failure is the configuration flag COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED being disabled by default. We will cover that gotcha below.

Test imported items in a sandbox workspace first, especially if you are running a team instance. Download the skill documentation (usually included in the import), read through the workflow or prompt logic, and run a test interaction before deploying to production.

Evaluating & Trusting Community Content

Not all published items are created equal. Here is how to assess safety and quality before importing. First, check for the blue Verified badge: it signals AnythingLLM's team reviewed the code and matched the description to reality. No verification means it is unvetted, which is not inherently bad, but requires more caution.

Read the item description carefully. Does it actually do what you need? Look for detailed documentation, example workflows, and author contact info. Items with sparse or vague descriptions are higher risk; good publishers explain inputs, outputs, dependencies, and any external API keys required.

Check the download count and publication date. Items with hundreds of downloads and recent updates signal maturity. A skill published six months ago with zero downloads could be abandoned, useful only for niche cases, or just undiscovered.

For agent skills, examine the handler.js file and plugin.json if available. You do not need deep coding skills: just look for red flags like hardcoded API keys, requests to unknown external domains, or manipulation of system files. The AnythingLLM team review catches egregious issues, but they do not do full security audits.

For system prompts, read the full prompt text before importing. Look for prompt injection risks or contradictory instructions that could produce unreliable outputs. A well-written prompt is specific, concise, and includes guardrails (e.g., "do not make up data").

If the item requires external API keys (Slack, database URLs, etc.), verify the documentation explains how to provide them securely. AnythingLLM encrypts stored keys, but leaked credentials at import time are not the platform's fault.

Self-Hosted Gotcha: Enabling Imports on Docker & Bare-Metal

If you run AnythingLLM self-hosted, imports may fail with an error: "Community Hub bundle downloads are not enabled on this instance." This is by design: the config flag COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED defaults to false for security. Admin must explicitly enable it.

To enable imports on Docker: add the environment variable to your docker-compose.yml or docker run command:

COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED=true

For bare-metal installations, set the same variable in your.env file or system environment before starting AnythingLLM. After setting it, restart the container or process. Imports will then work.

Why is this disabled by default? Because downloading and executing code from the internet carries risk if your team does not review items first. Enabling it trusts that your team will vet what they import. If you run a managed hosting service, check whether your provider has already enabled this flag. Opsily's managed AnythingLLM hosting comes with Community Hub downloads pre-configured, so you skip this step entirely.

If you want to keep downloads disabled but still share items within your team, see the next section on private sharing.

Creating & Uploading Your Own Skills to the Hub

If your team has built a skill or prompt that solves a reusable problem, publishing it expands AnythingLLM's ecosystem and gives you visibility. Here is the workflow: first, install the anythingllm-hub-cli tool via npm:

npm install -g anythingllm-hub-cli

Next, create your skill structure locally. An agent skill requires two files: a plugin.json manifest and a handler.js implementation. The manifest declares the skill's name, description, parameters, and version. The handler exports an async function that performs the work.

Get a Connection Key from your hub profile: log into hub.anythingllm.com, go to Settings > Connection Keys, and generate a new key. Copy it; you will use it to authenticate uploads.

From your skill directory, run:

anythingllm-hub-cli login --key <YOUR_CONNECTION_KEY>
anythingllm-hub-cli publish --path./my-skill --type agent_skill

The CLI validates your skill, uploads it to the Hub, and assigns it a public URL. You then choose: publish as public (anyone can see and import) or private (only your team members with your Connection Key can import). Private skills are excellent for internal tools you do not want to share externally.

After upload, the AnythingLLM team may contact you if they want to verify your skill for a badge. Verification typically takes 1-2 weeks. Until then, your skill is public but unverified.

For system prompts and slash commands, the upload process is similar but faster: you can upload directly from the Hub UI without CLI.

Sharing Privately with Your Team

If you run a team instance and want to share skills only within your organization, make items private on the Hub. Private items do not appear in public search and require authentication to import.

When publishing, toggle the "Private" setting in the Hub UI or add --private to your CLI upload command. Then invite your team members and share your Connection Key with them. They add the key to their AnythingLLM instance Settings > Community Hub, and private items become available to them.

This is ideal for internal-only workflows, company-specific system prompts, or skills that reference private APIs. Your team can iterate on the skill in the Hub (versioning is automatic), and every team member imports the latest version when they update their instance.

For production teams, establish a process: code review before publishing, versioning conventions, and deprecation notices for old skills. The Hub does not enforce this, but a loose governance avoids confusion and incompatible imports.

Troubleshooting & Common Issues

Failed to import agent skill: First, verify COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED is set to true if you are self-hosted. Second, check your internet connection; the instance must reach hub.anythingllm.com. Third, inspect your AnythingLLM logs for detailed error messages; "permission denied" usually means the config flag, while "validation failed" means the skill structure is malformed.

Connection Key not working: Connection Keys expire or can be revoked. Log into your Hub profile, verify the key is still active, and regenerate if needed. Ensure you passed the key correctly to your instance (Settings > Community Hub on desktop, or COMMUNITY_HUB_CONNECTION_KEY env var on self-hosted).

Item does not appear in Hub after upload: If you uploaded via CLI, check the command output for errors. If upload completed but the item is not searchable, it may be queued for indexing; wait 5 minutes and refresh. If the item is private, ensure you are logged in with the same account that published it.

System performance after importing many skills: Each imported skill adds a small memory footprint and can increase inference latency if skills run complex logic. Monitor your instance performance after bulk imports. If you notice slowdowns, audit which skills are actively used and uninstall unused ones. 10-20 skills is typical; 100+ starts to matter.

Frequently Asked Questions

What can I share on the AnythingLLM Community Hub?

Currently, you can publish agent skills (custom workflows and actions), system prompts (LLM instructions), and slash commands (chat shortcuts). Coming soon: workspaces, data connectors, and authentication providers. Check the official roadmap for details.

How do I know if a skill is safe to import?

Look for the blue "Verified" badge, which signals AnythingLLM team review. Read the full description and documentation. Check the download count and publication date as proxies for maturity. If the skill requires external APIs, verify the documentation explains secure setup. Test in a sandbox workspace first.

What's a Connection Key, and why do I need one?

A Connection Key is a personal authentication token generated from your Hub profile. Public items do not require it; private items do. You provide the key to AnythingLLM instances on your team so they can import private skills. Treat it like a password: do not share it publicly.

Why does importing an agent skill fail on my self-hosted instance?

The config flag COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED is disabled by default. Set it to true in your environment variables or.env file and restart AnythingLLM. This applies to Docker and bare-metal deployments.

Can I share skills privately with my team?

Yes. When publishing, toggle "Private" in the Hub or add --private to the CLI upload command. Share your Connection Key with teammates, and they can import items you have marked private. Versioning is automatic; every import pulls the latest version.

How do I build and upload a custom agent skill?

Install the anythingllm-hub-cli npm tool. Create a plugin.json manifest and handler.js implementation. Generate a Connection Key from your Hub profile. Run anythingllm-hub-cli login --key <KEY> and anythingllm-hub-cli publish --path./my-skill --type agent_skill. The Hub handles validation and hosting.

Is AnythingLLM Community Hub different for managed vs. self-hosted?

Functionality is the same, but self-hosted requires you to enable downloads via COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED. Managed hosting typically has this pre-configured. When evaluating AnythingLLM hosting, check whether the provider supports Community Hub imports out of the box.

The Bottom Line

The AnythingLLM Community Hub cuts your setup time by letting you import pre-built agent skills, prompts, and commands instead of building everything from scratch. Verified items are reviewed by the AnythingLLM team, but you should still vet what you import, especially in production. Self-hosted teams need to enable the COMMUNITY_HUB_BUNDLE_DOWNLOADS_ENABLED flag to use imports; managed hosting saves you that step. If you are evaluating AnythingLLM hosting options, compare who supports Community Hub out of the box versus who requires manual setup. See Opsily's managed AnythingLLM hosting to skip configuration entirely and start importing skills on day one.