Skip to main content

​
Introduction

Agent publishing is the process of making an agent version available in Agent HUB, a shared library accessible to all ELITEA users. Once published, an agent version is visible to the entire community and can be added to conversations without project membership. Key capabilities:
  • Publish agent versions to Agent HUB through a guided three-step wizard.
  • AI-powered automated validation runs before publishing. It identifies critical issues, warnings, and possible improvement areas and replaces manual pre-approval.
  • A validation token from the AI review is cryptographically linked to the publish request, ensuring that only validated versions can be submitted.
  • Attached Skills are retained on the published agent version. They are preserved for runtime use, but retained Skills are not published as separately discoverable catalog entries through this flow.
  • Unpublish your agent at any time to remove it from Agent HUB immediately.
  • Receive in-app notifications when your agent is published, rejected, or unpublished by a moderator.
Publishing is available only for classic agents (not pipelines). You must have the models.applications.publish.post permission in your role to see the Publish action.

​
How to Publish an Agent

​
Step 1 — Start

  1. Go to the Agents menu and open the agent you want to publish.
  2. Select the required version from the version selector.
  3. Click the three-dot menu (⋮) to open the actions menu.
  4. Select Publish from the dropdown. Publish wizard opened via the three-dot menu
  • Not visible at all — The option is hidden when your role does not include the publish permission (models.applications.publish.post).
  • Visible but greyed out — The option appears but is disabled with the tooltip “Publishing is blocked by platform policy”. This means is_publish_blocked is turned on for the platform and your project is not on the allowed list. Contact your platform administrator.

​
Step 2 — Preparation

The Publish version dialog opens on the first step, Preparation. This table describes the fields and actions on this step: Preparation step of the Publish wizard
The selected Category helps users discover your agent in Agent HUB. If no category is selected, the wizard does not allow you to continue.
Publishing terms contents:
Exclusions NoticeWhen publishing, certain components will be stripped from the public version for security and privacy:
  • Credentials and API keys
  • Private datasource connections
  • Environment-specific configurations
  • Module endpoints
Consumers of your published agent must provide their own credentials and configurations. Exception: attached Skills and sub-agents are not stripped — their instructions are embedded in the published agent. Retained Skills are never listed as separate entries in the catalog.Best Practices RequirementsTo ensure a quality experience for consumers, your agent should include:
  • A clear, descriptive name (not generic)
  • A detailed description explaining capabilities
  • At least one tag for discoverability
  • Comprehensive instructions for the model
  • Welcome message and conversation starters
Agents that do not meet these requirements may fail validation.Administrative RightsBy publishing, you acknowledge that:
  • Platform administrators may unpublish your agent at any time
  • You retain the ability to unpublish your own agent
  • Published agents are visible to all users in ELITEA Catalog
  • Usage metrics may be collected for published agents

​
Step 3 — Review Validation Results

When you click Continue, the wizard moves to the validation step and submits the selected agent version to the AI-powered validation engine. AI reviews this version against all publication requirements — no manual moderator pre-approval is needed at this stage. The validation snapshot includes the agent content, nested sub-agent content, and any attached Skills retained by the agent. If any of that content changes after validation, the validation token becomes invalid and you must validate again. While the validation runs, the dialog shows:
Reviewing your agent version to ensure it meets publication rules.
The validation engine uses an AI model to perform intelligent checks. If the AI service encounters a transient error, the system automatically retries the validation request twice before reporting a failure. You do not need to take any action during these retries.
In tabs, you can see possible validation outcomes.
The result panel shows three severity counters: Critical, Warnings and Suggestions. Each counter links to the matching section in the detail list. The detail list includes the field, optional context, the issue or suggestion, and a recommended fix when available.After validation:
  • If the status is PASS or WARN, the Publish button becomes turned on.
  • The response includes a validation token. Validation passed

​
Step 4 — Publishing

Click Publish to submit the agent for publication. The system sends both the version name and the validation token from the AI review in the same request — the token certifies that the version passed automated validation. The dialog transitions to the Publishing step and displays:
Publishing your agent…
On success, the dialog closes, a toast notification confirms:
The agent has been published
Once published, your agent appears in ELITEA Catalog, gets the Published badge on its card, and its version is marked as published.
If your agent uses nested sub-agents and some of those sub-agents could not be published, you will see this warning instead:
Agent published, but some sub-agents may not have been published.
Verify that all sub-agents meet publication requirements and publish them separately, if needed.

​
How to Publish an Agent from the Public Project (Admin Mode)

When you are working inside the public project (the admin context), the wizard uses a simplified single-step flow that bypasses the three-step wizard and AI validation entirely. The simplified dialog shows:
  • A heading, Publish version
  • A single prompt, Enter a version name to publish.
  • A Version name text field (same character rules apply: letters, numbers, ., -, _; max 50 characters)
  • A Category dropdown (required)
  • Publish button (turned on only when a valid version name is entered and a category is selected)
  • Cancel button
No publishing terms, no agreement checkbox, and no AI validation step are shown. Clicking Publish submits the version directly.

​
How to Unpublish an Agent

You can remove a published agent from Agent HUB at any time.

​
As the Agent Owner

  1. Open the published agent.
  2. Click the three-dot menu (⋮).
  3. Select Unpublish from the dropdown.
  4. In the Unpublish Agent dialog, review the confirmation message:
    Are you sure you want to unpublish {agent name} (version: {version name})? The agent will be removed from Agent HUB immediately. Existing conversations using this agent version may be affected.
  5. Click Unpublish to confirm, or Cancel to close without changes.
On success, a notification confirms:
Agent has been successfully unpublished!
Unpublish Agent dialog

​
As an Administrator (Moderation Space)

Administrators working in the moderation space see an additional Reason field in the Unpublish Agent dialog:
Provide clear explanation for the unpublishing decision.
This field is optional. Click Unpublish to confirm. The agent owner receives a notification:
“Unpublished agent version id: {linked version ID} from project id: {project ID}. Reason: {reason provided}” — links directly to the affected version.

​
Moderation Workflow

In the standard publishing flow, agents are published directly to published status after passing AI validation — no moderator pre-approval is required.

​
Admin Oversight

Administrators working in the public project or moderation space can unpublish any published agent at any time using the same Unpublish action described in As an Administrator (Moderation Space). This is a reactive oversight tool, not a publication gate. The agent author receives a notification when this happens.

​
Optional Pre-publication Moderation

A pre-publication moderation step can be turned on by platform administrators. When active, agents submitted for publishing enter on_moderation status and are queued for review in Moderation Space → Agents instead of publishing directly. The following table describes moderator actions when moderation is turned on:

​
Notifications

The following table lists notification events and their messages:

​
Best Practices

Address common validation failures before you begin:
  • Set a clear, specific agent name. Avoid generic names like “My Agent” or “Test.”
  • Write a detailed description that explains what the agent does and when to use it.
  • Add at least one tag to help users discover your agent in Agent HUB.
  • Write comprehensive model instructions that cover the agent scope, tone, and behavior.
  • Add a welcome message and at least one conversation starter so users know how to begin.
Agents lacking these elements are likely to receive validation warnings or failures.
The version name you enter in the Preparation step becomes the permanent identifier for this published snapshot. Use a name that communicates the release context, for example:
  • A semantic version number: 1.0.0, 2.1.3
  • A date-based label: 2026-04
  • A descriptor: initial-release, beta
Only letters, numbers, dots (.), hyphens (-), and underscores (_) are allowed. The name cannot be changed after publishing.
A WARN status allows you to proceed, but improvement suggestions indicate areas where your agent could provide a better experience for community users. Review each suggestion in the validation detail list and decide whether to address it before clicking Publish.
When your agent is published, all credentials, private datasource connections, environment-specific configurations, and internal module endpoints are stripped from the public version. Before publishing:
  • Ensure your agent instructions are written to guide users to provide their own credentials.
  • Document which credentials or configurations users must supply in the agent description.
  • Test the agent without credentials to verify that prompts and instructions still guide users effectively.
If your agent uses nested sub-agents, all sub-agents must meet publishing requirements or the parent agent may be published with a warning. Publish all required sub-agents to Agent HUB before publishing the parent agent to ensure the complete workflow is available to users.

​
Troubleshooting

Cause: Your role does not include the publish permission.Solution: Contact your project administrator to verify your role permissions include models.applications.publish.post.
Cause: The AI validation engine encountered repeated errors (ai_validation_failed). The system automatically retried twice but was still unable to complete the review.Solution: Close the wizard and try publishing again after a short wait. If the error persists, contact your platform administrator — the AI validation service may be temporarily unavailable.
Cause: The tooltip reads “Publishing is blocked by platform policy”, which means the platform is_publish_blocked setting is turned on and your project is not included in the allowed list.Solution: Contact your platform administrator to request that your project be added to the publishing allowlist.
Cause: One or more of these conditions is not met:
  • The Version name field is empty.
  • The version name contains characters other than letters, numbers, ., -, or _.
  • A version with the same name already exists for this agent.
  • No Category is selected.
  • The I agree with the Publishing Terms checkbox is unchecked.
Solution: Enter a valid, unique version name, select a category, and check the agreement checkbox.
Cause: The agent version does not meet one or more mandatory publication requirements. Critical issues shown in the validation result list must be resolved before publishing.Solution:
  1. Review the Critical issues in the validation detail list — each entry shows the affected field, context, the issue description, and a recommended fix.
  2. Close the wizard, update the agent accordingly, save changes, then re-open the Publish wizard.
  3. Re-run validation to confirm all critical issues are resolved.
Cause: The server returned a version_name_invalid error, meaning the version name does not pass server-side validation (for example, the name is already in use on the backend).Solution: Change the version name to a different value and click Continue again.
Cause: The validation token generated during Step 3 has a 5-minute expiry. If you wait too long between validation and clicking Publish, the token expires. Alternatively, if you navigated away and edited the agent between validation and publishing, the system detects a content change and invalidates the token.Solution: Click Back to return to the Preparation step, then click Continue again to trigger a fresh validation run. The token is refreshed automatically.
Cause: This is expected behavior. Agent publishing retains attached Skills for the published agent runtime, but does not publish them as separately discoverable catalog entries.Solution: Publish the Skill separately through the Skill publishing flow if you want it to appear as an independent catalog item.
Cause: One or more nested sub-agents could not be published alongside the parent agent. The warning message reads: “Agent published, but some sub-agents may not have been published.”Solution: Open each sub-agent used within your agent configuration and publish them individually following the same three-step wizard. Once all sub-agents are published, users of the parent agent will have access to the complete workflow.
Cause: A platform administrator unpublished your agent. The notification reads: “Unpublished agent version id: [linked ID] from project id: [project ID]. Reason: [reason]” and links directly to the affected version.Solution: Review the agent version to understand which content was flagged. Address the issue, create a new version with the corrections, and submit it for publishing again.
Cause: You have reached the platform limit for published versions on this agent. By default, a maximum of 3 published versions per agent are allowed at one time. The error returned is limit_reached.Solution: Unpublish one or more earlier published versions of the same agent before publishing a new one.
Cause: Your agent (or one of its sub-agents) is using an LLM model that is not a shared model from the public project. Only shared models are allowed in published agents so that all users can run the agent. The error returned is llm_not_shared.Solution: Open the agent (or the affected sub-agent) in the editor, go to the Model settings, and select a model that is shared from the public project. After saving, re-run the publish wizard.

​
Guides & References

  • ELITEA Catalog — Browse and use published agents and skills.
  • Agents — Manage your agents and track version statuses.