Publish to Marketplace

Share your custom assistants with the AI/Run CodeMie community by publishing them to the Marketplace. Follow the publishing process to ensure your assistant is properly configured and secure.

Marketplace review

The review flow described on this page applies only when a platform administrator has turned on the approval gate (MARKETPLACE_APPROVAL_GATE_ENABLED) together with Marketplace management (MARKETPLACE_MANAGEMENT_ENABLED). Both are backend environment variables that are off by default, require a restart of the API service, and cannot be switched in the UI. The approval gate has no effect unless Marketplace management is also enabled. With the approval gate on, every assistant published to the Marketplace passes automated quality checks and a manual review by a platform administrator before it becomes visible to other users. Administrators and maintainers who publish their own assistants skip the review queue, and their assistants are approved immediately. On a deployment where the gate is off, publishing goes straight to the Marketplace, and the review statuses, re-validation on edit, and Force-Publish described on this page are not available. See Enabling Marketplace Management.

Publishing Process

1. Create and configure your assistant in the Project Assistants section.

2. On the assistant card, open the actions menu (⋮) and select Publish to Marketplace:

   

3. Review credentials, prompt variables, and categories as described in Credential Policies and Select Categories.

4. The system validates the assistant configuration against the Marketplace quality checks:

   

5. If the assistant passes the checks, it is sent for manual review. The Sent for review dialog confirms that the assistant is pending approval and will not be visible in the Marketplace until an administrator approves it:

   

   If any check fails, a dialog lists the issues to fix. See Validation Checks.

6. Track the review result by the status badge on the assistant card. See Review Statuses.

Credential Policies

Personal/Project Credential Detection

Before publishing, the system checks for sensitive credentials:

• If your assistant uses personal or project-specific credentials, a warning modal will appear
• The modal guides you through the process and prevents accidental credential exposure
• You must review and confirm before proceeding with publication

Credential Security

Always review credential warnings carefully. Publishing assistants with personal credentials may expose sensitive information to other users.

Prompt Variables Review

When publishing an assistant with prompt variables, the system displays a Prompt Variables Review section showing how variables will be handled:

Variable Display:

• All Variables: Lists all prompt variables used in the assistant's system instructions
• Sensitive Variables: Displays sensitive variables separately with count (e.g., "Sensitive Variables (3):")
  • Sensitive variables are marked with orange tags
  • These variables are displayed as button-style tags with capitalized names

Security Notice:

The system displays an important warning:

info

Sensitive prompt variables will not be visible to other users, but will be injected into the system prompt when the assistant is invoked.

This ensures that:

• Users who clone your assistant from the marketplace will see placeholders for sensitive variables
• Sensitive variable values remain encrypted and hidden
• Users will need to provide their own values for sensitive variables when using the assistant
• The assistant functionality is preserved while protecting confidential information

Select Categories

Choose up to 3 relevant categories to help users discover your assistant:

Category	Description
Migration & Modernization	Cloud migrations, application modernization, legacy system updates
Support	Chatbots, virtual help desks, FAQ systems, customer support automation
Business Analysis	Analyzing business needs and gathering requirements
Product Management	Product roadmaps, lifecycle management, prioritization, stakeholder collaboration
Data Analytics	Data analysis, visualization, actionable insights from raw data
Engineering	Developer support, full-stack engineering, software engineering workflows
Quality Assurance	Testing, software reliability, automating QA processes
DevOps	Cloud systems, networking, infrastructure, DevOps workflows
Security	System security, vulnerability assessment, security automation
Compliance	Regulatory compliance, automating auditing processes
Project Management	Planning, tracking, project and task management
Customer Experience	Customer satisfaction, ticket routing, surveys, interaction automation
Monitoring & Alerts	System health monitoring, alert generation, proactive issue detection
UI/UX Design	Prototyping, wireframing, user experience research analysis
Knowledge Management	Organizational knowledge, documentation, information retrieval
Training	Onboarding, employee training, internal knowledge sharing
Architecture	Scalable and reliable technical architecture design
Presales	Client interactions, proposal generation, solution demonstrations

Category Management

Users with the isAdmin permissions can create, edit, and delete categories. Learn more in Assistant Categories Management.

Integration Requirements

• When publishing, your user integrations will be made available alongside the assistant
• If no suitable integrations are selected, you can set default integration fields
• The warning modal will not appear if no user credentials are selected

Integration Sharing

Be mindful that publishing shares your integration configurations. Review integration settings before publishing.

Validation Checks

Before an assistant is submitted for review, the system runs two validation steps. Each finding names the affected field (for example, Name or Description) and includes a recommendation on how to fix it.

Step 1: Deterministic Checks

The checks listed in the table below are applied first. If any of them fails, the Assistant Verification Failed dialog lists the findings:

• Edit assistant: open the assistant editor to fix the reported issues.
• Cancel: close the dialog without publishing.
• Force-Publish: override the CRITICAL findings with a justification. Shown only to platform administrators and maintainers. See Force-Publish for Administrators.

Each finding has one of two severity levels:

Severity	Effect
CRITICAL	Blocks submission. The issue must be fixed before the assistant can be sent for review.
OPTIONAL	Recommendation only. Fixing it improves the assistant's quality but does not block submission.

OPTIONAL findings open no dialog: the publish continues.

The following checks are applied to every assistant:

Check	Severity	Rule (default values)
Name length	CRITICAL	The name must be 5–60 characters long.
Name not generic	CRITICAL	The name must not be a generic name such as "assistant", "bot", "test", "demo", or "untitled".
Description length	CRITICAL	The description must be at least 100 characters long.
Description differs from name	OPTIONAL	The description must not repeat the name.
System prompt length	CRITICAL	The system prompt must be at least 200 characters long.
Category	CRITICAL	At least one category must be assigned.
Substance	CRITICAL	The assistant must have at least one tool, MCP server, skill, or data source, or a system prompt of at least 500 characters.
No inline credentials	CRITICAL	The system prompt and context must not contain credentials such as AWS keys, GitHub tokens, or Bearer tokens.

Step 2: AI Quality Check

After the deterministic checks pass, an AI-based quality analysis reviews the configuration and suggests improvements, for example a clearer name or description. If it has recommendations, the Assistant Quality Validation Failed dialog lists them in the Analysis Summary section:

• Manual Edit: open the assistant editor to apply the suggested changes.
• Cancel: close the dialog without publishing.
• Publish Anyway: skip the AI recommendations and continue publishing. When the analysis contains only optional recommendations, or when Marketplace management is not enabled on the deployment, the button is available to every user. When Marketplace management is enabled and the analysis contains CRITICAL items, only administrators and maintainers see it, and a reason for the override must be entered first.

Publish Anyway skips only the AI check. It does not override failed deterministic checks.

note

Platform administrators can change the default thresholds on the Config tab of Marketplace Management, so the exact limits depend on the deployment.

Review Statuses

After an assistant is submitted, a status badge next to the author name on the assistant card shows where it is in the review process. Assistants that were never submitted to the Marketplace show no badge.

Status	Meaning
REVIEW	The assistant passed the automated checks and is waiting for an administrator's decision. It is not visible in the Marketplace yet.
VERIFIED	An administrator approved the assistant. It is published and visible to all users in the Marketplace.
REJECTED	An administrator rejected the assistant. The assistant is not published.

Pending review:

Verified: the ✓ VERIFIED badge is also displayed on the assistant card in the Marketplace and in the assistant details header, so users can distinguish reviewed assistants.

Rejected: hover over the REJECTED badge to see the reason provided by the administrator.

To resubmit a rejected assistant, update its configuration according to the rejection reason and publish it to the Marketplace again.

Withdrawing an Assistant from Review

While an assistant is in REVIEW, the actions menu shows Remove from Marketplace review queue instead of Publish to Marketplace. Selecting it opens the Remove from Review Queue? dialog. After confirmation, the assistant leaves the review queue and stays in the author's Project Assistants.

Editing a Published or In-Review Assistant

An edit to an assistant that is published (VERIFIED) or waiting for review (REVIEW) is validated again. A VERIFIED assistant whose configuration changed goes back to review.

1. Edit the assistant and click Save.

2. The Save changes dialog warns that saving will unpublish the assistant and send it for re-validation. The assistant stays unavailable in the Marketplace until the review is complete:

   

   • Click Cancel to discard the save. The assistant keeps its current status.
   • Click Save to save the changes and run the validation checks.

3. If the updated assistant passes the checks, the changes are saved and a confirmation message appears. A VERIFIED assistant whose configuration changed goes back to REVIEW and leaves the Marketplace until it is approved again. An assistant that is already in REVIEW stays in the queue. A save without configuration changes keeps the VERIFIED status.

4. If the updated assistant fails the checks, the Assistant Verification Failed dialog shows the findings, and a banner describes the current Marketplace state of the assistant:

   

   Choose one of the options:

   • Keep editing: return to the editor and fix the reported issues.
   • Save and remove (assistant with the REVIEW status): save the changes and remove the assistant from the review queue.
   • Save & Unpublish (assistant with the VERIFIED status): save the changes and unpublish the assistant from the Marketplace.

   After either save option, the assistant can be edited further outside the Marketplace and published again later.

The same flow applies when an assistant is edited from a chat. Clicking Configuration in the chat header opens the Configure & Test panel; clicking Save in the panel for an assistant with the REVIEW or VERIFIED status shows the same Save changes warning and runs the same validation checks:

tip

Group several changes into a single edit. Each save of a published assistant that changes its configuration removes it from the Marketplace until an administrator approves it again.

Force-Publish for Administrators

Access Required

This functionality is only available to platform administrators and maintainers. Users with the isAdmin role can publish assistants without waiting in the review queue and override CRITICAL validation findings. Regular users do not see these options.

Platform administrators and maintainers follow a shortened publishing flow:

• If their assistant passes the validation checks, it is published directly with the VERIFIED status, without waiting in the review queue.
• If CRITICAL findings block the assistant, it can be force-published with a justification:
  1. When CRITICAL checks fail on publish or on save, the Assistant Verification Failed dialog opens. Click Force-Publish.
  2. In the Why is it safe to publish anyway? field, describe why the checks should be overridden. The field appears only after Force-Publish is clicked, and the button then changes to Confirm Force-Publish. The justification must be at least 20 characters long; the button stays disabled until then.
  3. Click Confirm Force-Publish. The assistant is published with the VERIFIED status.

The justification, the administrator who overrode the checks, and the bypassed CRITICAL findings are stored with the publishing decision for audit purposes. The force-publish options are not available to regular users.

Pending assistants whose gate findings became CRITICAL after submission can also be force-published from the review queue. See Force-Publishing from the Review Queue.

For information about reviewing submitted assistants, see Marketplace Management.

After Publishing

Once approved (or published directly when the approval gate is off), your assistant appears in the Marketplace tab:

You maintain full control with these options:

• View Details: See how users interact with your assistant
• Edit: Update configuration and settings (triggers re-validation and review, see Editing a Published or In-Review Assistant)
• Delete: Permanently remove the assistant
• Remove from Marketplace: Return the assistant to your Project Assistants tab

Maintenance

Regularly update your published assistants to fix issues, improve performance, and incorporate user feedback.