Marketplace Moderation
The Marketplace is the install-only Agent Recipe Catalog — a free catalog of ready-made AI agent recipes. Nothing is bought or sold, so there is no order, refund, or dispute flow to moderate.
There is no dedicated admin moderation route in the current build. A recipe becomes publicly listed the moment its published flag is true, which the recipe's own author sets by publishing one of their agents via POST /api/v1/catalog/agent-recipes/publish. Listing is self-service; there is no review gate in front of it.
What an admin can do today
- Review what is public. Admins see the catalog through the same endpoint users hit:
GET /api/v1/catalog/agent-recipes. Each entry carriesname,description,category,tags,role,persona,system_prompt, and itspublished/install_countstate. There is no separate admin view — audit the public list directly. - Suspend an abusive creator. Deactivate the offending User record at Users by setting
isActive: false. That account can no longer publish new recipes or install existing ones, which is the effective lever for stopping abuse at the source.
What is missing / roadmap
- A dedicated moderation queue UI.
- The ability for an admin to directly unpublish or take down a single recipe (today the only path is deactivating the author).
- Category curation.
Until those land, moderation is account-level: identify the creator behind a problematic recipe and deactivate them. Because installs are free and produce an inactive cloned agent in the installer's own workspace, the blast radius of a bad recipe is bounded — it does nothing until a user reviews it and turns it on.