The public catalog, publishers and moderation

How an organization becomes a publisher, why a public skill is listed only after a moderator approves it, how anyone can report one, and what a takedown does to the skill.

By The SKaaS team. Last updated . Markdown version

The public catalog is the library on this site. Anyone can browse it, read a skill's guide and its audit, and report a skill, without an account. An organization appears in it through a publisher profile, and a skill appears in it after a person has reviewed its first public release.

Edition and plan: every edition and every plan can publish. The moderation routes exist only on a deployment that has operator credentials, which is how the hosted library is run. A self-hosted deployment without them serves a catalog with no moderation.

How does an organization become a publisher?

Create a publisher profile. It has a slug that becomes the public address, a display name and an optional home page. The slug never changes, so choose it once. You need the repos:admin scope.

curl -sS -X POST "https://api.skaas.io/v1/orgs/acme/publisher" \
  -H "authorization: Bearer $SKAAS_TOKEN" -H "content-type: application/json" \
  -d '{ "slug": "acme", "display_name": "Acme Tools", "homepage_url": "https://example.com" }'

A profile is pending, verified or suspended. You can change the display name and the home page later with PATCH on the same address. A suspended publisher cannot change its profile. The verified mark is a separate step. See Verified publishers.

When does a skill appear in the catalog?

Set a repository to public. A public skill is listed only after a moderator has approved the first public release of that skill. Making a repository public asks for a review of every skill in it. Publishing a new release of a skill in a public repository asks again if the earlier review was refused. Until a review is approved, the public catalog answers as if the skill did not exist, and it stays out of public search. The members of your organization read it as usual.

The queue holds an item for each reason: a new publisher, a first public publish, an abuse report, an audit finding, a model-assisted review or a manual flag. A moderator claims an item, which holds it for 15 minutes so that two people do not decide the same one, and resolves it as approved, rejected or escalated, with a note. An escalated item stays open for another moderator. A refusal deletes nothing. The skill stays out of the catalog, and you can fix it and publish again.

How do I report a public skill?

Open the skill in the library and use the report form, or send a report to the catalog route without an account. Pick a category: malware, a leaked credential, impersonation, copyright, spam, something illegal, or other. Describe the problem in up to 2,000 characters. You can name the release that the report is about. A deployment may ask for a check that a person files the report. You get a receipt with a report identifier.

The organization that owns the skill sees the reports about it, without the identity of the reporter. A report is open, then triaged, then actioned or dismissed. A dismissed report changes nothing else.

What can an operator do about a report?

An operator signs in with a credential of at least 32 characters, which the deployment lists by name, so that each action records who did it. The operator reads reports and the moderation queue across organizations and resolves them. To take a skill down, the operator records the case reference and a reason. The skill is quarantined and held, so it leaves search and listings, and every read of it is refused. Nothing is deleted. The hold records the case and the operator, and only an operator can lift it. An operator can also place a hold on a skill that is already quarantined, and lift one, through two routes of their own: POST /v1/operator/orgs/{org_id}/artifacts/{artifact_id}/hold and .../hold/lift. See the runbook on a takedown request in docs/runbooks/ of the repository.

What does a hold stop?

While a hold is in place, no role of the owning organization and no key changes the state of the skill, publishes to it, moves its channels or makes its repository public. Each of these is refused with FORBIDDEN, with the message that an operator holds the artifact and the reason operator_hold. The public catalog answers for the skill as for one that does not exist. The same refusal stops the owner from withdrawing a release of the skill and from restoring one, because either would change what the case is about. A repository that holds a skill cannot be deleted at all, so a hold cannot be escaped that way either. The skill stays quarantined whatever else happens.

What does the owning organization see?

The skill shows as quarantined. The activity log of the organization carries an artifact.held event with the case reference, and a skill.state_changed event with the reason when the skill was not quarantined before. The dashboard raises its notice that a skill was quarantined. The organization sees the reports about the skill without the identity of the reporter. Administrators see the hold in the list of quarantined artifacts: the Quarantine page under administration in the dashboard, or GET /v1/orgs/{org}/quarantine with the repos:admin scope. Each artifact shows when it was quarantined, the reason that was recorded, whether an operator holds it and the case reference of the hold or the takedown. A held artifact shows a sentence that only an operator can lift the hold, and it has no control. When the operator lifts the hold, the log carries an artifact.hold_lifted event.

Can I change a quarantine?

It depends on who set it. An admin of the owning organization can quarantine a skill and can set it back to active, which is a security action that needs the repos:admin scope. The Quarantine page of the dashboard does it for an artifact that no operator holds, with a button on the row of the artifact and an optional reason. A skill that an operator took down carries the hold. The owning organization cannot lift that hold, and an attempt to set the skill back to active is refused, so a takedown cannot be undone by the party that it was made against. If you disagree with a takedown, say so to the operator, quoting the case reference that the skill shows. If the operator agrees, the operator lifts the hold. Lifting it does not reactivate the skill: it stays quarantined until an admin of the owning organization sets it back to active. It is public again if its first public release was approved.

Public guide and resource responses, including pinned reads, may stay in a shared cache for up to one minute after a security withdrawal or takedown. The origin refuses new reads immediately. There is no cache-purge operation. Recheck a pinned read after that minute.