Marketplace Content Policy

1. What this policy covers

This policy applies to everything published on the Dija Marketplace: the prompt text of an agent, the steps of a playbook, the tools a listing declares, and the name, description and images shown with it.

It is part of the Marketplace Author Terms. Authors accept it when they become authors. Anyone else reading it is being told the rules of the place rather than signing up to them.

"We" and "us" mean Dija Limited, a company incorporated in Cyprus, registration number HE 483340, with its registered office at Inomenon Ethnon, 2 Anastasia Court, Floor 4, 6042 Larnaca, Cyprus.

A listing from the marketplace runs inside somebody else's workspace, with access to their data and their tools. That is the reason this policy exists, and the reason the rules below are firm.

2. Nothing illegal

Do not publish a listing that is illegal, or that is built to help somebody do something illegal, in the places it would realistically be used. That includes fraud, unauthorised access to computer systems, surveillance or stalking of individuals, and helping anyone evade the law.

3. No spam, and no tools for deceiving people

Do not publish a listing that:

• sends to people who never agreed to hear from the sender, including scraped, harvested or bought lists;
• hides or fakes who is sending, who a message is from, or what it is about;
• invents a prior relationship, an introduction or a referral;
• works around an unsubscribe, a send limit or a rate limit;
• impersonates a person, a company, a product or us, or uses a name, handle, description or icon meant to be mistaken for somebody else's.

4. No malware, and no attacks on the systems a listing runs in

Do not publish a listing that:

• contains or installs malicious code, or is designed to damage, disable or take over a system;
• tries to override the instructions, safety rules or configuration of the workspace it runs in, including telling the model to ignore its own instructions or writing text that pretends to come from the system;
• tries to reach tools, data or accounts the installer did not give it;
• sends the installing workspace's data anywhere the installer did not choose, whether by message, by a web request, or by hiding it inside something that leaves the workspace;
• tries to manipulate another system, service or model into acting outside its own rules.

5. Nothing that infringes somebody else's rights

Do not publish work that is not yours to publish. That covers copyright, trade marks, trade secrets, database rights and anything else somebody else holds.

Where you build on another author's listing, credit it. Republishing another author's listing unchanged is not allowed.

Complaints about infringement are handled under the Copyright and Takedown Policy.

6. No personal data, and no credentials

Do not put other people's personal data in a listing: no names, email addresses, phone numbers, postal addresses or other details of real people, in the prompt, in the description, or anywhere else. Use obviously made-up examples.

Do not publish a listing whose purpose is to collect, scrape or process personal data in a way the law does not allow.

Do not put credentials in a listing: no API keys, tokens, passwords, connection strings or private keys, yours or anybody else's. A key inside a listing is compromised the moment somebody installs it.

Do not publish anything confidential to your employer, your client or your customers.

7. No misleading claims

What a listing says it does is what it has to do. Do not:

• claim abilities, integrations or data sources the listing does not have;
• describe a listing as approved, verified, certified or endorsed by us. We review listings and we do not endorse them;
• promise results, reply rates, revenue or performance you cannot back up;
• misdescribe the data a listing touches or the tools it uses. The summary an installer sees is built from what you declared.

8. Nothing that targets children

The marketplace is for business use by adults. Do not publish a listing aimed at children, one built to collect or profile data about children, or one that sexualises, endangers or exploits a child in any way. We refer material of that last kind to the appropriate authorities.

9. Do not try to hide a breach

Trying to hide any of the above from our checks is a breach in itself, whatever was being hidden. That includes look-alike characters, invisible characters, and breaking a phrase up so it reads differently to a machine than to a person. Text is put into a standard form before it is checked. Hiding something turns a finding a reviewer might have read generously into evidence that you meant it.

10. How this policy is applied

Three things happen, and it is worth being exact about what each one can do.

Automated checks. Every submission is checked by software before a person sees it. These checks can refuse a submission and send it back for changes, and they can flag things for the reviewer. They cannot approve anything. A submission that has passed them has been approved by nobody.

A person reads it. Every version is read by a reviewer before it is approved. No listing reaches the marketplace without that, and there is no path on which software publishes a listing on its own.

Reports. Once a listing is live, anyone can tell us about it. See section 11.

We do not publish the detail of what our checks look for. Approval is not an endorsement and not a guarantee that a listing is safe, so read what you install, as you would anywhere else.

11. Reporting a listing

There are two ways, and one of them asks nothing of you.

By email. Write to legal@dija.ai. You do not need a Dija account and you do not need to be a customer. Tell us which listing (the web address of its page is the most useful thing you can give us), what is wrong with it, and how to reach you if we need more.

In the product. If you are signed in to a Dija account, every listing page has a report control. You choose a reason and describe what you found.

We read every report. We may not be able to reply to you individually, and we do not always tell a reporter what we decided. Reports are not a weapon: filing them in bad faith, or to damage a competitor, is itself a breach of this policy and we may act on it.

12. What happens when a listing breaks this policy

Depending on what it is, we may ask the author for changes, refuse a version, withdraw a version or a whole listing, stop the author publishing, or ban the author.

Where a listing is unsafe, infringing or seriously in breach, we can withdraw it in a way that also switches off copies people have already installed and cancels work running against them at that moment. The workspaces affected are told. Nothing in their workspace is deleted. This has limits, and we would rather state them: withdrawals travel by a feed that deployments check for themselves, so they are not instant, and a deployment that never connects to us never receives one.

We can act on a first breach where the risk calls for it, and we do not run a fixed number of strikes.

An author whose listing or account we act against can write to legal@dija.ai and ask us to look again. A person will.

13. Changes, and contact

We may change this policy as the product and the law change. A change takes effect when we publish it here with a new date, and a listing that was acceptable under an older version can be re-examined against the current one.

This policy is part of the Marketplace Author Terms, which are governed by the law of Cyprus, with the non-exclusive jurisdiction of the courts of Cyprus. Nothing here takes away rights the law of your own country gives you and does not allow you to give up.

Anything about this policy, including a report about a listing: legal@dija.ai.