This repository publishes the cloudpress-api Agent Skill — documentation of the CloudPress REST API, plus a small installer that writes it into a project or home directory. It is not the CloudPress platform, and it does not run anything on CloudPress infrastructure.

That distinction decides where a report should go, so the two cases are separated below. If you are unsure which applies, report it privately anyway and we will route it.

If you have found a vulnerability in CloudPress itself — the API, the control panel, authentication, the MCP server, or anything that could affect another customer's account, sites, or data — email security@cloudpress.com with the subject CloudPress security report.

Do not open a public issue in this repository. Do not include credentials, API keys, access tokens, customer data, or logs containing any of those.

Report privately even if you found the issue by reading this documentation. A documented endpoint that behaves differently from its documentation is worth telling us about, because one of the two is wrong and we do not yet know which.

Testing limits. You may test against your own CloudPress account and your own sites. Do not test against accounts, sites, or domains that are not yours, and do not run load, denial-of-service, or brute-force testing against CloudPress infrastructure. If you need latitude beyond your own account to demonstrate an issue, ask us first at the address above.

Use the repository's private vulnerability reporting feature from the Security tab, or email security@cloudpress.com with the subject cloud-press/skills security report.

In scope:

• The installer (bin/install.mjs) — anything that writes outside the paths it reported, follows a symlink it should refuse, or clobbers a file without saying so.
• The published artifacts — the @cloud-press/skills npm package, the plugin marketplace content, or a GitHub Release asset whose contents do not match this repository at the corresponding tag.
• Security-relevant errors in the documentation. This is a real category, not a courtesy one. The skill is read by AI coding agents that will follow it literally and at scale, so documentation that understates an authorization requirement, describes a weaker authentication pattern than the API actually supports, or shows an example that mishandles a token or key is a defect worth this channel rather than a normal issue.
• Anything published here that should not have been — an internal hostname, an internal repository path, a credential, or a non-public implementation detail.

Ordinary documentation mistakes with no security consequence — a wrong parameter name, a stale response field, a broken link — are welcome as normal issues.

Only the most recent published version is supported. Fixes ship as a new release; there are no backports to earlier versions. See CHANGELOG.md for what changed between them.

• Which of the two cases above your report falls under.
• The affected version — the skill release (see CHANGELOG.md) or the commit, and for platform reports, the endpoint and request involved.
• Impact: what an attacker gains, and who is affected.
• Reproduction steps, with the smallest sanitized evidence needed to investigate.

You should receive an acknowledgment within five business days. We will coordinate validation, remediation, disclosure timing, and credit with you. Please allow a reasonable remediation period before disclosing publicly.