Content Policy

Every published version runs through automated checks before it goes live. This policy describes what's allowed.

Not allowed

• Malware or destructive payloads. Skills that exfiltrate data, install backdoors, or damage systems.
• Leaked secrets. Archives are scanned for credentials, private keys, and tokens; matches fail the build.
• Path-unsafe archives. Absolute paths or .. traversal in archive entries are rejected.
• Impersonation & malicious naming. A version's manifest name must match the skill it's published under.

Enforcement

Publishes that fail checks never become an installable version. Platform admins may block skills, suspend organizations, and suspend accounts that violate this policy. Every moderation action is recorded in the audit log.