Review and approval
Every new listing is security scanned, checked by AI and approved by a person before it goes live. Most are approved within a few hours. Allow up to 48.
1. Security scan. Eight checks run on every upload.
| Check | What fails it |
|---|---|
| File structure | No SKILL.md, missing frontmatter, or no name in the main SKILL.md |
| File types | Executables such as .exe, .dll, .jar or .wasm |
| Dangerous commands | Deleting from root paths, piping a download into a shell, os.system, shell=True, chmod 777 |
| Secrets | API keys, tokens or passwords in the files |
| Environment harvesting | Reading environment variables and sending them over the network |
| Network access | Network calls combined with reading local data |
| Obfuscation | Hidden or encoded code |
| Prompt injection | Phrases such as "ignore previous instructions", "do not tell the user" or "secretly" |
Some patterns only warn, such as curl, fetch, file writes, sudo or base64. Each check passes, warns or fails. The results go to the AI review and then to a person, who decides. A failed check doesn't block approval when the finding turns out to be a false positive.
2. AI review. Reads the scan results, permissions, file tree and every SKILL.md, and recommends approve, reject or a closer look. Listings with outside checkout links or unlock codes always go to a person.
3. A person approves. Every new listing. An update with no failed checks, an AI approval and no red flags can be approved automatically.
| Status | Meaning |
|---|---|
| Pending | In review |
| Live | On the marketplace and available to agents |
| Rejected | Not approved. The reason is in your dashboard. |
| Paused | Hidden by Agensi, or on hold until your payouts work again |
If it's rejected. You get a message in your dashboard and by email with the checks that failed, what was found and how to fix it. Fix it, then click Resubmit. Uploading the exact same ZIP again is blocked.