Attack Chain

1. An attacker crafts a malicious .praison recipe bundle.
2. The attacker modifies the manifest.json file within the bundle, injecting ../ sequences into the name field, for example setting "name": "../../outside-dir".
3. The attacker sends a POST request to the /v1/recipes/{name}/{version} endpoint, uploading the malicious bundle, seemingly targeting a safe location such as /v1/recipes/safe/1.0.0.
4. The RegistryServer._handle_publish() function parses the request and writes the uploaded .praison file to a temporary path.
5. The LocalRegistry.publish() function opens the tarball, reads the attacker-controlled manifest.json, and constructs a path based on the malicious name and version fields: self.recipes_path / name / version.
6. The server creates directories based on the manipulated path using recipe_dir.mkdir(parents=True, exist_ok=True), effectively traversing out of the intended registry root.
7. The server copies the uploaded bundle to the out-of-root destination.
8. The server then validates the manifest values against the URL parameters. Because they don't match, the server attempts to delete the file but it still persists due to the path traversal. The server returns an HTTP 400 error, but the malicious file has already been written outside the registry root.

Impact

This vulnerability allows for arbitrary file write operations outside the intended registry root, creating a high integrity risk. Successful exploitation can lead to the creation or overwriting of critical system files, potentially leading to denial-of-service or privilege escalation. While the provided PoC demonstrates a single file write, the ability to write to arbitrary locations significantly broadens the attack surface. Impacted parties include registry operators running the PraisonAI recipe registry service and any deployment that allows remote recipe publication. If adjacent writable locations contain sensitive application data or service files, the impact can be substantial.

Recommendation

• Deploy the Sigma rule PraisonAI Recipe Registry Path Traversal Attempt to detect attempts to exploit this vulnerability based on the creation of files with path traversal sequences in their names (file_event logs).
• Upgrade to a patched version of PraisonAI that includes validation of manifest.json name and version fields prior to any filesystem write.
• Monitor web server logs for HTTP 400 errors associated with recipe publish requests and investigate the associated uploaded files for path traversal attempts (webserver logs).
• Implement input validation for recipe names and versions on the client-side to prevent users from creating recipes with malicious names.