The Agent Skills Directory

[DATA_EXFILTRATION]: The CFile::MakeFileArray() method is identified as a critical vector for Arbitrary File Read and Server-Side Request Forgery (SSRF). The skill warns that this API accepts local filesystem paths and remote URLs; passing unvalidated user input to it allows an attacker to read sensitive files or perform internal network probes.
[COMMAND_EXECUTION]: The documentation describes the use of CFile::Delete(), which executes irreversible physical file deletion on the server. The skill notes this should be handled carefully to avoid orphaned files or accidental data loss.
[DYNAMIC_EXECUTION]: The skill highlights a stored Cross-Site Scripting (XSS) risk where files like .html, .svg, and .js saved in the /upload/ directory are served directly by the web server in the application's origin. It mandates extension whitelisting to prevent malicious script execution.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection via user-supplied files and remote content.
Ingestion points: Untrusted data enters via multipart form uploads (rules/uploads.md) and the ui.uploader component (rules/ui-uploader.md).
Boundary markers: The skill recommends using explicit whitelists for file extensions and validating MIME types against file content rather than client headers.
Capability inventory: The skill utilizes CFile::SaveFile for writing to storage, CFile::Delete for removing files, and HttpClient for fetching remote resources.
Sanitization: The instructions include requirements for HTML escaping (htmlspecialcharsbx) and content-based type detection (MimeType::getByFilePath).
[CREDENTIALS_UNSAFE]: The documentation specifically identifies /bitrix/.settings.php as a sensitive file containing configuration data that must be protected from unauthorized access through the file APIs. This highlights the risk of credential exposure if file path validation is bypassed.

bitrix-files