sharepoint-api

Warn

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The scripts/session-manager.sh script contains a force_cleanup function that executes pkill -9 Chrome, which terminates all Google Chrome processes on the host system regardless of their association with this skill. It also recursively deletes all SingletonLock files within the ~/agent-browser-profiles directory.
  • [DYNAMIC_EXECUTION]: Operational scripts such as list-files.sh, download-file.sh, and upload-file.sh use agent-browser eval to execute dynamically generated JavaScript code blocks within a browser session to perform authenticated SharePoint actions.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes file metadata from SharePoint and allows user-provided input to influence API queries.
  • Ingestion points: File metadata and folder paths retrieved via RenderListDataAsStream in scripts/list-files.sh and scripts/upload-file.sh.
  • Boundary markers: No explicit delimiters or boundary instructions are used when passing external data to subsequent agent actions.
  • Capability inventory: The skill possesses capabilities for reading/writing files, performing network operations, and capturing network traffic via HAR capture.
  • Sanitization: The build_viewxml function in scripts/list-files.sh directly interpolates user-provided shell variables (e.g., $FILE_TYPE, $MODIFIED_AFTER) into a CAML XML query string without validation or escaping, which can lead to query injection in the SharePoint API.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Aug 15, 2026, 05:15 AM
Security Audit — agent-trust-hub — sharepoint-api