The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes data from untrusted external sources and SharePoint resources, creating a potential surface for indirect prompt injection attacks.
Ingestion points: The uploadFile and replaceFile tools (scripts/uploadFile.ts, scripts/replaceFile.ts) download content from user-specified URLs. The skill also retrieves content from SharePoint sites, lists, and pages via scripts like findFiles.ts and findListItems.ts.
Boundary markers: Input validation is performed using Zod schemas to ensure data integrity. However, the skill does not include specific instructions or markers to the agent to treat processed content as untrusted data.
Capability inventory: The connector provides capabilities to perform network fetches and write to SharePoint (creating folders, uploading files, updating list items).
Sanitization: The skill correctly sanitizes path segments and identifiers using encodeURIComponent before including them in API requests.
[COMMAND_EXECUTION]: The cli.js entry point uses spawnSync to execute the connector's logic. This is a standard dispatching mechanism for this connector's architecture and is restricted to executing internal scripts.

microsoft-sharepoint