The Agent Skills Directory

Indirect Prompt Injection Consideration: The skill is designed to ingest data from user requests, retrieval tools, and conversation history to generate file content (SKILL.md Section 0). While it lacks explicit boundary delimiters for the generated content, it incorporates several security controls, including strict filename sanitization and path validation (Section 1) to prevent path traversal. The skill utilizes file-writing capabilities to the /app/created/ sandbox and invokes the SharePoint Create file tool (Section 3). The inherent risk of the agent following instructions embedded in the sourced data is managed through these structural constraints.
Data Security Best Practices: The workflow is designed to avoid streaming raw file bytes through the LLM context (Section 3). By using internal path references and allowing the platform's connector to resolve file content server-side, the skill reduces the risk of accidental sensitive data exposure within the model's response stream.
Constrained File Operations: Access to the filesystem is restricted to the /app/created/ directory (Section 1). Mandatory file size verification is performed using standard filesystem APIs (Section 2) prior to any upload, ensuring that only files within a specific size threshold (under 4,000,000 bytes) are processed by the SharePoint connector.

sharepoint-file-creation