The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates and patterns for ingesting and processing untrusted file data (Pattern 1, Pattern 2, and Pattern 5), which represents a surface for indirect prompt injection if an agent uses these patterns to handle files containing malicious instructions.
Ingestion points: The IFormFile parameters in FileUploadValidator.ValidateAsync and ImageProcessingService.ProcessAndUploadAsync, as well as MultipartReader in StreamingFileUploadModel.OnPostAsync (found in SKILL.md).
Boundary markers: The provided patterns focus on file metadata and signature validation but do not define delimiters or specific instructions for the agent to ignore embedded natural language instructions within the file contents.
Capability inventory: The patterns include capabilities for file system writes (File.Create), file deletion (File.Delete), directory creation (Directory.CreateDirectory), and network-based uploads to cloud storage (BlobContainerClient.UploadAsync).
Sanitization: The skill demonstrates good security practices including path traversal mitigation using Path.GetFileName, file signature (magic number) verification, and enforcement of file size limits.
[EXTERNAL_DOWNLOADS]: The skill references and integrates with external libraries and cloud services within its code patterns.
Evidence: References to the ImageSharp library for image processing and Azure Blob Storage for remote file persistence.
Context: These are well-known technology services and libraries used for their intended purpose in standard development workflows.

file-handling