The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for handling untrusted file uploads. It includes a complete evidence chain for security mitigation: ingestion points are defined by Active Storage attachments; boundary markers are established via model-level validations for byte size and content-type; the capability inventory is limited to standard Rails storage operations; and sanitization is addressed by recommending magic-byte sniffing with the Marcel library rather than trusting user-provided headers.\n- [EXTERNAL_DOWNLOADS]: The skill references established Ruby gems like image_processing and marcel, along with official documentation for AWS S3 and Google Cloud Storage. These references are documented neutrally as they represent industry-standard components for the stated functionality.\n- [CREDENTIALS_UNSAFE]: Configuration examples correctly demonstrate using the Rails credentials system for managing cloud provider keys, avoiding the risk of hardcoded secrets in version-controlled files.

activestorage-uploads