The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates that ingest and process untrusted metadata from external images, creating a potential surface for indirect prompt injection or schema confusion.
Ingestion points: Untrusted data enters the context through the metadata parameter in upload functions located in templates/upload-api-basic.ts and templates/direct-creator-upload-backend.ts.
Boundary markers: Absent in raw templates; implementations are responsible for isolating metadata from instructions.
Capability inventory: The skill scripts utilize network access via fetch to interact with the Cloudflare Images API.
Sanitization: The skill's frontend implementation correctly uses textContent to prevent DOM XSS, though API templates rely on implementers for data sanitization.
[COMMAND_EXECUTION]: Local shell scripts are provided to automate API testing and signature generation.
Evidence: scripts/analyze-usage.sh and scripts/generate-signed-url.sh execute curl, openssl, and jq commands.
Note: These scripts implement security best practices, such as unsetting sensitive environment variables like CF_API_TOKEN and CF_IMAGES_SIGNING_KEY after use to prevent credential leakage.
[EXTERNAL_DOWNLOADS]: The skill facilitates data transfer between the local environment and well-known Cloudflare service domains.
Evidence: Network operations target api.cloudflare.com and imagedelivery.net for image ingestion, management, and serving.

cloudflare-images