The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references several external dependencies including aws4fetch, @aws-sdk/client-s3, and the wrangler CLI.
These are well-known and trusted resources within the Cloudflare and AWS ecosystems.
While some version numbers (e.g., wrangler@4.59.2) appear to be forward-dated or non-existent, the sources are reputable official registries.
[COMMAND_EXECUTION]: The documentation provides numerous CLI commands for managing R2 storage.
Commands such as npx wrangler r2 bucket create and wrangler r2 object put are standard development operations.
The instructions correctly identify that these commands should be used in a secure environment with appropriate API token permissions.
[PROMPT_INJECTION]: The skill describes an attack surface for indirect prompt injection through its data processing templates.
Ingestion points: Untrusted data enters the application via c.req.formData() and c.req.arrayBuffer() in the provided Hono worker templates (found in common-patterns.md and templates/r2-simple-upload.ts).
Boundary markers: There are no explicit delimiters or system instructions in the templates to prevent an LLM from executing malicious commands found within the retrieved objects.
Capability inventory: The skill provides full administrative capabilities for R2 buckets, including put, get, delete, and list across all scripts.
Sanitization: The templates implement basic file validation (MIME type checks and UUID-based key generation) but do not perform deep content inspection or sanitization of the object bodies.

cloudflare-r2