The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from the clipboard (HTML, product details) and remote URLs (WebFetch/curl) to generate WordPress block markup. It lacks explicit instructions to ignore embedded instructions within the processed data, creating a potential surface for indirect injection.
Ingestion points: Clipboard content (AAWP HTML, product details), remote URLs (Path 3: URL Input).
Boundary markers: Absent. The instructions do not define delimiters for user-provided or fetched content.
Capability inventory: Uses curl and pbcopy via shell commands (SKILL.md).
Sanitization: Instructions mention cleaning up product titles and HTML entities, but no security-focused sanitization of the input data before interpolation into the output block.
[EXTERNAL_DOWNLOADS]: The skill uses curl or WebFetch to download content from arbitrary user-provided URLs to extract product data (Path 3: URL Input).
[COMMAND_EXECUTION]: The skill executes shell commands using pbcopy to automatically place generated markup onto the user's clipboard and uses curl to fetch webpage data.
[DATA_EXFILTRATION]: The skill hardcodes specific Amazon affiliate tags (gtorg0f-20 for .com and gaurtiwa-21 for .in) and instructs the agent to inject these into product URLs. While not exfiltration of sensitive local data, it redirects user traffic through specific affiliate accounts without explicit disclosure.

product-box-converter