The Agent Skills Directory

[DATA_EXFILTRATION]: The skill is configured to perform extensive telemetry. Specifically, it instructs the agent to capture the user's prompt verbatim, base64-encode it, and transmit it to a Shopify-owned endpoint (shopify.dev/mcp/usage) as part of the code validation process. While this involves sending user data externally, it is a disclosed feature of an official vendor tool intended for service improvement and analytics.
[INDIRECT_PROMPT_INJECTION]: The skill utilizes a documentation search tool (scripts/search_docs.mjs) to retrieve context for code generation. This introduces a potential surface for indirect prompt injection if the retrieved content were to contain malicious instructions. This is a low-risk architectural characteristic common in retrieval-augmented generation systems.
[COMMAND_EXECUTION]: The skill requires the use of bundled local scripts via a bash tool for its core operations (searching docs, validating code, and logging feedback). These scripts are internal to the skill and facilitate its intended functionality without invoking unauthorized external resources or exhibiting suspicious shell behavior.
[OBFUSCATION]: Several asset files (.json.br) were flagged by static analysis for potential steganography. These are Brotli-compressed binary blobs containing Shopify Function schema data and documentation indices. Given the compression format and the skill's context, these are determined to be false positives rather than intentional obfuscation.

shopify-functions