The Agent Skills Directory

[OBFUSCATION]: The skill includes bundled and minified JavaScript files (scripts/log_feedback.mjs, scripts/search_docs.mjs, and scripts/validate.mjs). This form of code obfuscation prevents clear inspection of the logic being executed by the agent and makes security auditing more difficult.
[DATA_EXFILTRATION]: Telemetry scripts (scripts/track-telemetry.sh and track-telemetry.ps1) capture metadata and the verbatim user prompt, sending this information to https://shopify.dev/mcp/usage. Although the destination is a trusted vendor domain, the transmission of user-provided content represents an external data flow that requires monitoring.
[COMMAND_EXECUTION]: The skill configuration in SKILL.md utilizes a PostToolUse hook to automatically execute shell commands (scripts/track-telemetry.sh) within the agent's operating environment following tool usage.
[INDIRECT_PROMPT_INJECTION]: The skill handles untrusted user input within its core tools, though it applies mitigation techniques to reduce risks.
Ingestion points: The verbatim user prompt is ingested from the agent's context and passed as a command-line argument to the validate.mjs and search_docs.mjs scripts.
Boundary markers: The skill instructions mandate that the agent base64-encode the user prompt before passing it to the validation script, providing a structural boundary to prevent command injection.
Capability inventory: The skill utilizes the bash tool and included scripts that perform network requests and local file system operations.
Sanitization: Base64 encoding of the user prompt is used as a sanitization layer to ensure the input is handled as a safe string when passed to the underlying tool scripts.

shopify-polaris-customer-account-extensions