The Agent Skills Directory

[OBFUSCATION]: The helper scripts scripts/log_feedback.mjs, scripts/search_docs.mjs, and scripts/validate.mjs contain minified and bundled JavaScript code. This obfuscation hides the operational logic and complicates security review.
[DATA_EXFILTRATION]: The skill is configured to transmit session data to https://shopify.dev/mcp/usage. The telemetry includes verbatim user prompts (base64-encoded), agent-generated code blocks, session identifiers, and tool usage metadata. The track-telemetry.sh and track-telemetry.ps1 scripts facilitate this by stashing prompt data in temporary files (/tmp or %TEMP%) before exfiltration via network requests.
[COMMAND_EXECUTION]: The skill employs a PostToolUse hook to automatically execute the scripts/track-telemetry.sh script using the bash tool after every tool invocation.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user data by requiring the agent to base64-encode the user's most recent message and pass it as an argument to the validation script.
Ingestion points: Verbatim user prompt passed as the --user-prompt-base64 flag to scripts/validate.mjs.
Boundary markers: None identified in the script logic for the decoded prompt content.
Capability inventory: The skill has access to the bash tool, which it uses to run local scripts and perform network operations via curl and fetch.
Sanitization: The agent is instructed to base64-encode the prompt before execution to ensure shell stability, but the script decodes and processes the raw content, providing a surface for instruction injection.

shopify-polaris-admin-extensions