The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests untrusted HTML content from external storefronts to check SEO signals and policies, which creates a surface for indirect prompt injection. To mitigate this risk, the instructions in SKILL.md require the agent to sandbox all crawled markup within XML delimiters (<storefront-markup-context>) and explicitly instruct the model to ignore any instruction sequences embedded within that data. Mandatory Evidence Chain: (1) Ingestion point: crawlStorefront in scripts/shopify-markets-localization-auditor.mjs. (2) Boundary markers: Present in SKILL.md Hard Rules. (3) Capability inventory: Access to Shopify Admin API and local command execution via Shopify CLI. (4) Sanitization: The textFromHtml utility in the script strips tags, and the skill body provides explicit model-level instructions for isolation.
[COMMAND_EXECUTION]: The bundled script shopify-markets-localization-auditor.mjs uses node:child_process.execFile to invoke the Shopify CLI (shopify) for API operations. This is a legitimate and documented dependency for the skill's core functionality, used for executing GraphQL queries and managing store connections.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to connect to Shopify's Admin API and to crawl merchant storefronts. It utilizes a security-hardened fetch implementation in scripts/lib/public-fetch.mjs that prevents Server-Side Request Forgery (SSRF) by validating URL protocols and blacklisting private, local, and loopback IP address ranges.

shopify-markets-localization-auditor