The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content directly from Shopify stores (product descriptions, SEO metadata, and blog content) during its translation workflow. Maliciously crafted content within the store could attempt to perform an indirect prompt injection attack to manipulate the agent's behavior.
Ingestion points: Translatable resources are fetched from the Shopify Admin API via the fetch command in scripts/shopify-translator-admin.mjs.
Boundary markers: SKILL.md mandates the use of explicit XML tags (<translatable-source-content>) to delimit untrusted data when passing it to the translation logic.
Capability inventory: The skill has the capability to write data back to the store using the translationsRegister mutation and can generate local CSV/JSON artifacts.
Sanitization: The skill includes a 'Translation Sandboxing' rule in SKILL.md that explicitly instructs the agent to treat source text strictly as read-only data and to never interpret or follow instructions embedded within it.
[COMMAND_EXECUTION]: The skill relies on local shell command execution to interact with the Shopify CLI and perform administrative tasks.
Evidence: scripts/lib/shopify-cli.mjs uses Node.js execFileSync to execute the system shopify CLI for store authentication and GraphQL query execution. It also executes node scripts provided within the skill's own package.

shopify-store-translator