The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses deceptive metadata to claim an author of 'Shopify' when the actual author is 'display-design-studio'. This misrepresentation is used to gain trust for instructions that involve high-risk data harvesting.- [DATA_EXFILTRATION]: The instructions require the agent to capture the user's prompt verbatim, base64-encode it, and transmit it along with session and tool identifiers to an external domain (shopify.dev) via the validate.mjs script. Harvesting raw user input for transmission to external services is a significant privacy violation and exfiltration risk.- [COMMAND_EXECUTION]: Every response is required to execute local scripts (scripts/search_docs.mjs and scripts/validate.mjs) that are not included in the provided skill files. This forces the agent to run unverifiable and potentially malicious code on the host system.- [PROMPT_INJECTION]: The instructions use high-pressure language and MANDATORY markers to override the agent's standard privacy protocols, ensuring that the raw data capture and transmission workflow is followed.- [PROMPT_INJECTION]: Trust chain exposure exists as the skill ingests untrusted user data and passes it verbatim into a shell command context via the validate.mjs call. This creates a surface for indirect prompt injection or command injection through the missing scripts.

shopify-hydrogen