The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Inclusion of external resources flagged as malicious. The skill references and includes snippets that load external scripts from domains flagged by automated scanners. Specifically, a code snippet in references/performance-optimization.md demonstrates loading a script from https://third-party.com/widget.js, which is flagged for botnet activity.
[METADATA_POISONING]: Deceptive metadata and references to blacklisted infrastructure. The skill frontmatter in SKILL.md identifies the author as Jeffallan and links to documentation on jeffallan.github.io, a domain flagged by scanners as blacklisted. This contradicts the provided author context of emmraan. Furthermore, the SKILL.md file itself is flagged for malware reputation issues.
[REMOTE_CODE_EXECUTION]: Patterns for dynamic script loading from untrusted external sources. The performance optimization reference provides an implementation for injecting third-party scripts at runtime (script.src = 'https://third-party.com/widget.js'), which, combined with the flagged reputation of the target domain, presents a significant remote code execution risk.
[INDIRECT_PROMPT_INJECTION]: Vulnerable data ingestion surfaces without boundary protection. The skill implements various patterns for processing untrusted data from metafields and user inputs across themes, apps, and checkout extensions. There is a lack of mandatory evidence chain components such as explicit boundary markers or sanitization logic in the provided code examples to prevent the agent from acting on instructions embedded within Shopify store data.

shopify-expert