The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to fetch and process content from external, untrusted ecommerce websites using curl. This data ingestion creates a surface for indirect prompt injection, where malicious instructions embedded in a target website could attempt to influence the agent's behavior.
Ingestion points: curl commands targeting external domains (e.g., https://store.com) in SKILL.md.
Boundary markers: Absent from the provided command examples.
Capability inventory: Network access via curl and text processing via grep.
Sanitization: None specified for the fetched HTML content.
[COMMAND_EXECUTION]: The skill contains several examples of shell commands intended to be executed by the agent to identify store platforms and catalogue sizes.
Evidence: The automated alert flagging 'Remote code piped to Sh' is a false positive. The command curl -s https://store.com | grep -ioE 'cdn\.shopify\.com|Shopify\.theme|shopify-features' pipes data to grep for pattern identification. The scanner likely misidentified the string Shopify as a pipe to a shell (| Sh).
[EXTERNAL_DOWNLOADS]: The skill facilitates the fetching of data from external domains via curl. These operations are consistent with the skill's primary purpose of ecommerce prospecting and research.
Evidence: Examples include fetching store homepages and sitemap XML files (e.g., sitemap_products_1.xml) to estimate catalogue size.

ecommerce-store-prospecting