The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data from multiple sources to generate a final report, creating a surface for indirect prompt injection.
Ingestion points: The skill accepts a natural-language product_feature description and performs web research on public sources (Step 1B), as well as retrieving records from external patent databases (Step 2).
Boundary markers: The instructions include explicit directives to "Escape all user, web, and patent text" and to clearly differentiate between "Official product statements," "Analytical inference," and "Secondary reports."
Capability inventory: The skill utilizes MCP tool calls for patent searching, performs web-based research, and writes a self-contained HTML file to the local environment.
Sanitization: The skill provides strong remediation guidance by requiring the agent to escape all content, validate URLs (using https only), and prohibit the loading of external JavaScript, CDNs, or trackers.
[DYNAMIC_EXECUTION]: The skill generates a self-contained HTML report that includes embedded CSS and minimal JavaScript for a dimension filter.
Evidence: Step 3 instructs the agent to "Embed CSS and minimal filter JavaScript in the single file" and warns "Never inject retrieved HTML or JSON as executable code."
Risk: While the generated script is for UI filtering, the process of code generation from external data requires strict adherence to the provided sanitization rules to prevent cross-site scripting (XSS) in the generated report.

map-product-features-to-patents-ip