The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided URLs as input and uses a browsing tool to visit these sites, extracting layout and style data to inform its design specifications. This creates a surface for indirect prompt injection where a malicious website could contain hidden instructions designed to manipulate the agent's output or influence the behavior of downstream engineering agents during the handoff process.
Ingestion points: inspiration-intake.md (Step 1, processing content from external URLs via browse skill).
Boundary markers: Absent. The instructions do not specify delimiters or explicit warnings to ignore natural language instructions found within the processed external sites.
Capability inventory: The skill has the ability to Write artifacts and use the Skill tool to delegate tasks to other agents (nextjs-senior-engineer, react-vite-tailwind-engineer).
Sanitization: Absent. There is no evidence of filtering or validation of the extracted data for potential injection patterns.
[DYNAMIC_EXECUTION]: The inspiration-intake.md reference file explicitly instructs the agent to use JavaScript evaluation (browse JS eval) on external, untrusted websites to extract computed CSS properties like fontFamily, fontSize, and borderRadius. While this is used for legitimate design analysis, performing dynamic execution on third-party web content is a sensitive operation.

frontend-design-ui-ux