The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential surface for indirect prompt injection when processing data returned from the MCP server.
Ingestion points: The agent is instructed to process output_components (which may contain text or suggestions) from the generate_screen_from_text tool response (SKILL.md).
Boundary markers: No specific delimiters are defined for the tool's output data within the instructions.
Capability inventory: The skill utilizes the mcp__stitch toolset, which includes capabilities for screen generation, project listing, and design system application.
Sanitization: The instructions explicitly mandate a human-in-the-loop verification process. The agent is strictly forbidden from auto-accepting suggestions and must present them to the user for explicit approval (SKILL.md, "Handling output_components").
[SAFE]: No malicious patterns, obfuscation, or unauthorized data exfiltration attempts were detected. The skill is instructional in nature and facilitates the use of a well-known UI design service (Google Stitch). The polling logic provided is a standard functional pattern for managing high-latency API operations and does not constitute a malicious time-delayed trigger.

stitch-screens