The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes a bash script scripts/fetch-stitch.sh to download design content from external URLs. These URLs are dynamically provided at runtime by the Stitch MCP get_screen tool. While intended for Google Cloud Storage resources, the destination is computed during execution.\n- [COMMAND_EXECUTION]: The skill provides and instructs the agent to execute a local bash script to perform network operations. This script uses curl to fetch remote resources and save them to the local file system.\n- [PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection as it processes untrusted HTML and JSON data from external sources.\n
Ingestion points: Data retrieved via stitch-mcp-get-screen and the contents of downloaded HTML files (e.g., temp/source.html).\n
Boundary markers: No explicit delimiters or 'ignore' instructions for the ingested content are present in the skill's guidelines.\n
Capability inventory: The skill has access to Bash, Read, and Write tools, enabling it to perform shell commands and modify files based on its interpretation of external data.\n
Sanitization: There are no documented validation or sanitization routines for the design metadata or HTML content before it is processed by the agent.

stitch-uview-components