The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and project identifiers to generate visual screens via MCP tools. This ingestion of untrusted data creates a surface for indirect prompt injection if the user input or the data retrieved from external tools contains malicious instructions.
Ingestion points: User-provided prompt and projectId arguments in SKILL.md and references/workflow.md; data retrieved from list_screens and get_screen MCP tool outputs.
Boundary markers: The skill suggests a structured prompt format (e.g., [Device] [Mode] [Screen Type]. [Style]. [Layout]. [Components].) but does not implement explicit delimiters or instruction-ignore blocks for data interpolated into tool calls.
Capability inventory: Uses the generate_screen_from_text, list_screens, get_screen, and get_project MCP tools to create and manage UI designs.
Sanitization: The skill includes instructions to replace real customer data (names, phone numbers, order IDs) with demonstration values and to strip credentials or session cookies before outputting reports.
[EXTERNAL_DOWNLOADS]: The skill references an external code repository for implementation logic.
Evidence: Links to a specific commit at https://github.com/google-labs-code/stitch-skills/tree/0337446dadde6f8c94210444e2aa9d546126480f as a basis for the skill's implementation. This points to a well-known service.

stitch-mcp-generate-screen-from-text