The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from websites, images, and Figma links, creating a surface for indirect prompt injection attacks where hidden instructions in the source material could manipulate agent behavior.\n
Ingestion points: Ingestion occurs via the web_fetch tool and scripts/extract_css_vars.py for URLs, and through Figma MCP tools (get_design_context) as defined in references/capture-flows.md.\n
Boundary markers: Absent. The instructions do not require the agent to use delimiters or specific warnings to ignore instructions found within extracted design content.\n
Capability inventory: The skill has access to bash for executing internal Python/TypeScript scripts, write_to_file for generating design documentation, and web_fetch for network requests.\n
Sanitization: Absent. No mechanisms are described for sanitizing or filtering extracted HTML content or CSS variables to prevent them from being interpreted as natural language instructions.

designscope