The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes scripts for local environment configuration and diagnostics. \n
scripts/setup.sh executes claude mcp add to register the local MCP endpoint.\n
scripts/health-check.py uses subprocess.run to call pgrep and claude mcp list to verify the application status and registration.\n- [PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection through ingested design content.\n
Ingestion points: Untrusted design data (HTML, JSX, text) enters the context via tools such as get_node_info, get_jsx, and get_tree_summary as documented in SKILL.md and references/tool-reference.md.\n
Boundary markers: No delimiters or instructions are used to distinguish design content from system instructions or to ignore embedded commands.\n
Capability inventory: The skill possesses the capability to modify design state via write_html and update_styles, and it includes scripts capable of system command execution.\n
Sanitization: There is no evidence of content sanitization or validation for the data retrieved from the design artboards.

paper-design