The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the user to run npx -y @gethopp/figma-mcp-bridge, which fetches a package from the external NPM registry. It also provides links for manual download of a Figma plugin from a GitHub repository (github.com/gethopp/figma-mcp-bridge/releases).\n- [REMOTE_CODE_EXECUTION]: The execution of npx -y @gethopp/figma-mcp-bridge results in the immediate execution of remote code on the host machine to start the MCP server.\n- [COMMAND_EXECUTION]: The save_screenshots tool allows the AI agent to specify an outputDir absolute path to save files. Without strict path validation or sandboxing, this capability could be abused to write files to sensitive system directories or overwrite existing files.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests design data, text, and metadata from external Figma documents which could be controlled by third parties. This data is then processed by the agent, potentially allowing for embedded instructions to influence agent behavior.\n
Ingestion points: get_document, get_node, get_selection, get_styles, and get_metadata tools in SKILL.md.\n
Boundary markers: None identified in the skill instructions.\n
Capability inventory: The skill has file system write capabilities via the save_screenshots tool.\n
Sanitization: No evidence of data sanitization or instruction filtering is provided for the ingested design content.

figma-mcp-bridge