The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation process involves cloning a remote repository and executing scripts via pnpm install and pnpm tools-dev, which downloads and runs external code from the project source.
[COMMAND_EXECUTION]: The skill uses child_process.spawn and exec to run various coding agent CLIs (e.g., claude-code, cursor-agent, gemini) that are detected on the user's system PATH.
[EXTERNAL_DOWNLOADS]: Installation procedures require cloning a repository from https://github.com/nexu-io/open-design.git and downloading binary installers for macOS and Windows from https://open-design.ai.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied prompts and discovery form data which are interpolated into the system prompts sent to the design engine.
Ingestion points: User request prompts and discovery questionnaire answers within the QuestionForm component.
Boundary markers: The provided prompt templates lack explicit delimiters or instructions to isolate user input from system instructions in the provided snippets.
Capability inventory: The skill can write files to the local project workspace, execute shell commands to spawn agents, and perform network requests to AI provider APIs.
Sanitization: While the system includes SSRF validation for proxy URLs, no specific sanitization or escaping of user-provided text input is documented.
[PRIVILEGE_ESCALATION]: Troubleshooting documentation includes instructions for the user to manually execute sudo apt-get install to provide necessary system dependencies for the PDF export functionality.

open-design-ai-prototyping