The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to utilize npx @agent-native/core@latest to install and execute the vendor's CLI tools from the npm registry as part of the standard setup and maintenance workflow.
[COMMAND_EXECUTION]: The skill relies on shell commands for setup, authentication, and local plan previewing. Key commands include plan local serve and plan local verify, which are used to host and validate visual plans on the user's machine.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text plans to generate complex visual artifacts, which constitutes an attack surface for indirect prompt injection.
Ingestion points: The skill gathers input from chat history, pasted content, or referenced files to build its visual proposals.
Boundary markers: Instructions emphasize rewriting plans as standalone proposal documents, helping to isolate the source intent from the agent's internal planning logic.
Capability inventory: The skill can write MDX files to the local repository, execute shell commands via the local bridge, and communicate with the vendor's hosted service (plan.agent-native.com).
Sanitization: The documentation explicitly states that HTML wireframe content is re-sanitized during the rendering process to prevent malicious code execution from injected content.
[DYNAMIC_EXECUTION]: The skill dynamically generates and updates MDX files and HTML wireframe fragments at runtime. Tools like patch-visual-plan-source and update-visual-plan are used to manipulate these artifacts based on evolving plan data.

visual-plan