The Agent Skills Directory

[COMMAND_EXECUTION]: The skill writes a .mcp.json configuration file to the workspace. This file directs the agent platform to execute an MCP server (npx -y @formio/mcp) to interface with the Form.io API. The process is gated by a user approval step that displays the full JSON content before writing.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to fetch and run the @formio/mcp package from the npm registry. As this is a package from the skill's own vendor, it is considered a standard dependency for the tool's operation.
[DATA_EXFILTRATION]: The agent is instructed to perform a PUT network request to a project URL provided by the user. This request is used to set public configuration values (like appUrl) within the Form.io project. This is a legitimate configuration step and is preceded by a user confirmation preview showing the target URL and data.
[INDIRECT_PROMPT_INJECTION]: The skill processes natural language input from users to generate application models, which introduces a surface for indirect prompt injection. The skill implements mitigation strategies by using structured data formats (template.json) for handoffs and requiring user approval at multiple stages (planning, importing, and scaffolding).
Ingestion points: User descriptions of application features in SKILL.md and INTENT.md are passed to a planner skill.
Boundary markers: Explicit approval gates are required before writing configuration (MCP_CONFIG.md), importing data (IMPORT.md), and routing to frameworks (FRAMEWORK.md).
Capability inventory: Capabilities include writing .mcp.json, calling the project_import and project_set tools, and making network requests to the Form.io API.
Sanitization: The skill relies on a separate planner to convert natural language into a structured JSON template, preventing direct interpolation of untrusted text into command arguments or sensitive files.

formio-application