The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines an architecture for building applications that ingest and process data from external sources such as APIs, databases, and user uploads.
Ingestion points: Untrusted data enters the system via Core.Net.HttpRequest in the robot flow and is displayed in the UI through components like DataTable, Thread, and Markdown.
Boundary markers: The instructions mandate the use of the Markdown component for rendering external content, noting that it "sanitises what it renders."
Capability inventory: The robot backend has capabilities for file system access (Core.FileSystem) and network operations (Core.Net.HttpRequest).
Sanitization: The skill explicitly forbids the use of dangerouslySetInnerHTML and provides clear guidance on using sanitizing components for all external data to prevent injection attacks.
[COMMAND_EXECUTION]: The skill utilizes the robomotion CLI for project management, authentication, and deployment. These operations are standard for the vendor's development environment and are performed using official tooling.
[DYNAMIC_EXECUTION]: The skill facilitates the generation of React and TypeScript code for application development. This process is constrained by "Hard Rules" that enforce the use of a pre-approved component library (@robomotion/app-kit) and generated typed stubs, minimizing the risk of arbitrary code execution.
[CREDENTIALS_UNSAFE]: The skill includes instructions for managing authentication sessions. It follows security best practices by explicitly warning against committing session files (.robomotion/session.json) to version control and advising the agent to use browser-based approval flows instead of asking for API keys.

building-app