The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill leverages the fmt_evaluate_dart_expression tool to execute code at runtime within the application's VM. This is an intended feature for developer inspection of application state.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with external data sourced from a running Flutter application, which represents a potential attack surface.
Ingestion points: Application state and errors are ingested via fmt_semantic_snapshot, fmt_get_app_errors, and tool outputs described in SKILL.md.
Boundary markers: No specific delimiters or safety instructions are defined to separate ingested data from agent instructions.
Capability inventory: The skill includes tools to simulate user input, modify code via hot reload, and execute expressions.
Sanitization: No explicit data sanitization or filtering is described for the information retrieved from the target application.
[COMMAND_EXECUTION]: The documentation mentions using local CLI commands (flutter-mcp-toolkit permissions) to manage macOS screen recording permissions, which is a standard procedure for UI testing and inspection tools.

flutter-mcp