The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The evaluate_dart_expression tool allows for the evaluation of arbitrary Dart code strings within the application's running process. While intended for inspecting live app state (e.g., AgentState.instance.value), it provides a mechanism for dynamic code execution at runtime.
[COMMAND_EXECUTION]: The instructions direct the agent to execute various shell commands using the flutter-mcp-toolkit CLI, such as doctor, exec, permissions, and snapshot, to manage the debug session and troubleshoot connection issues.
[DATA_EXFILTRATION]: The skill facilitates access to potentially sensitive application data through multiple vectors:
get_recent_logs retrieves the last 100 log lines from the application, which may contain sensitive runtime information or PII.
evaluate_dart_expression can be used to serialize and extract any object or state currently held in the application's memory.
[INDIRECT_PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect instructions through data processed from the application.
Ingestion points: Untrusted data enters the context via get_recent_logs and the output of evaluate_dart_expression (SKILL.md).
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within the logs or app state.
Capability inventory: The agent has the ability to execute arbitrary Dart code in the app and run host-level CLI commands via flutter-mcp-toolkit (SKILL.md).
Sanitization: There is no mention of sanitizing or escaping the content retrieved from the app before it is presented to the model.

flutter-mcp-toolkit-debug