The Agent Skills Directory

[DYNAMIC_EXECUTION]: The evaluate_dart_expression tool allows for the execution of arbitrary Dart code within the running Flutter application's main isolate. While intended for debugging and state inspection, this capability allows the execution of code that can modify the live application environment or access internal memory.
[COMMAND_EXECUTION]: The 'Error envelope playbook' section contains multiple instructions for the agent to execute shell commands using the flutter-mcp-toolkit CLI tool (e.g., flutter-mcp-toolkit doctor --json, flutter-mcp-toolkit exec --name status). These commands are used for environmental diagnostics and error recovery.
[DATA_EXFILTRATION]: The skill provides tools to retrieve internal application state (evaluate_dart_expression) and recent log output (get_recent_logs). If the target application logs sensitive information (such as credentials, PII, or tokens) or stores them in reachable variables like AgentState, these tools can expose that data to the agent.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The agent is instructed to read and interpret data provided by the application via logs, Dart expression results, and error envelopes (error.details).
Boundary markers: No specific delimiters or safety instructions are provided to help the agent distinguish between diagnostic data and potentially malicious instructions embedded within logs or error messages.
Capability inventory: The skill possesses the capability to execute shell commands (via the toolkit CLI) and arbitrary Dart code within the app isolate based on its diagnosis of the ingested data.
Sanitization: There is no evidence of sanitization or filtering applied to the application logs or error details before they are processed by the agent.

flutter-mcp-toolkit-debug