marionette-flutter-drive-app
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from the target Flutter application's UI and logs, which creates a surface for indirect prompt injection if the application displays content controlled by an external attacker.\n
- Ingestion points: Data is ingested via
get_interactive_elements,take_screenshots, andget_logs(documented in SKILL.md).\n - Boundary markers: Absent. There are no specific instructions or delimiters provided to the agent to distinguish between legitimate UI content and potential malicious instructions.\n
- Capability inventory: The agent can perform UI interactions (
tap,enter_text), trigger developer workflows (hot_reload,hot_restart), and execute shell commands via themarionetteCLI bridge.\n - Sanitization: Absent. There is no evidence of filtering or sanitization of the retrieved UI text or log data.\n- [COMMAND_EXECUTION]: The skill requires the execution of shell commands for installation and operation.\n
- Evidence: Instructions include
flutter pub add marionette_flutter,dart pub global activate marionette_mcp, and usage of themarionetteCLI (e.g.,marionette tap --key ...).\n - Context: These commands target vendor-owned resources (
marionette_flutter,marionette_mcp,marionette_cli) from the authorleancodepland represent standard integration procedures.
Audit Metadata