Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Remote Package Invocation: The skill utilizes
npxto execute theemulatetool from the npm registry. This pattern involves fetching and running code at runtime, which is a common practice for development utilities but introduces an external dependency. - Local Command Execution: The skill employs
Bashto run emulator commands andcurlto simulate API requests. These operations are scoped to local listeners (defaulting to127.0.0.1) to facilitate testing without connecting to live Google services. - Mock Credential Management: Examples in the documentation include mock OAuth
client_secretand user data for seeding the emulator. These values are used for local configuration and are not real production secrets. - Indirect Prompt Injection Surface: The skill allows seeding the emulator with mock data (e.g., email bodies in
messages, file contents indrive_items) withinSKILL.md. This data enters the agent's context during testing. While the emulator facilitates local development, the absence of explicit boundary markers or sanitization in the mock data examples means that any instructional text embedded in these fields could potentially influence an agent processing the emulator's output.
Audit Metadata