argent-device-interact

Warn

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: MEDIUMCREDENTIALS_UNSAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [CREDENTIALS_UNSAFE]: The file references/harmonyos.md includes the hardcoded password 123456 for the OpenHarmony.p12 test keystore. Although this is documented as a public test key for development purposes, hardcoding any functional credential in skill instructions is a security concern.
  • [COMMAND_EXECUTION]: The skill provides documentation for building and signing HarmonyOS Application Packages (HAP), which includes shell commands such as ohpm install, node hvigorw.js, and java -jar hap-sign-tool.jar. These commands are intended to be executed in the user's local development environment to facilitate app deployment.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests data from external, untrusted sources (application UIs) and possesses high-privilege capabilities.
  • Ingestion points: Untrusted data enters the agent context through tools like describe, native-find-views, and debugger-component-tree which read UI trees and properties.
  • Boundary markers: The skill does not provide clear delimiters or instructions to the agent to ignore or sanitize content found within the application UI elements.
  • Capability inventory: The skill has access to sensitive operations including entering secrets (keyboard), executing JavaScript on web targets (debugger-evaluate), and opening arbitrary URLs (open-url).
  • Sanitization: No sanitization or validation logic is defined for the content retrieved from device screens before the agent processes it.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 10, 2026, 10:42 PM