local-ai-app-integration
Fail
Audited by Gen Agent Trust Hub on Aug 26, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill directs the agent to download binary archives (Windows .zip and Linux .tar.gz) from the
lemonade-sdk/lemonadeGitHub repository. This repository is not identified as a trusted source. - [REMOTE_CODE_EXECUTION]: The skill provides reference launcher implementations in Python and Node.js that use
subprocess.Popenandspawnto execute the downloadedlemondbinary as a persistent private subprocess. - [COMMAND_EXECUTION]: The integration sequence involves running several shell and PowerShell commands to create directory structures, fetch release metadata from the GitHub API, download artifacts, and perform file system operations like
Expand-Archiveandtarto install the binary into the application's vendor tree. - [PROMPT_INJECTION]: The instructions explicitly command the agent to bypass application-level security controls, stating that 'any onboarding wall, validator, or startup check that demands [an API key] must not block local-mode users' and to 'auto-satisfy the key-entry screen'.
Recommendations
- AI detected serious security threats
Audit Metadata