local-ai-app-integration
Warn
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: MEDIUMEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads binary artifacts (lemonade-embeddable) from the lemonade-sdk GitHub repository, which is not a recognized trusted vendor or a known resource of the skill author.
- [REMOTE_CODE_EXECUTION]: The instructions guide the agent to download, extract, and subsequently execute the lemond binary using the subprocess module in Python or child_process in Node.js.
- [COMMAND_EXECUTION]: The skill provides a reference launcher implementation that uses subprocess.Popen to spawn a private server process, managing it via command-line arguments and environment variables.
- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. Ingestion points: user application code (SKILL.md Step 1) and API responses from the local server (reference.md). Capability inventory: subprocess calls in lemond_launcher.py and binary downloads in SKILL.md. Boundary markers: Absent. Sanitization: Absent. The skill ingests untrusted data from the user's repository and from the lemond server's API responses without explicit sanitization or boundary markers.
- [DYNAMIC_EXECUTION]: The skill involves the dynamic generation of a launcher module (lemond_launcher.py) that orchestrates the execution of external binaries and handles networking tasks like port probing and health checks at runtime.
Audit Metadata