lemo-opuscar

Warn

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill's scripts/setup.sh clones an external repository from https://github.com/lemomo-ai/lemo-opuscar.git and executes code from it. It performs automated installation of dependencies using npm install and pip install based on remote package.json and requirements.txt files.
  • [EXTERNAL_DOWNLOADS]: The skill downloads approximately 191 MB of library data, scripts, and media styles from an external GitHub repository. It also fetches binary models for speech synthesis using a remote tools/fetch.sh script.
  • [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to follow logic and instructions contained within $LIB/AGENTS.md and $LIB/DIRECTOR.md. Since these files are downloaded at runtime from a third-party source, they represent a surface for indirect prompt injection.
  • Ingestion points: SKILL.md (Step 3) directs the agent to read $LIB/AGENTS.md.
  • Boundary markers: None present in the instructions to distinguish remote instructions from system constraints.
  • Capability inventory: The skill has access to shell execution, node.js execution, python execution, and file system writes.
  • Sanitization: No sanitization of the remote markdown instructions is performed before ingestion.
  • [DYNAMIC_EXECUTION]: The skill installs Node.js and Python packages at runtime and executes various scripts from the library (e.g., node core/render/still.mjs). It also compiles or runs code provided in STYLE.md files within the downloaded library.
  • [COMMAND_EXECUTION]: The agent is instructed to run setup scripts (sh setup.sh) and production tools with various arguments. While it includes a warning not to install system-level software, it grants the library tools broad execution capabilities within the project directory.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 10, 2026, 01:40 PM