lucida-remove-background

Fail

Audited by Gen Agent Trust Hub on Aug 15, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to install a command-line interface (CLI) tool directly from a third-party GitHub repository (git+https://github.com/egeorcun/lucida.git) using the uv tool installer. This facilitates the execution of remote code not vetted by a standard package registry.\n- [COMMAND_EXECUTION]: The skill requires the execution of shell commands (bgr remove, uv tool install) to perform its primary functions. It specifically instructs the agent to run these commands in the "user's normal terminal" to avoid sandbox timeouts, which is an attempt to bypass environment-specific security restrictions.\n- [EXTERNAL_DOWNLOADS]: The skill initiates the download of machine learning model weights from Hugging Face (egeorcun/lucida and ZhengPeng7/BiRefNet_HR) during initial execution.\n- [PROMPT_INJECTION]: The skill processes external image files, providing a potential vector for indirect prompt injection attacks.\n
  • Ingestion points: User-provided image files processed via file paths in SKILL.md.\n
  • Boundary markers: None present to delimit untrusted data or warn against embedded instructions.\n
  • Capability inventory: Includes shell command execution (uv, bgr) and file system write operations (SKILL.md).\n
  • Sanitization: No sanitization or validation of the input image content or file metadata is performed.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Aug 15, 2026, 06:03 PM
Security Audit — agent-trust-hub — lucida-remove-background