Skills
skills/majesticlabs-dev/majestic-marketplace/external-llm-review/Gen Agent Trust Hub

external-llm-review

Warn

Audited by Gen Agent Trust Hub on Mar 24, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill gathers source code changes via git diff and project context from CLAUDE.md and .agents/lessons/, then transmits this information to external APIs through the codex and gemini command-line tools. This represents a potential exposure of proprietary code to third-party services.\n- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection. It interpolates unvalidated content from git diff directly into a review prompt without boundary markers or sanitization, which could allow malicious code comments to manipulate the model's review verdict.\n
  • Ingestion points: git diff and local project documentation (SKILL.md)\n
  • Boundary markers: Absent\n
  • Capability inventory: Shell execution of external LLM CLIs (codex exec, gemini)\n
  • Sanitization: Absent\n- [COMMAND_EXECUTION]: The skill invokes several local system commands to gather context and interact with external tools, including git, cat, ls, codex, and gemini.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Mar 24, 2026, 09:57 PM