arabic-agent-eval

Warn

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: MEDIUMEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill directs the user to install the arabic-agent-eval package via pip.\n
  • Evidence: pip install arabic-agent-eval in SKILL.md.\n
  • Context: This package is not associated with a verified or trusted vendor. Installation of third-party packages from public registries without verification carries a supply chain risk, as the package content could be modified or malicious.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process Arabic natural language data (dialects and function definitions) for benchmarking, which exposes the agent to indirect prompt injection.\n
  • Ingestion points: The skill utilizes a dataset described in SKILL.md containing 51 evaluation items and 22 function schemas across multiple Arabic dialects.\n
  • Boundary markers: No boundary markers or instructions to ignore embedded commands are present in the provided instructions.\n
  • Capability inventory: The aae command-line tool performs function-calling evaluations, which involves the agent processing the untrusted text and interacting with external LLM APIs (OpenAI, Anthropic).\n
  • Sanitization: There is no evidence of sanitization or validation of the Arabic text dataset before it is processed by the agent.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 11, 2026, 04:09 PM