ai-agents

Fail

Audited by Gen Agent Trust Hub on Oct 16, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill provides links to official GitHub documentation for OpenAI's agents SDK (e.g., openai.github.io). These are legitimate references to well-known service providers and do not involve untrusted executable content.
  • [INDIRECT_PROMPT_INJECTION]: Usage tracking and evaluation scripts process local log data containing prior user interactions. While this constitutes an ingestion surface for untrusted data, the scripts are restricted to token accounting and pass-rate analysis, do not process the data through an LLM, and lack network exfiltration capabilities.
  • [DYNAMIC_EXECUTION]: Test scripts (scripts/test_codex_usage.py) use subprocess.run to execute Python code snippets extracted from the skill's own markdown documentation. This execution is limited to the local testing environment for verification of documentation accuracy and does not impact the runtime security of the agent.
Recommendations
  • Contains 7 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Oct 16, 2026, 03:04 PM