business-logic-vulnerabilities

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for the agent to ingest and analyze untrusted data from external web applications (e.g., API responses, headers, and body content) to identify security flaws. This process inherently exposes the agent to an indirect prompt injection attack surface if an adversary embeds malicious instructions in the target application's output. However, this is an expected risk for a tool designed for security auditing. Evidence: The skill instructs the agent to evaluate external fields such as 'status', 'is_login', and 'prizeIdList' from captured network traffic (SKILL.md, CHECKLIST.md).- [COMMAND_EXECUTION]: The playbook documents numerous shell commands (curl, bash) and script examples (Python, PHP) as reproduction steps and attack patterns. These commands are intended to be executed by the auditor against a target system and do not target the host environment or the agent's internal configuration. Evidence: Examples include race condition scripts using Turbo Intruder and curl commands for parameter tampering (SCENARIOS.md, SKILL.md).- [EXTERNAL_DOWNLOADS]: The documentation references several well-known and legitimate security tools and public GitHub repositories, such as Burp Suite, Tesseract OCR, Gopherus, and host_scan. These are provided as part of a recommended toolkit for security professionals and do not constitute a supply chain risk. Evidence: References are found in the methodology and scenario sections (METHODOLOGY.md, SCENARIOS.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 07:15 PM