command-injection-advanced

Fail

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill documentation references numerous open-source security tools hosted on GitHub, such as commix, tplmap, and nosqlmap, which are standard utilities for security assessment and vulnerability detection.
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to run security binaries and custom scripts (e.g., nosqli-scan.py, bypass-generator.py) for the purpose of demonstrating and testing injection vulnerabilities.
  • [DATA_EXFILTRATION]: Instructions are provided for using out-of-band exfiltration services such as Burp Collaborator, interact.sh, and webhook.site to confirm blind injection vulnerabilities, representing legitimate testing methodology.
  • [INDIRECT_PROMPT_INJECTION]: The skill utilizes the WebFetch tool to retrieve external data from targets, making it susceptible to indirect prompt injection if the ingested data contains malicious instructions.
  • Ingestion points: Target web responses and API data retrieved through the WebFetch tool.
  • Boundary markers: Absent; the instructions do not define delimiters for separating target data from prompt logic.
  • Capability inventory: Capabilities include Bash, Read, Write, Edit, and WebFetch tools.
  • Sanitization: No input validation or sanitization mechanisms are implemented for the agent's processing of retrieved data.
  • [SAFE]: The automated findings regarding malicious URLs (HackTricks) and exfiltration destinations (webhook.site) are false positives in this context, as these are recognized resources for security research and out-of-band vulnerability testing.
Recommendations
  • Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Sep 19, 2026, 04:36 PM