waf-bypass-agent

Pass

Audited by Gen Agent Trust Hub on Mar 9, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The instructions direct the agent to perform active network requests against a user-defined target endpoint. This includes sending various payload variants involving encodings, structural changes, and content-type smuggling to observe server behavior.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection due to its processing of external data.
  • Ingestion points: The agent is instructed to capture and analyze HTTP response bodies and headers in Phase 1 (Signal Collection) and Phase 3 (Branch A/B/C) of the SKILL.md file.
  • Boundary markers: The instructions do not define boundary markers or provide warnings to the agent to disregard instructions potentially embedded within the analyzed HTTP responses.
  • Capability inventory: The agent has the capability to perform network requests and modify its testing logic based on the data it receives from the target throughout all phases.
  • Sanitization: The instructions lack any requirement for sanitizing or validating external content before it is processed by the agent to generate new hypotheses or test variants.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 9, 2026, 10:07 PM