JWT Security Testing

Warn

Audited by Gen Agent Trust Hub on Nov 6, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent to clone the jwt_tool repository from an external GitHub account and install its dependencies using pip, which involves executing third-party code.
  • [DYNAMIC_EXECUTION]: Several phases of the workflow use shell here-documents (python3 << 'EOF') to dynamically generate and execute Python code for token forgery and cryptographic attacks.
  • [COMMAND_EXECUTION]: The workflow heavily relies on shell commands to invoke system tools such as bash, curl, hashcat, and john for penetration testing tasks.
  • [EXTERNAL_DOWNLOADS]: The skill downloads public keys from remote target servers using curl and clones software from GitHub repositories.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external JWT tokens, which could be crafted to influence agent behavior through the processing logic.
  • Ingestion points: User-provided tokens are interpolated directly into shell commands and Python script blocks in SKILL.md.
  • Boundary markers: The instructions do not use specific delimiters or instructions to prevent the agent from obeying data contained within the tokens.
  • Capability inventory: The skill has access to network operations (curl), file system manipulation, and shell execution.
  • Sanitization: There is no evidence of token validation or sanitization before the data is processed by the agent's internal scripts.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 6, 2026, 03:30 PM