web-auth-jwt

Pass

Audited by Gen Agent Trust Hub on Sep 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute various CLI tools, including jwt_tool, hashcat, and john, passing external data (tokens) as arguments to these commands.
  • [INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection and command injection. It ingests untrusted data in the form of JWT tokens from target web applications and processes them through shell-based tools.
  • Ingestion points: External JWT tokens are captured from HTTP headers (e.g., Authorization: Bearer) or cookies and processed in SKILL.md and cheatsheet.md.
  • Boundary markers: The skill does not provide specific instructions to the agent to sanitize or validate the tokens before passing them to the shell, nor does it use explicit boundary markers for the data.
  • Capability inventory: The agent is empowered to run shell commands (jwt_tool, hashcat) using the captured tokens as inputs across both documentation files.
  • Sanitization: There are no explicit sanitization or escaping steps defined for the external tokens before interpolation into command-line arguments.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 20, 2026, 11:01 PM
Security Audit — agent-trust-hub — web-auth-jwt