What problem does it solve?
AI agents that handle file uploads, exports, or archive extraction may accept user-controlled paths, allowing files to be written outside the intended workspace. This Skill probes an agent's file handling behavior to confirm whether path traversal is possible before attackers exploit it.
Core Features & Use Cases
- Save Path Traversal Probes: Sends dialogue prompts requesting the agent to save output to traversal paths like "../../../../tmp/aig_report.txt" and checks whether the agent confirms writing outside the sandbox.
- Filename Injection Testing: Simulates uploaded archives containing traversal filenames such as "../../etc/cron.d/pwn" to test extraction handling.
- Structured Vulnerability Reports: Emits confirmed findings as XML <vuln> blocks with evidence, risk type, severity level, and remediation suggestions.
- Use Case: A security engineer red-teaming a file-handling AI agent runs this Skill to verify the agent normalizes paths and enforces a fixed base directory before deployment.
Quick Start
Test this agent for path traversal vulnerabilities by probing its file save and upload handling with traversal paths.