TabletopExercise

Fail

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONOBFUSCATIONCREDENTIALS_UNSAFEDYNAMIC_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides 'technical atomics' that instruct users or agents to download and execute shell scripts from an external IP address (203.0.113.50) using shell piping (e.g., curl | sh). Evidence is found in README.md and ATOMICS-LIBRARY.md.\n- [OBFUSCATION]: The ATOMICS-LIBRARY.md file contains a multi-layer Base64-encoded PowerShell payload. The command uses [Convert]::FromBase64String and appears to contain Gzip-compressed content, a technique commonly used to hide malicious code from static analysis.\n- [CREDENTIALS_UNSAFE]: Hardcoded credentials (testuser / TestPass123!) are provided in ATOMICS-LIBRARY.md for a simulated victim workstation, which is a poor security practice.\n- [DYNAMIC_EXECUTION]: The skill generates 'executable' action sequences and provides pre-written malicious payloads (remote shell pipes and obfuscated PowerShell), increasing the likelihood of dangerous code being executed.\n- [COMMAND_EXECUTION]: The PDF generation component (e.g., pdf-generator/package.json) requires the installation of external dependencies like playwright and the execution of shell commands, which expands the potential attack surface.\n- [EXTERNAL_DOWNLOADS]: The skill facilitates downloading scripts from an untrusted remote IP address (203.0.113.50), which is a critical vector for delivering second-stage malware.
Recommendations
  • HIGH: Downloads and executes remote code from: http://203.0.113.50/stage2.sh - DO NOT USE without thorough review
  • AI detected serious security threats
Audit Metadata
Risk Level
CRITICAL
Analyzed
Nov 9, 2026, 12:00 PM