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 inREADME.mdandATOMICS-LIBRARY.md.\n- [OBFUSCATION]: TheATOMICS-LIBRARY.mdfile contains a multi-layer Base64-encoded PowerShell payload. The command uses[Convert]::FromBase64Stringand 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 inATOMICS-LIBRARY.mdfor 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 likeplaywrightand 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