java-deserialization-methodology
Pass
Audited by Gen Agent Trust Hub on May 16, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONDATA_EXFILTRATIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: The skill includes the well-known default Apache Shiro AES encryption key (
kPH+bIxk5D2deZiIxcaaaA==) in its instructions and evaluation files for use in vulnerability identification and exploit examples. - [COMMAND_EXECUTION]: Provides specific command-line instructions for executing security tools such as
ysoserial.jar,JNDIExploit.jar, and various Python exploitation scripts against target systems. - [DATA_EXFILTRATION]: Recommends the use of external DNSLog services (e.g.,
dnslog.cn) to perform out-of-band (OOB) verification of vulnerabilities, which involves sending DNS queries to external infrastructure. - [REMOTE_CODE_EXECUTION]: Contains functional exploit payloads for JNDI injection (targeting Log4j and Fastjson) and Java serialization gadgets (via ysoserial) designed to achieve code execution on target systems.
- [PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) by processing untrusted data from target server responses without explicit boundary markers.
- Ingestion points: Target server responses including Cookies (
rememberMe), HTTP Headers, and POST bodies as identified inSKILL.mdandreferences/jndi-fastjson.md. - Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present in the provided templates.
- Capability inventory: The agent is instructed to perform network requests (
http_request) and is given context to suggest/execute local commands as detailed inreferences/ysoserial-gadgets.md. - Sanitization: There is no evidence of sanitization or validation of the data retrieved from external targets before processing.
Audit Metadata