The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides fully functional exploit payloads for multiple vulnerabilities, including Spring Expression Language (SpEL) injection (CVE-2022-22963, CVE-2022-22947) and Spring4Shell (CVE-2022-22965), which allow for arbitrary OS command execution on a target system.
[DATA_EXFILTRATION]: Specifically targets the extraction of sensitive information, including plaintext passwords, API keys (AWS, Stripe), and bearer tokens, from application endpoints (/actuator/env) and JVM memory dumps (/actuator/heapdump).
[PERSISTENCE]: Includes instructions for deploying a JSP-based webshell into the target's web server directory to maintain persistent unauthorized access.
[COMMAND_EXECUTION]: Utilizes shell commands and piping to interact with remote targets. The skill includes patterns where remote content is piped directly to python3 -m json.tool. While used for formatting, this pattern is flagged for potential subversion if the target output is manipulated to exploit the interpreter.
[INDIRECT_PROMPT_INJECTION]: The skill processes unvalidated content from remote targets via curl. 1. Ingestion points: curl responses from $TARGET in all phases. 2. Boundary markers: Absent. 3. Capability inventory: curl, grep, strings, python3. 4. Sanitization: Absent. The lack of robust sanitization creates an attack surface where a malicious target could influence agent behavior.
[REMOTE_CODE_EXECUTION]: Documents the H2 Console RCE technique, providing a specific SQL 'CREATE ALIAS' payload designed to execute Java code and system commands on the host system.

hunt-springboot