The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains numerous payloads for achieving Remote Code Execution (RCE) across multiple Java template engines, including Freemarker, Velocity, Spring Expression Language (SpEL), Thymeleaf, Pebble, and Groovy. Evidence includes:
Freemarker: Use of freemarker.template.utility.Execute to run system commands.
Velocity: Reflection-based payloads using java.lang.Runtime to execute OS commands.
SpEL: Usage of the T() operator to access java.lang.Runtime or java.lang.ProcessBuilder for code execution.
Groovy: Direct execution via ".execute().text".
[COMMAND_EXECUTION]: The skill instructs the user or agent to employ external security tools to facilitate exploitation. Specifically, it references sstimap, tplmap, and tinja for automated vulnerability scanning and shell acquisition.
[DATA_EXFILTRATION]: Provides specific instructions and payloads for reading sensitive system files from the host environment. Evidence includes a Freemarker payload designed to resolve and read /etc/passwd via the java.net.URI and openStream() method chain.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection attack surface by consuming data from an external state management server.
Ingestion points: Data enters the context via the get_state_summary() function from the state MCP server (file: SKILL.md).
Boundary markers: No delimiters or instructions to ignore embedded commands are present when processing the state summary.
Capability inventory: The skill provides high-impact RCE payloads and utilizes external exploitation tools (file: SKILL.md).
Sanitization: No validation or sanitization logic is specified for the data retrieved from the state summary.
[REMOTE_CODE_EXECUTION]: Includes advanced evasion techniques to bypass Web Application Firewalls (WAFs) and security sandboxes. This includes character-by-character string construction in SpEL and obfuscation using Freemarker's lower_abc built-in.

ssti-freemarker