The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches security analysis tools from third-party GitHub repositories.
Downloads ysoserial-all.jar from github.com/frohoff/ysoserial.
Clones the phpggc project from github.com/ambionics/phpggc.
Clones the JNDI-Exploit-Kit from github.com/pimps/JNDI-Exploit-Kit.
[REMOTE_CODE_EXECUTION]: Executes third-party code downloaded from external sources during runtime.
Executes the downloaded ysoserial-all.jar binary using java -jar.
Executes scripts from the cloned phpggc repository using php.
[COMMAND_EXECUTION]: Performs extensive system operations using standard binaries to facilitate attack methodology.
Uses curl and wget for network operations and file retrieval.
Employs java, python3, php, dotnet, and git to generate payloads and execute security tools.
[DYNAMIC_EXECUTION]: Generates and executes code snippets at runtime for payload construction.
Uses python3 -c to dynamically assemble a Python pickle payload that includes system commands.
[INDIRECT_PROMPT_INJECTION]: Creates an attack surface by interpolating external variables directly into shell commands.
Ingestion points: Variables such as $TARGET and COLLAB_HOST are ingested from the environment or user input for use in SKILL.md.
Boundary markers: None identified; untrusted data is directly interpolated into command strings without delimiters.
Capability inventory: Significant command execution capabilities including curl, wget, java, python3, php, dotnet, and git.
Sanitization: No evidence of input validation or shell escaping for the interpolated external variables.

hunt-deserialization