The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a comprehensive set of complex shell commands for the SQLMap tool. These commands are intended to be executed by the agent to automate the detection and exploitation of database vulnerabilities.
[DATA_EXFILTRATION]: The instructions describe methods for extracting large volumes of sensitive data. Evidence includes commands for dumping entire databases (--dump-all), extracting specific sensitive columns like username and password, and reading local system files from the target server such as /etc/passwd using the --file-read flag.
[REMOTE_CODE_EXECUTION]: The skill includes advanced exploitation techniques that allow for arbitrary code execution on target systems. Specifically, it provides commands for using --os-shell and --os-cmd, which grant the agent the ability to execute operating system commands on the remote web server.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted external data, creating a significant attack surface for indirect injection if the tool's output or the ingested data is manipulated.
Ingestion points: The agent is instructed to ingest data from a Target URL, Request File (Burp Suite captures), logfile, and bulkfile.txt.
Boundary markers: No delimiters or instructions to ignore embedded commands within the processed data are present.
Capability inventory: The agent has the capability to execute shell commands, perform network operations, and read/write files via the SQLMap tool.
Sanitization: There is no evidence of input validation, escaping, or sanitization for the URLs or file paths provided as input.
[EXTERNAL_DOWNLOADS]: The skill suggests downloading the SQLMap tool from GitHub if it is not already present on the system. While GitHub is a well-known service, this introduces a dependency on external code that may be executed by the agent.

SQLMap Database Penetration Testing