The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides complete Python script templates in references/blind-injection.md and references/union-and-error.md designed to automate data extraction. It explicitly instructs the agent to copy, modify, and execute these scripts locally using python3 to handle complex injection scenarios such as 32-character truncation in EXTRACTVALUE functions.
[COMMAND_EXECUTION]: The methodology directs the agent to use various command-line tools for security testing, including sqlmap for automated exploitation and curl for manual parameter integrity checks. The agent is encouraged to use these tools to interact with remote web servers.
[DATA_EXFILTRATION]: The skill's primary focus is the exfiltration of data from remote databases. It provides detailed procedures for identifying and extracting sensitive information like 'flags' and database schemas through UNION-based, error-based, and blind injection techniques, including out-of-band (OOB) methods using DNSLog services.
[REMOTE_CODE_EXECUTION]: The documentation includes payloads for achieving code execution on target database servers. Examples include using xp_cmdshell in MSSQL, DBMS_SCHEDULER in Oracle, and COPY FROM PROGRAM in PostgreSQL to execute shell commands and establish reverse shells.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because the automation scripts provided in the references process untrusted data retrieved directly from target HTTP responses.
Ingestion points: Data is ingested via the requests library in the Python scripts when parsing HTTP response bodies (e.g., using r.text and re.search for XPath errors).
Boundary markers: There are no boundary markers or instructions to isolate the data received from the target web server.
Capability inventory: The agent has the capability to execute Python scripts, run shell commands (sqlmap, curl), and perform network operations.
Sanitization: The scripts do not implement sanitization or validation of the data retrieved from the target before it is processed by the agent's logic.

sql-injection-methodology