The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides comprehensive instructions and payloads for exfiltrating sensitive data from databases using Out-of-Band (OOB) techniques. It details how to use DNS (UTL_INADDR), HTTP (UTL_HTTP), SMB (UNC paths), and SMTP (UTL_SMTP) for exfiltration. Evidence is found in Section 5 of SKILL.md and Sections 1 and 5 of SCENARIOS.md.
[COMMAND_EXECUTION]: The documentation includes techniques for escalating database access to full operating system command execution using features such as MSSQL's xp_cmdshell, MySQL User Defined Functions (UDF), and Oracle's Java stored procedures. Evidence is located in SKILL.md Section 6 and SQLMAP_ADVANCED.md Section 1.6.
[PERSISTENCE]: The skill provides functional payloads for maintaining access to a host by injecting reverse shell commands into system crontabs via SQLite's ATTACH DATABASE command. Evidence is found in SCENARIOS.md Sections 11 and 17.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external SQL databases, which creates a surface for tool output poisoning. * Ingestion points: The agent is instructed to read and process output from various database query results. * Boundary markers: There are no instructions to use delimiters or ignore embedded commands in database outputs. * Capability inventory: The skill documents high-impact capabilities including shell execution, network exfiltration, and file writing. * Sanitization: The skill does not discuss or implement sanitization for the retrieved database content.
[REMOTE_CODE_EXECUTION]: The playbook references advanced automated tools like SQLMap for acquiring remote shells, performing binary file I/O, and writing webshells to the target server's filesystem. Evidence in SQLMAP_ADVANCED.md Section 4.

sqli-sql-injection