The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on a wide variety of command-line tools for scanning, authentication testing, and exploitation, including nmap, nxc (NetExec), hydra, mssqlclient.py, sqsh, and wmic.
[PRIVILEGE_ESCALATION]: Instructions within the skill direct the agent to utilize sudo when running network-level tools like responder and impacket-smbserver, which require elevated privileges on the system running the agent.
[EXTERNAL_DOWNLOADS]: The skill contains a reference to a third-party GitHub repository (github.com/IamLeandrooooo/SQLServerLinkedServersPasswords/) for downloading specialized password extraction tools, which is not a verified or trusted vendor.
[REMOTE_CODE_EXECUTION]: The methodology describes techniques for executing remote code on target systems, notably using xp_cmdshell to download and execute PowerShell reverse shell payloads.
[PERSISTENCE]: The skill documents how to establish persistent access on target MSSQL servers by configuring startup stored procedures through the sp_procoption command.
[DATA_EXFILTRATION]: The skill outlines methods for extracting sensitive information, including SQL login hashes and capturing NetNTLMv2 hashes by forcing target systems to authenticate to an attacker-controlled SMB server using xp_dirtree.
[INDIRECT_PROMPT_INJECTION]: The skill ingests results from target databases during enumeration phases, which presents a surface where malicious database content could influence the agent's behavior.
Ingestion points: Results from SQL queries used for enumeration in SKILL.md and references/mssql-techniques.md.
Boundary markers: None identified in the instructions to separate data from instructions.
Capability inventory: Extensive subprocess calls for external tools, network operations, and file writing capabilities.
Sanitization: No explicit validation or escaping of data retrieved from the database is mentioned.

mssql-pentesting