The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides specific SQL payloads designed to exfiltrate database information to external domains such as 'attacker.com' and 'attacker-server.com' using DNS and HTTP protocols. It also includes instructions for extracting sensitive data such as usernames and passwords from the 'users' table.
[PRIVILEGE_ESCALATION]: The skill includes instructions for 'Authentication Bypass' using SQL injection techniques, which aim to gain unauthorized administrative access to a target application.
[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically construct and execute complex SQL queries for various injection techniques, including UNION-based, error-based, and time-based extraction.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external web applications (HTTP responses and database error logs), creating a surface for indirect prompt injection. A malicious target could potentially embed instructions within these responses to manipulate the agent's logic.
Ingestion points: Web application parameters, HTTP request/response logs.
Boundary markers: A mandatory confirmation gate is present for executing offensive actions, but no specific boundary markers are defined for the parsing of external content.
Capability inventory: Execution of SQLMap and dynamic generation of SQL injection payloads.
Sanitization: The skill does not describe mechanisms for sanitizing or filtering content retrieved from the target application servers.
[REMEDIATION_GUIDANCE]: Ensure all testing is performed only in isolated, non-production environments with explicit authorization. Implement strict output sanitization when the agent processes data from external web responses. Use structured data formats (like JSON) with strict schema validation for internal processing to minimize the impact of potentially malicious strings in target responses. Maintain the human-in-the-loop confirmation requirement for all offensive tool execution.

sql-injection-testing