The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is centered around executing complex shell commands via the sqlmap utility. It provides instructions for obtaining interactive operating system shells (--os-shell) and executing arbitrary commands (--os-cmd) on target systems, which represents a high-risk capability if misdirected or abused.
[DATA_EXFILTRATION]: Several commands demonstrate the extraction of sensitive information. This includes dumping entire database tables (--dump-all), extracting password hashes (--passwords), and reading sensitive system files from the target server (e.g., --file-read="/etc/passwd").
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection (Category 8). It ingests untrusted data from multiple external sources, including target URLs, Burp Suite request files (-r), and bulk target lists (-m).
Ingestion points: Untrusted data enters the agent context through the --url parameter and local file reads of request/log files specified in the documentation.
Boundary markers: The instructions lack delimiters or specific warnings to ignore embedded instructions within the data being processed by SQLMap.
Capability inventory: The skill allows for network operations, local file reading/writing, and shell command execution.
Sanitization: There is no mention of sanitizing or validating the content of external request files or target responses before the agent processes them.
[CREDENTIALS_UNSAFE]: The workflow relies on processing HTTP request files (-r) and log files (-l) from tools like Burp Suite. these files typically contain sensitive session tokens, cookies, and authentication headers, which the agent must read from the local filesystem to execute the commands.

sqlmap-database-pentesting