The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted HTTP request files, which creates a vulnerability surface where malicious instructions could be embedded in data processed by the agent.
Ingestion points: The skill instructs the agent to load HTTP requests from external files such as request.txt and profile_request.txt using the -r and --second-req flags (found in SKILL.md).
Boundary markers: There are no instructions or delimiters provided to the agent to ensure it ignores any potentially malicious instructions hidden within these external request captures.
Capability inventory: The skill enables high-impact capabilities via sqlmap, including interactive OS shells (--os-shell), command execution (--os-cmd), and file system manipulation (--file-read, --file-write).
Sanitization: No sanitization or validation steps are defined for the content of the ingested request files before they are passed to the tool.
[COMMAND_EXECUTION]: The skill provides templates for executing powerful commands that can interact directly with the operating system and file system of a target server.
The --os-shell and --os-cmd flags allow the tool to execute shell commands on the remote database server.
The --file-read and --file-write flags enable reading sensitive files (e.g., /etc/passwd) or writing potentially malicious files (e.g., webshells) to the target's web root.
[DYNAMIC_EXECUTION]: The skill documents the use of the --eval flag, which allows for the execution of arbitrary Python code at runtime.
An example in SKILL.md demonstrates using --eval to import the hashlib module and execute code to compute a hash for a parameter on every request.

sqlmap