The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation includes default credentials for the DVWA (Damn Vulnerable Web Application) lab target (admin / password). These are standard, publicly known defaults for this specific training application and do not constitute a leak of sensitive or private information.
[COMMAND_EXECUTION]: The skill describes the use of project-specific wrapper scripts (e.g., scripts/<tool>/<use-case>.sh) and make targets to execute pentesting tools and manage Docker-based lab infrastructure. It specifies an -x flag for command execution.
[INDIRECT_PROMPT_INJECTION]: The skill defines conventions for processing user-provided target notation (IP addresses, hostnames, and URLs).
Ingestion points: SKILL.md (Target Notation section) describes how the agent should interpret target strings.
Boundary markers: The Safety Rules section explicitly requires validation against a .pentest/scope.json file before execution.
Capability inventory: SKILL.md (Output Expectations section) references shell script execution via the -x flag.
Sanitization: Present; the skill mandates a manual allowlist process where targets must be added to .pentest/scope.json before any tools are run.

pentest-conventions