The Agent Skills Directory

[COMMAND_EXECUTION]: The utility scripts (code_security_scanner.py, dependency_auditor.py, security_scanner.py) execute external security CLI tools including bandit, npm, pip, and pip-audit via the subprocess.run module. These calls use list-based arguments and maintain a default of shell=False, which prevents command injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The dependency_auditor.py script includes a fallback mechanism that attempts to install the pip-audit package from the official Python Package Index (PyPI) if it is not present in the environment. Additionally, setup_pentest_env.sh is provided to install dependencies from requirements.txt during the initial environment setup.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external scan targets, creating a theoretical attack surface for indirect prompt injection.
Ingestion points: security_scanner.py fetches HTTP response headers and body content; code_security_scanner.py reads source code files; dependency_auditor.py ingests output from third-party audit tools.
Boundary markers: The skill does not currently implement explicit boundary delimiters or "ignore previous instructions" warnings when processing the data it scans.
Capability inventory: The orchestrator is configured with broad tool access including file manipulation (Read, Write, Edit, Glob) and command execution for security tooling (Bash(python3:*), Bash(pip:*), Bash(npm:*), Bash(bandit:*), Bash(pip-audit:*)).
Sanitization: The processing scripts utilize structured JSON parsing and regular expression filtering, which provides a layer of structural validation for the external content.

performing-penetration-testing