The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted source code and dependency manifests which presents an indirect injection surface.
Ingestion points: scripts/vulnerability_scanner.py and scripts/dependency_auditor.py read files from the target project directory. scripts/pentest_report_generator.py ingests user-supplied findings JSON files.
Boundary markers: No specific boundary markers or instructions are provided to separate the scanned code content from the agent's control logic.
Capability inventory: The skill facilitates file reading, regex-based analysis, and report generation.
Sanitization: Input is processed through regex filters or JSON parsing, but the agent's view of this content could be influenced by instructions hidden in comments or strings within the scanned files. This is considered low severity as it is inherent to the scanning purpose.
[OBFUSCATION]: The references/attack_patterns.md file contains several examples of obfuscated URLs and payloads intended for testing SSRF bypasses.
Evidence: The documentation provides IP addresses encoded in hex (0x7f000001), decimal (2130706433), and octal (0177.0.0.1).
Context: These are legitimate educational examples for security professionals to verify if a target application correctly handles encoded internal addresses. They do not represent an attempt to hide malicious intent within the skill itself.
[COMMAND_EXECUTION]: The SKILL.md file provides CLI examples for interacting with the included Python scripts. These are standard tool usage patterns and do not involve unauthorized shell access or privileged commands.

security-pen-testing