The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/generate_report.py file contains a generate_script method that takes arbitrary text input and writes it to a .sh file. The script then calls file_path.chmod(0o755) to make the generated file executable. This allows the skill to dynamically create and prepare shell scripts for execution based on runtime input.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted data from external codebases and infrastructure configurations (e.g., AWS Terraform or CloudFormation files).
Ingestion points: The skill identifies and reads source code repositories and configuration files in the target environment as part of its 'Analyze the Target' step.
Boundary markers: The provided scripts lack delimiters or instructions to treat findings from external files as untrusted content.
Capability inventory: The skill has the ability to write files and execute Bash commands via the Bash(security:*), Bash(scan:*), and Bash(audit:*) tools, and it includes a Python utility for generating executable shell scripts.
Sanitization: There is no evidence of sanitization or escaping of the content read from external files before it is used in the generate_report.py script to generate output files.
[DATA_EXFILTRATION]: The skill instructions and examples explicitly direct the agent to access sensitive infrastructure components, including AWS configuration files and the source code of payment applications. While necessary for a PCI-DSS audit, this workflow involves the systematic reading of highly sensitive data which could be exposed if the skill's reporting mechanism is compromised.

validating-pci-dss-compliance