The Agent Skills Directory

[SAFE]: The skill functions as a security scanner. Analysis of scripts/scan_skill.py shows it utilizes regular expressions to scan local file contents for known vulnerability patterns. It does not perform any unauthorized network operations, credential harvesting, or destructive file modifications.
[COMMAND_EXECUTION]: The skill utilizes the standard system binaries jq and python3 as declared in the manifest. These tools are used for processing JSON data and executing the scanner logic, respectively, which is appropriate for its intended functionality.
[INDIRECT_PROMPT_INJECTION]: The scanner constitutes a surface for indirect prompt injection because it processes untrusted data from other skills. Ingestion points: The scan_skill.py script reads file contents from any directory provided via the --skill-dir argument. Boundary markers: The scanner does not wrap findings in specific delimiters to prevent the agent from following instructions contained in the scanned files. Capability inventory: The script's capabilities are limited to reading files and printing results; it does not have inherent write or network capabilities. Sanitization: The tool mitigates potential injection risks by truncating detected lines to 100 characters in its report, which helps prevent the execution of long or multi-line malicious instructions.

agentic-security-scanner