The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill provides complex bash one-liners and a standalone shell script (recon.sh) that execute multiple third-party security tools (Amass, Subfinder, Nuclei, ffuf, etc.). This gives the agent direct control over the underlying system's shell.
DATA_EXFILTRATION (LOW): While intended for reconnaissance, the workflow is designed to extract information about target infrastructure and potentially transmit data to external endpoints via tools like curl and various scanners.
EXTERNAL_DOWNLOADS (LOW): The skill fetches data from external, untrusted sources such as bgp.he.net, waybackurls, and gau. This content is directly processed by further tools in the pipeline.
Indirect Prompt Injection (LOW): The skill possesses a high vulnerability surface for indirect injection.
Ingestion points: Output from waybackurls, gau, and subfinder (found in sections 5, 7, and 10).
Boundary markers: Absent; data is handled as raw strings in shell pipes.
Capability inventory: bash subprocess execution, curl network requests, and nuclei vulnerability scanning.
Sanitization: None; the scripts pipe external data directly into powerful command-line tools without escaping or validation.

Red Team Tools and Methodology