osint-autopilot

SUSPICIOUS: the skill’s footprint matches its stated purpose as an offensive OSINT/recon autopilot, but that purpose itself gives the agent high-risk security-scanning capability. Main concerns are autonomous recon, multi-agent processing of untrusted external content, and unpinned third-party CLI installs; data flows are mostly consistent, with no clear credential theft or hidden exfiltration.

The fragment is an automated reconnaissance/endpoint-and-secret-harvesting workflow: it enumerates hosts and historical URLs (gau/waybackurls), fetches first-party HTML/JS, extracts sensitive-looking patterns into a “secrets” file, and actively scans public IPs with nmap across many ports. There is no direct evidence of a backdoor or destructive sabotage in the shown code, but the functionality is strongly aligned with offensive cyber operations and would be high risk if distributed as part of a software supply chain. Recommend treating this as a potentially malicious capability set and restricting/validating its execution context.

This module is not a passive utility; it orchestrates active host enumeration and content discovery (ffuf + curl + grep/regex) and is explicitly designed to hunt and return sensitive “secret-like” strings and high-interest exposure paths (e.g., admin panels, swagger/graphql/actuator, and files like .env/.git). It also writes evidence artifacts to a directory derived from user input and propagates potentially sensitive findings into a downstream synthesis prompt. While it claims “authorized red-team recon,” there is no in-module enforcement or robust input validation, and its use of agent-driven command instructions with untrusted parameters plus `curl -sk` increases risk. Treat as high security-risk capability for recon/credential harvesting if used outside tightly controlled environments.