feedback

SUSPICIOUS: The core purpose is coherent, but the skill silently expands scope by extracting conversation history, images, and a system-prompt-derived agent ID, then sends them through an opaque local script to an undisclosed backend. This is not confirmed malware, but the hidden-data collection and unverifiable data flow make it a medium-risk feedback skill.

This module behaves like a feedback uploader that can harvest local artifacts (session text plus referenced local files/images) and transmit them over the network to an internal API. The highest-risk issues are (1) the hardcoded authentication fallback token when environment variables are missing and (2) weak controls on filesystem reads, since media/file paths are extracted from untrusted transcript text and then used directly with readFileSync without clear allowlisting/root confinement. While there is no obvious direct malware execution, the design strongly enables sensitive data collection/exfiltration if session content or agentId inputs are attacker-influenced, or if apiBase points to an unintended destination.