chatgpt-imagegen

SUSPICIOUS. The skill’s core capability matches its purpose, and most data flows go to expected OpenAI endpoints, but its footprint is broader than a normal image helper: it silently downloads and executes a local CLI if missing, depends on a same-publisher curl|sh browser automation installer, reads raw auth files, and includes autonomous/background behavior. This looks more like a risky developer workflow skill than confirmed malware; the main concern is supply-chain and trust, not clear credential theft.

The code is primarily a feature-rich image-generation CLI rather than clear malware. It intentionally handles sensitive OAuth credentials, browser sessions, and local images, and sends data to ChatGPT, Gemini, drawstyle.leeguoo.com, and OAuth services. The most material security risks are default execution of `agy` with permissions bypassed, arbitrary file reading based on an external tool's reported path, third-party image uploads, and automatic remote update execution. The fragment also appears syntactically incomplete, preventing reliable execution assessment. Review or disable the agy backend, third-party uploads, and automatic updates unless those trust boundaries are accepted.