paper-explain-figures

The code appears to implement a local figure-analysis orchestration tool, not malware. The main security concerns are broad inherited-environment exposure to external AI runners, automated execution with approval/permission checks disabled, processing of untrusted image files by external converters, incomplete secret redaction, and a potentially unsafe symlink-following cleanup routine. The fragment also contains apparent runtime errors from undefined names and omitted definitions. No explicit exfiltration or backdoor behavior is shown.

The fragment is a configuration for a local paper-figure analysis workflow, not direct malware. It contains notable security risks: Codex has workspace-write capability despite a read-only claim, and Claude runs with permission checks disabled. No concrete data exfiltration, credential theft, or destructive behavior is shown, but the omitted worker implementation should be reviewed before use, especially to enforce read-only filesystem access and constrain external process capabilities.