--dangerously-skip-permissions in CI: when AI agent YOLO flags leak your secrets

An issue opens, a workflow starts, and an AI agent reads it: run: | claude -p "Fix this issue:...