Claude Code hid timezone and proxy checks. what else can coding agents read

I used to let Claude Code break coursework and repo chores into little sub-agent threads. Not blind trust, but close enough that it felt like a normal dev tool. then the hidden tracker story landed and yeah, that comfort got weird. From the writeups and Thariq's post, Claude Code was checking timezone, proxy, and custom endpoint signals through obfuscated logic, then dropping a classification marker into the system prompt. Anti-abuse makes sense. Hiding the read is what bugs me. if a coding agent can quietl