Stealth Agents Write Public Code

Leaked Claude Code shows a feature that can make 'stealth' (undercover) contributions to public repositories and an always‑on agent that monitors chat for 'frustration' words. That combination can alter who visibly authors code, evade attribution or review, and create opaque supply‑chain and moderation pathways. — This matters because it threatens open‑source provenance, increases software‑supply‑chain and security risks, and raises privacy and governance questions about platforms’ automated agents.