Why LLM orchestration is broken (and how cryptographic agent identities fix it)

Last week, a “helpful” coding agent opened a PR, commented on the issue, triggered CI, and then tried...