Message your other Claude Code sessions

The strongest reaction was not surprise that this exists. It was how many people had already built their own version. The practical reason kept coming up: persistent specialist sessions are a cheap way to avoid re-paying the token cost of large context or specialized setup. One agent can absorb the expensive background once, then other sessions can query it when needed. Several people also argued that this beats brute-force prompt compaction. Instead of squeezing an entire chat into a brittle summary, keep transcripts and artifacts searchable, or keep subagents alive and resumable, so context can be retrieved on demand instead of guessed in advance. Where people got sharper was on failure modes. Manual handoff documents were called a trap because agents write them with too much confidence, then future agents treat them as ground truth and inherit bad assumptions. Two-way messaging and searchable logs were seen as more grounded because they preserve the ability to ask follow-up questions and inspect original context. Security-minded commenters also pointed out the obvious tradeoff: once agents can call across session boundaries and trigger hooks, you have created another remote-execution path, even if the product presents it as convenience. The net view was enthusiastic but very unsentimental. This is useful infrastructure for serious agent workflows, but it pushes users closer to operating an ad hoc distributed system with all the usual problems of coordination, observability, and trust boundaries.