Claude agents communicate across open CLI sessions, consuming unexpected usage

**Summary**

Users on Reddit are discussing an issue with Anthropic's Claude where agents can communicate across multiple open CLI (command-line interface) sessions, allegedly triggering stale sessions and causing unexpected usage consumption. The framing suggests this is a recurring problem ("done it again"), implying prior similar issues.

**Dominant framing**

The discussion frames this as a bug or design flaw that catches users off-guard—specifically, that maintaining multiple open CLI instances creates unintended cross-session communication that depletes API usage quotas unexpectedly. The tone is cautionary, warning other users to avoid the practice.

**Notable context**

Engagement is minimal (2 Reddit mentions, no mainstream coverage), and the posts provide no technical details, source links, or official Anthropic statements. No users appear to be disputing the claim or offering alternative explanations. The "done it again" phrasing suggests community memory of prior incidents, but no specific prior issues are cited or linked in the visible posts. Without verification from Anthropic or detailed reproduction steps, the scope and severity of the issue remain unclear from the available discussion.