Suspected Message Dropping Bug in Claude Code
BFJ-Concerns · reddit · 2026-07-18
This Reddit thread discusses a potential issue where messages in Anthropic's Claude Code are silently dropped server-side.
- The author argues that what appears as "Fable/Claude being quiet" is actually text disappearing before reaching the client.
- Using local session files and scripts, they analyzed data across two machines: 6,166 out of 23,295 messages dropped text (26%). Scanning 73,000 messages across Opus 4.8, Opus 4.7, Sonnet 5, and Haiku 4.5 with the same method yielded zero hits.
- The trigger is consistent: text is most likely deleted when the model thinks, outputs text, thinks again, and then calls a tool. Text immediately adjacent to tool calls or at the end of a turn is usually safe.
- The thread explains this text is erroneously routed through a "protected/encrypted" channel; the model thinks it sent it, but the client never receives it, and it may be permanently lost after context compression.
- The author notes the earliest and most active related issues have been open for a long time with no maintainer response, urging others to run the script to see if the drop rate universally sits around 25%.
More from coding & agent
- LangChain ships Managed Deep Agents 0.9, walkthrough from Interrupt London keynote — LangChain · 2026-10-26
- Knowledge Worker Kit adds a full section on organizing projects — dSebastien · 2026-10-26
- Spring AI 2.1.0-RC1 adds tool-call confirmation, pausing agent loops before real-data access — therealdanvega · 2026-10-26
- METR red-teams OpenAI's agent monitoring, finds bypasses and gaps — thione · 2026-10-26
- OpenAI: coding agents weakened safety monitoring, one lied about it — thione · 2026-10-26
- ConvFill: A 0.6B Talker Buys Time While a Large Reasoner Whispers the Real Answer — andimarafioti · 2026-10-26