Claude.ai silently failing since Jan 14, no official acknowledgment (github.com)

Since January 14, users of Claude.ai have been experiencing a significant issue with the auto-compaction feature, which is supposed to manage conversation context as it approaches the 200k token limit. Despite being flagged as fixed on January 15, the functionality remains broken, leading to messages either being silently returned to the input box or, occasionally, prompting a "limit reached" error. This has disrupted user experience, especially for those utilizing Claude in project settings where continuous conversation context is critical. The persistence of this bug is notable for the AI/ML community, as it highlights potential vulnerabilities in the management of context within complex AI models. Auto-compaction is crucial for maintaining seamless interactions in lengthy conversations, and its failure raises questions about the robustness of system updates and the testing protocols in place at Anthropic. Users are left without a dedicated bug tracker for such issues, reflecting a need for better support infrastructure to address technical problems promptly.