Anthropic released memory for Claude Managed Agents into public beta on the Claude Platform, enabling agents to retain and build on knowledge across separate sessions.
Anthropic published a post-mortem identifying three issues behind a month-long Claude Code quality slip, confirming all are resolved in v2.1.116 and that usage limits have been reset for all subscribers.
An Anthropic team member shared a post-mortem examining the root causes behind recent Claude Code issues, offering transparency into what went wrong and likely next steps.
Anthropic confirmed the recent quality issues originated in Claude Code and the Agent SDK harness — including Cowork, which runs on the SDK — while the underlying models and Claude API remained unaffected.
Anthropic launched a /schedule command allowing users to queue one-time tasks from the CLI or the Routines UI's Once Trigger, with example uses including delayed feature flag cleanup and post-launch reporting.
Anthropic clarified that one-time tasks created via the new scheduling feature draw from users' normal rate limits, directing users to documentation for further details.
Anthropic team member noahzweben confirmed that the /schedule command can be used to set up recurring loops, expanding its utility beyond one-time scheduling.
Anthropic confirmed that memories in Claude Managed Agents are portable, allowing developers to export them as files or manage them directly through the API for full programmatic control.
Anthropic detailed the architecture behind Managed Agents memory: stores are scoped per workspace, saved as text files, and mounted inside the container when attached to a session, giving Claude read and write access.
Anthropic outlined prevention measures following the Claude Code regression, including internal dogfooding with user-matching configs and a broader eval suite run against isolated system prompt changes.
Anthropic's bcherny described the recent Claude Code quality investigation as the most complex the team has undertaken, citing non-obvious root causes and numerous confounding factors.
Anthropic confirmed separate reports of Opus 4.7 problems inside Claude Code, saying the team is actively working on improvements to be rolled out over the coming days.
