Anthropic publishes reference implementation for agent automations
Anthropic's Claude developer blog published a walkthrough of a reference implementation built on Claude Managed Agents (beta) that reads custom sources such as Slack channels and GitHub repos on a schedule, tracks what changed since the last run, and posts a brief to Slack. The post focuses on common failure modes — treating a failed source read as "nothing new," using fixed time windows that create gaps or repeats, and recording a post before it is confirmed — and describes agent.md rules, per-source bookmarks, memory stores, vault credentials, and per-run spending caps to address them. It also provides a /claude-api managed-agents-onboard command to configure the agent interactively in Claude Code.
It turns the engineering details of making scheduled agents reliable — read confirmation, bookmarks, memory and scoped permissions — into a reusable reference implementation.