Claude Code Digest — 2026-03-19 23:24:57
Version updates
2.1.80 delivers crucial quality-of-life improvements for local workflows. Anthropic added the rate_limits field to statusline scripts. Developers can now track their 5-hour and 7-day usage limits directly in their terminal prompts. This eliminates the persistent guessing game around used_percentage and resets_at timelines, allowing you to pace intensive refactoring sessions.
The update also adds source: 'settings' as a plugin marketplace source. You can now declare plugin entries inline directly within settings.json. This removes the friction of maintaining external registries for local, one-off MCP scripts.
What the docs reveal
Anthropic fundamentally changed the operational model of Claude Code today. The assistant transitioned from a passive, synchronous CLI tool into an asynchronous, event-driven daemon.
The era of event-driven agents
The new Channels feature allows external MCP servers to push events directly into a running Claude session. A failed CI build, a Datadog alert, or an incoming Slack message can now wake your terminal agent. Claude can parse the incoming context, formulate a fix, and ping you on your phone via Telegram. By exposing an explicit reply tool, Anthropic enables two-way interaction. Developers can approve terminal commands remotely while away from their keyboards.
Anthropic simultaneously released Scheduled Tasks to handle basic polling. You can instruct Claude to run prompts at fixed intervals to monitor deployment statuses. However, the documentation explicitly recommends Channels as the superior architecture. Polling burns context window limits and API credits; pushing payloads via Channels triggers computation only when necessary.
Massive guardrails against prompt injection
Opening a local AI agent with filesystem access to external internet payloads presents an extreme security risk. A malicious Telegram message could theoretically trick Claude into executing rm -rf on your workspace. Anthropic understands this threat and engineered strict security boundaries.
Channels require multi-layered authorization. Team and Enterprise administrators must globally enable the channelsEnabled setting. Individual users must explicitly pass the --channels startup flag. Furthermore, developing custom listener plugins requires the sharply named --dangerously-load-development-channels flag. Anthropic enforces a strict sender-allowlist pairing process to verify remote users. Never bypass these verification constraints; failing to authenticate external inputs leaves your local machine vulnerable to remote execution.
Visualizing agent swarms
The introduction of the --teammate-mode flag reveals Anthropic's direction for complex workflows. With display options including tmux, in-process, and auto, Claude Code prepares the terminal UI to handle multiple specialized subagents running concurrently. Claude no longer executes as a singular thread. It orchestrates a swarm.