Claude Code watches for oversights: a plugin adds an extra agent

Enabling the assistant. The command `/plugin enable cc-plugin-you-should-know@builtin` enables You should know. The plugin works in sessions connected directly to Anthropic with telemetry enabled.

Anthropic also added Claude Mods, plugins that change the agent's internal behavior. In September, the company offered an experimental launch option through `CLAUDE_CODE_ENABLE_FUNCTION_HOOKS=1 claude` and open-sourced three built-in mods. The demos included a plugin that hides secrets in tool output before it reaches the model.

Getting the update. For native and npm installations, use `claude update`; for installations through the corresponding Homebrew cask, use `brew upgrade claude-code@latest`. Check the installed version with `claude --version`. The stable channel usually trails by about a week.

In the agent list, the `n:<text>` filter now searches session names and tasks, including collapsed sections. Enter opens the first match. In VS Code, a running command or subagent can be sent to the background with the Run in background button.

Previously, Remote Control could stop receiving messages for several minutes after a failed reconnection. Claude Code now waits 30 seconds for a response and retries. Anthropic also fixed the loss of the previous conversation when a cloud session restarted during context compaction.

MCP servers using protocol 2025-11-25 can now request that a URL be opened, for example to sign in to an account. If a server stops connecting after the update, Anthropic recommends adding `"bareElicitationCapability": true` to its configuration.

OpenTelemetry now includes a `prompt_text` field that duplicates `prompt`. If your log collector removes or masks `prompt`, apply the same rule to `prompt_text`.

On Bedrock, Vertex, Foundry, and the Claude application gateway, Opus 4.7+ and Fable now use a 1 million-token context window by default.