Anthropic releases Claude Code v2.1.290 with expanded plugin hooks, managed agents improvements, and bug fixes

Anthropic released version 2.1.290 of Claude Code, a command-line AI coding assistant, with new plugin hook capabilities and bug fixes. The update adds serverToolUses to the result of a mod's turn.step hook, enabling plugins to inspect the tool calls the API executed, including each call's id, name, input, start, and end times.

Plugin hooks now receive an agentId field in the tool.check event, allowing hooks to differentiate between permission checks from a main session and those from a subagent. The release also introduces ThemeKey and Color types to the plugin hooks typings, enabling mods to reference the theme colors an editor can display.

Two new /claude-api managed-agents-onboard commands simplify setup for the Managed Agents pattern: one takes a URL to configure managed agents, and another builds a Console quickstart template (e.g., deep-researcher) using the ant CLI.

The changelog documents fixes for long sessions with hundreds of images getting stuck, requests failing behind certain proxies and gateways, and sessions compacting without triggering scheduled tasks. Additional fixes address WebFetch dropping page text beyond 100,000 characters, nested list parsing crashes, and headless --json-schema runs exiting non-zero after successful structured output delivery.

Other improvements include a Deny button on the sign-in approval page to terminate pending sign-ins quickly, support for partial session names in claude attach and claude logs commands, and warnings when managed settings files are symbolic links outside the managed settings folder.

Bug fixes target Windows-specific issues like multi-line shell blocks failing with CRLF line endings, Claude Code hanging when opening the /permissions tab in fullscreen search mode, and /ultrareview failing to upload uncommitted changes under specific git configurations.