v2.1.172
Claude CodeSummary
This release introduces significant hierarchical agent capabilities, allowing sub-agents to spawn nested sub-agents up to five levels deep. Numerous bug fixes address stability issues related to 1M context sessions, model selection across various interfaces, and background agent operations. Performance has also been improved, particularly in long conversations and reducing idle CPU consumption.
New Features
- Enabled sub-agents to spawn their own sub-agents, supporting up to 5 levels of nesting.
- Added a search bar when browsing plugins in the marketplace via the /plugin command.
- Amazon Bedrock now reads the AWS region from `~/.aws` config files if `AWS_REGION` is unset, aligning with AWS SDK precedence.
Bug Fixes
- Fixed sessions using 1M context without usage credits getting permanently stuck; sessions now automatically compact under the standard context limit.
- Fixed a repeating error message about an image being removed when the conversation contained multiple images.
- Fixed the agents view incorrectly keeping a session under 'Working' with a busy spinner for up to 30 seconds after the worker responded.
- Fixed background agents potentially reading another directory's project settings (approvals, trust) when dispatched to a pre-warmed worker.
- Fixed background-session attach failing with EAUTH for sessions started on an older version after the daemon auto-updated.
- Fixed a background sub-agent remaining stuck as 'active' in the agent panel after a nested agent it spawned was stopped.
- Fixed /model suggestions in the `claude agents` dispatch input rendering with an incorrect slash prefix and showing models disabled for the organization.
- Fixed `availableModels` restrictions not being applied correctly to subagent model overrides, the agent dispatch model picker, and the advisor model.
- Fixed `availableModels` allowlists incorrectly hiding Opus and Sonnet 1M rows in the /model picker when version-specific IDs were used.
- Fixed the /model picker on Bedrock offering models the provider does not serve, which previously caused silent model switching.
- Fixed model IDs receiving a doubled 1M-context suffix when `ANTHROPIC_DEFAULT_OPUS_MODEL` already included one.
- Fixed the `opusplan` model setting not correctly enabling 1M context in plan mode for entitled users.
- Fixed wildcard domain rules (`WebFetch(domain:*.example.com)`) and file permission rules with mid-pattern wildcards not matching correctly.
- Fixed the up-arrow prompt history showing the main agent's prompts while a subagent's chat tab was active.
- Fixed memory recall failing to find mounted team memory stores (`CLAUDE_MEMORY_STORES`) in remote sessions.
- Fixed workflow validation incorrectly rejecting scripts whose prompt strings or comments contained `Date.now()` or `Math.random()`.
- Fixed the /plugin marketplace list losing cursor position after navigating out of a long list, and Esc from the plugin browser returning to the wrong tab.
- Fixed PowerShell tool calls rendering as raw JSON instead of a command display and permission dialog in VSCode, and stripped ANSI escape codes from displayed shell output.
Improvements
- Added the `model` attribute to the `claude_code.lines_of_code.count` OTEL metric.
- Disabled mouse tracking on Windows consoles that do not fully support it.
- Improved performance in long conversations by removing redundant message normalization and avoiding full history transforms when streaming tool-use state is unchanged.
- Reduced idle CPU usage by preventing the /goal status chip from re-rendering the terminal at 5 Hz while idle, and reducing UI re-renders when subagents run in parallel.
- Improved Claude in Chrome tool loading by batching browser tool calls into a single request.
- Improved the non-interactive Usage Policy refusal message to suggest starting a new session or changing the model.
- The /code-review command now keeps the `ultra` option visible when not signed into claude.ai, with an explanation that cloud review requires an account.
- Shortened the Remote Control footer indicator to "/rc active" and hid it on narrow terminals.
- Stopped promoting /loop in remote sessions, as pending loops do not keep the container alive.