Claude Code v2.1.229: Remote Control, Plugin Marketplaces, and Critical Fixes

Claude Code v2.1.229 (released Aug 12) brings remote session improvements, plugin marketplace support, and a batch of stability fixes for streaming, Windows paths, and MCP OAuth.
Remote Control & Plugin Marketplaces
claude remote-control --continueis now documented to resume the most recent Remote Control session.- Self-hosted runner sessions now support server-supplied Claude Code hooks, matching managed-environment behavior.
- SSE keepalive pings were added to gateway streaming responses during long thinking pauses, preventing idle-timeout disconnects on Vertex and Bedrock upstreams.
- Plugin marketplace
commandsources let a local command (e.g., an IDE) print the plugin directory; it's re-resolved each session and applied without restart. Usemode: "link"to use it in place. ListAgentsnow marks disconnected Remote Control sessions asofflineand labels cloud sessions ascloud.
Key Fixes
- Fixed long responses disappearing while streaming and being printed twice in the terminal.
- Fixed a crash (including on
--resume) when a tool call had a non-stringglob,file_path, orcommandvalue. - Fixed a RangeError crash when progress bars or markdown tables rendered in very narrow terminals (also affected
--continue/--resume). - Fixed a crash on Windows when referencing extended-length (
\\?) or UNC paths. - Fixed auto mode failing for users who disable the attribution header via
CLAUDE_CODE_ATTRIBUTION_HEADER(direct Anthropic API connections). - Fixed
/modelrejecting Sonnet/Opus 1M for claude.ai subscribers using a customANTHROPIC_BASE_URLgateway. - Fixed MCP OAuth with strict authorization servers by using
127.0.0.1instead oflocalhostin the redirect URI. - Fixed Remote Control clients showing a stuck spinner after a slash command typed in the laptop terminal.
- Fixed the Claude Code Review workflow (from
/install-github-app) completing without posting its review. - Fixed multi-second UI stalls after editing a file with thousands of IDE diagnostics while the extension is connected.
- Fixed one-shot
claude plugincommands leaving a stray liveness file that prevented cleanup of outdated plugin versions. - Fixed dynamic workflows inside CPU-limited containers using the host's core count instead of the container's CPU limit.
- Fixed a file-watcher handle leak after atomic file replacements, and an uncaught error on Windows when the scheduled-tasks watcher failed on network/virtual filesystems.
- Fixed SDK and
--input-format stream-jsonsessions getting a 400 error on whitespace-only messages. - Fixed conversations exceeding the 32 MB request limit retrying compaction when no images/documents can be stripped; they now fail once with a clear message.
- Fixed OpenTelemetry export from Claude Desktop being rejected by the Desktop-managed gateway.
- Fixed self-hosted runner/remote sessions exiting at startup when
managed-mcp.jsonis deployed; servers are now skipped with a warning. - Fixed self-hosted runner repository preparation hanging on Git Credential Manager prompts; git now fails fast when credentials are missing.
Improvements & Changes
- Workflow fan-outs now stagger same-prefix sibling agents so subsequent agents reuse the cached prompt prefix instead of re-paying it. Disable with
CLAUDE_CODE_WORKFLOW_PREFIX_STAGGER_MS=0. - Improved "prompt is too long" errors to explain why automatic compaction could not recover, instead of just suggesting
/compact. - Sandbox: IPv6 literals in network domain lists are now bracketed (e.g.,
[::1]:443), and ambiguous spellings are enforced fail-closed and flagged by/doctor. /loginnow repeats theCLAUDE_CODE_OAUTH_TOKENoverride warning after successful login./commit-push-prno longer auto-approves git/gh commands with dangerous flags (--force,--amend,--no-verify).- Self-hosted runner on Windows now requires an explicit
--base-dir; there is no default checkout directory. - VSCode: "Report a problem" and
/bugopen the built-in feedback dialog instead of a retired survey link. - VSCode: the
/btwside-question panel is now resizable by dragging its boundary. - VSCode: added session groups in the sidebar.
This release is particularly relevant for teams running Claude Code in remote/self-hosted environments, those using plugin marketplaces, and anyone hitting streaming or Windows-specific crashes. Full details on GitHub.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Werld: Open-Ended Artificial Life Simulation with Evolving Neural Networks
Werld is a real-time artificial life simulation where agents with NEAT neural networks evolve their own neural architecture, sensory processing, and behaviors without hardcoded rules or reward functions. The simulation starts with 30 agents on a Watts-Strogatz small-world graph with 64 sensory channels, 7 continuous motor functions, and 29 heritable genome traits.

Introducing NetViews 2.3: A Robust Network Diagnostic Tool for macOS
NetViews 2.3 combines host discovery, Wi-Fi insights, and real-time monitoring with a streamlined GUI for better network diagnostics on macOS.

Krasis LLM Runtime Shows 8.9x Prefill and 4.7x Decode Speed Improvements Over Llama.cpp
Krasis LLM runtime now runs both prefill and decode entirely on GPU with different optimization strategies, achieving 8.9x faster prefill and 4.7x faster decode than llama.cpp on Qwen3.5-122B with a single 5090 GPU.

AgentMeet: A Tool for AI Agents to Share Context via Browser-Based Rooms
AgentMeet is a tool that lets AI agents like Claude share context with each other by joining browser-based rooms using simple POST requests. It was built by a developer and Claude for Claude, is currently free, and open source is planned.