Anthropic News | Claude Code v2.1 /config and Python SDK Updates
The Clauder Navi editorial team brings you the latest news surrounding Anthropic on June 19, 2026, delivered in a news program style. Today's top highlights are the new /config syntax introduced in Claude Code v2.1.181 and the code_execution tool support added in Python SDK v0.110.0. We also cover the multi-agent research system construction methodology published on the Anthropic Engineering Blog.
Claude Code v2.1.181 introduces a new /config key=value syntax, enabling real-time switching of configuration values directly from the command line. Automatic retry on API disconnection has also been strengthened, improving reliability for long-running processing tasks. This is a high-impact update for engineers looking to shorten their trial-and-error cycles.
In the Python SDK, support for the code_execution_20260120 tool was added in v0.111.0 / v0.110.0, making it possible to implement API calls to the code execution tool. A bug affecting stream event type preservation for Bedrock users was also fixed at the same time, and SDK users are encouraged to update. A tagging feature for refusal-fallback middleware has also been newly introduced.
The Anthropic Engineering Blog published details on a research system where a lead researcher agent formulates plans and multiple sub-agents conduct parallel exploration. The figures show a 90.2% performance improvement over a single-agent setup, providing development teams considering multi-agent implementations with practical design guidelines.
目次 (5)
[1] Claude Code v2.1.181 — New /config Syntax and Connection Stability
Claude Code v2.1.181 was released on June 17, 2026 at 22:07 UTC (source: https://github.com/anthropics/claude-code/releases/tag/v2.1.181 ). Impact level for engineer readers: ★★★.
The biggest change is the addition of /config key=value syntax. Intuitive commands like /config thinking=false allow you to switch runtime settings on the fly without editing configuration files, enabling easier experimentation. Automatic retry on API disconnection has also been implemented, and long paragraphs in streaming are now displayed line by line. For macOS, an opt-in setting for sandbox.allowAppleEvents and a CLAUDE_CLIENT_PRESENCE_FILE environment variable have also been added. This is a highly impactful update for engineers who use Claude Code on a daily basis.
[2] Python SDK v0.111.0 / v0.110.0 — code_execution Tool Support
Python SDK v0.110.0 (17:18 UTC) and v0.111.0 (17:31 UTC) were released on June 18, 2026 (source: https://github.com/anthropics/anthropic-sdk-python/releases/tag/v0.110.0 ). Impact level for engineer readers: ★★★.
v0.110.0 adds support for the new code_execution_20260120 tool. Teams implementing API calls to code execution tools can access this feature by updating the SDK. The subsequent v0.111.0 introduces a tagging feature for refusal-fallback middleware. A bug affecting stream event type preservation for Bedrock users has also been fixed, improving the reliability of stream processing.
[3] Anthropic Engineering Blog — Multi-Agent Research System Construction Methodology
On June 13, 2026, the Anthropic Engineering Blog published a detailed guide on building multi-agent research systems (source: https://www.anthropic.com/engineering/multi-agent-research-system ). Impact level for engineer readers: ★★.
The article explains an architecture where a lead researcher agent formulates plans and multiple sub-agents conduct parallel exploration. A 90.2% performance improvement over a single-agent setup was achieved, and the production environment adopts a rainbow deployment approach designed to avoid interrupting running agents. Practical decision-making criteria are presented — including small-scale testing first, automated evaluation using LLM judges, and production deployment policies — making this a valuable reference for teams implementing multi-agent systems in-house.
What to Watch Next
The /config syntax in Claude Code is expected to support an expanding set of configurable options going forward. The code_execution tool in the Python SDK serves as a gateway for implementing code execution calls via the API, so demand for practical articles combining it with specific use cases will likely grow. The Anthropic multi-agent research system is a valuable real-world example that teams considering automated research task pipelines for enterprise use can reference as a design guide. By continuing to track the latest versions of SDK releases, you can safely incorporate new features and known bug fixes.
Sources
- GitHub — Claude Code v2.1.181 Release Notes: https://github.com/anthropics/claude-code/releases/tag/v2.1.181
- GitHub — anthropic-sdk-python v0.110.0: https://github.com/anthropics/anthropic-sdk-python/releases/tag/v0.110.0
- Anthropic Engineering Blog — Multi-Agent Research System: https://www.anthropic.com/engineering/multi-agent-research-system
