Claude Code v2.1.224 Adds Self-Hosted Runners, Cross-Session Messaging
Anthropic released Claude Code v2.1.224, a feature-packed update that introduces self-hosted environments, cross-session messaging, and a new plugin source, along with a heap of fixes for sandbox and remote-control issues.
Key Additions
- Self-Hosted Runners: The new
claude self-hosted-runnercommand turns your own machines or containers into a place where Claude Code web, mobile, and desktop sessions can run. Available on Team and Enterprise plans. - Archive Plugin Source: You can now install plugins from a zip over HTTPS without needing git or npm. Optional SHA-256 pinning is supported for security.
- Cross-Session Messaging: Claude Code sessions can now message each other across machines (macOS and Linux) with
SendMessage, and you can useListAgentsto discover available sessions. - Bedrock Region Prefix: New
ANTHROPIC_BEDROCK_REGION_PREFIXenv var lets you prefer a specific cross-region inference profile over the one derived fromAWS_REGION.
Beefed-Up Sandbox Credential Masking
The sandbox now supports more credential-masking options:
extractandonExtractNoMatchfor structured env valuesdecode: "jwt"withmaskClaimsfor JWT-aware maskingawsPairs/sigv4for AWS SigV4 re-signing
These options require network.tlsTerminate and are honored only from user, managed, or --settings settings.
Notable Fixes
- Fixed long (>200 char) project paths resolving to another project's session directory – session list, rename, fork, delete, and
/resumeno longer cross projects. - Fixed sandbox filesystem deny entries with trailing slashes (e.g.,
denyRead: "~/.aws/") being bypassed on Linux and macOS. - Sandbox violation details now appear in Bash tool results, so Claude sees which file or network access was denied and why.
- Fixed
SendMessagereporting success when writes to a teammate's inbox actually failed. - Removed the 200-subagent-per-session spawn cap – long-running sessions no longer refuse new agents (concurrency and depth limits still apply).
- Remote Control now shows compaction progress and connection failures persistently instead of silent pauses or 8-second toasts.
Who It's For
If you're running Claude Code in production, on remote servers, or in a multi-session workflow, this release adds meaningful flexibility and plugs several security holes.
📖 Read the full source: GitHub Claude-Code
👀 See Also

Simple Self-Distillation Method Improves LLM Code Generation
Researchers show that fine-tuning LLMs on their own sampled outputs (simple self-distillation) improves code generation performance, boosting Qwen3-30B-Instruct from 42.4% to 55.3% pass@1 on LiveCodeBench v6.

Goldman Sachs Analysis Shows Minimal AI Impact on 2025 US GDP Growth
Goldman Sachs economists report AI investment contributed 'basically zero' to US GDP growth in 2025, citing imported hardware and unmeasured productivity impacts as key factors.

Stanford Study: Law Professors Prefer AI Answers Over Peers 75% of the Time
In a blind evaluation of 3,000 comparisons, law professors rated AI-generated answers significantly higher than peer-written ones. AI responses were flagged as harmful only 3.5% of the time vs 12% for humans.
New Orleans Tests Carbyne AI for 911 Call Triage — What It Means for Emergency Tech
New Orleans is testing Carbyne's AI Emergency Call Triage to handle repeated calls about the same incident, reducing dispatcher workload. The AI asks callers if they're calling about a known incident; yes routes to AI info, no transfers to a human.