2026-08-11 · Anthropic

v2.1.228

securityprotocolsmodels

read at source ↗ github.com

v2.1.228

Source: Anthropic Claude Code Date: 2026-08-11 URL: https://github.com/anthropics/claude-code/releases/tag/v2.1.228

Summary

Patch release fixing session-redraw hangs, Windows git-path detection, /tui model reversion, cross-session inbox/messaging bugs, self-hosted-runner checkout and session-cleanup failures, and a marketplace settings-merge bug. It also hardens claude.ai-synced skills so they can’t shadow local commands or MCP prompts, and their bodies can’t run ! shell commands or @ file expansions on the local machine.

Implications

Feeds dev-tooling-ecosystem thread — routine maintenance, not a capability shift. The skills-hardening item is the notable line: closing a prompt-injection-adjacent vector in synced skills is a small fail-closed move inside the tool itself, consistent with the broader early-August hardening cadence tracked elsewhere in the stack (fnox, aube, mise).

← all signals