Claude Code v2.1.47 Released: 40+ Bug Fixes, Windows Overhaul, Memory Improvements, and Faster Startup
Anthropic ships Claude Code v2.1.47 with over 40 bug fixes targeting Windows rendering bugs, long-session memory leaks, a ~500ms startup improvement, plan mode preservation after compaction, and a revamped background agent kill shortcut.
Anthropic published Claude Code v2.1.47 to npm on February 18, 2026 (UTC). The release is a broad maintenance update containing more than 40 bug fixes alongside several performance and usability improvements. No new headline features are introduced; the focus is squarely on reliability.
Windows Reliability Overhaul
Windows users receive the densest cluster of fixes in this release.
Terminal rendering was broken by os.EOL (\r\n) line endings leaking into display code, causing line-count displays to show 1 instead of the correct value. Bold and colored Markdown text was also shifting to wrong characters on Windows for the same reason. Both are now fixed.
Bash tool output was being silently discarded for users running MSYS2 or Cygwin shells. A separate fix stops CWD-tracking temp files from accumulating indefinitely (issue #17600). Image pasting now works on WSL2 systems where Windows copies images as BMP format (#25935). Worktree session matching now handles mixed drive-letter casing correctly.
Memory Improvements for Long Sessions
Three independent memory leaks in long-running agent sessions are addressed:
- API stream buffers, agent context, and skill state are now released after use.
- Agent task message history is trimmed after tasks complete.
- An O(n²) message accumulation in agent progress updates is eliminated.
Together these changes meaningfully reduce the memory footprint of multi-hour coding sessions.
Startup Performance
SessionStart hook execution is now deferred, reducing time-to-interactive by approximately 500 milliseconds.
VS Code Plan Preview Improvements
The plan preview panel in the VS Code extension now auto-updates as Claude iterates on a plan. Commenting is enabled only when the plan is ready for review, and rejecting a plan keeps the preview open so Claude can revise rather than closing it.
Background Agent Control Change
The shortcut to kill all background agents has changed. Previously, double-pressing ESC terminated them. Now, ctrl+f kills all background agents while a single ESC cancels only the main thread, leaving background agents running. This gives finer control over agent lifecycle during complex multi-agent sessions.
Session Management
Several session-resume issues are resolved:
/resumeno longer fails to find sessions whose first message exceeds 16KB (#25920, #25721).- Custom session titles set with
/renameare no longer lost after resuming a conversation (#23610). - Session names set via
/renamenow survive context compaction (#26121). - The session resume picker now shows 50 sessions on first load, up from 10 (#26123).
Context Compaction Fixes
Compaction was failing when a conversation contained many PDF documents; document blocks are now stripped alongside images before sending to the compaction API (#26188). A separate fix prevents plan mode from being lost after compaction, which was silently causing the model to switch from planning to implementation mode (#26061).
Agent and Skill Fixes
- User-defined agents now load all files on NFS/FUSE filesystems that report zero inodes, not just the first one (#26044).
- Plugin agent skills referenced by bare name (rather than fully-qualified plugin name) no longer silently fail to load (#25834).
- Project-level
.claude/agents/and.claude/skills/directories are now discovered when running from a git worktree. - The custom
modelfield in.claude/agents/*.mdis now respected when spawning team teammates (#26064). - Concurrent agent sessions no longer trigger API 400 errors (“thinking blocks cannot be modified”) caused by interleaved streaming content blocks.
Other Notable Fixes and Additions
@ file mentions: The file suggestion index is pre-warmed on startup and cached per session with background refresh, making file suggestions appear noticeably faster.
Bash permission classifier: A fix prevents the classifier from generating hallucinated match descriptions that don’t correspond to actual input rules — a subtle but security-relevant correctness bug.
Hook inputs: last_assistant_message is added to Stop and SubagentStop hook inputs, allowing hooks to access the final assistant response without parsing transcript files.
New keybinding: chat:newline is added as a configurable action for inserting literal newlines in multi-line input (#26075).
Statusline JSON: added_dirs is now exposed in the workspace section, making directories added via /add-dir visible to external scripts (#26096).
CJK layout: Wide CJK characters no longer cause misaligned timestamps and layout elements in the TUI (#26084).
Thinking mode: alwaysThinkingEnabled: true in settings.json now correctly activates thinking mode on Bedrock and Vertex providers (#26074).
MCP: MCP servers now appear in the MCP Management Dialog after deferred loading.
macOS FSEvents loop: Read-only git commands no longer trigger file watcher loops on macOS; --no-optional-locks is now passed (#25750).
Background agent results: Background agents now return the agent’s final answer instead of raw transcript data (#26012).
Upgrading
npm install -g @anthropic-ai/claude-code@2.1.47
Or from within Claude Code:
claude update
v2.1.47 was published to the npm registry at 20:15 UTC on February 18, 2026. No GitHub release tag has been created as of this writing. Full changelog is available in the CHANGELOG.md file.
Related Articles
Claude Code v2.1.49 Released: Background Agent Kill Fix, Startup Speedups, and Yoga WASM Memory Leaks Squashed
Anthropic ships Claude Code v2.1.49 with roughly 20 changes, including a fix for Ctrl+C and ESC being silently ignored during background agent execution, multiple startup performance improvements via MCP batching and auth-failure caching, an end to Yoga WASM memory growth in long sessions, file editing in simple mode, and new SDK capability fields.
Claude Code v2.1.50: Worktree Hooks, Agent Memory Overhaul, Opus 4.6 Gets 1M Context
Anthropic ships Claude Code v2.1.50 with WorktreeCreate/WorktreeRemove hook events, declarative worktree isolation in agent definitions, a fix for session data loss on SSH disconnect, a native module compatibility fix for older glibc Linux systems, an avalanche of memory leak fixes targeting long-running sessions, and Opus 4.6 fast mode now supporting the full 1M context window.
Claude Sonnet 4.6 Becomes the Default for Free and Pro Users — Outperforms Opus 4.5 on Coding Agent Benchmarks
Anthropic has made Claude Sonnet 4.6 the default model for claude.ai's Free and Pro plans. Released February 17, 2026, it matches Sonnet 4.5 pricing at $3/$15 per million tokens while internal Claude Code evaluations show it beating the previous frontier model, Opus 4.5, 59% of the time on agentic coding tasks.
Popular Articles
868 Agentic Skills, One Command: Antigravity Awesome Skills Becomes the Cross-Tool Skill Standard
Antigravity Awesome Skills (v5.4.0) delivers 868+ battle-tested skills for Claude Code, Gemini CLI, Codex CLI, Cursor, GitHub Copilot, and five other AI coding assistants via a single npx command. With official skills from Anthropic, Vercel, OpenAI, Supabase, and Microsoft consolidated under one MIT-licensed repository, it's emerging as the portable skill layer for the fragmented AI coding agent landscape.
How Claude Sonnet 4.6 Agent Teams Achieve 4x Productivity: Practical Insights from Anthropic's Own Research
Two Anthropic studies—a survey of 132 internal engineers and an analysis of 1M+ real-world agent interactions—reveal the precise delegation strategies and autonomy patterns that enable high-performing teams to multiply output with Claude Sonnet 4.6 agent teams.
What Actually Makes OpenClaw Special: The Full Story from VibeTunnel to 200k+ GitHub Stars
The three-stage VibeTunnel→Clawdbot→OpenClaw evolution, Pi runtime philosophy, why HEARTBEAT is the real differentiator from Claude Code, and the ClawHub supply chain attack (12% of skills were malicious). An unvarnished look at the most used and most misunderstood OSS agent.
Latest Articles
Two AI Agent Communication Projects Hit Hacker News Simultaneously, Targeting MCP's Blind Spots
Aqua and Agent Semantic Protocol appeared on Hacker News on the same day, both tackling the same unsolved problem: how AI agents communicate directly without a central broker, across network boundaries, and asynchronously.
Claude Sonnet 4.6 Becomes the Default for Free and Pro Users — Outperforms Opus 4.5 on Coding Agent Benchmarks
Anthropic has made Claude Sonnet 4.6 the default model for claude.ai's Free and Pro plans. Released February 17, 2026, it matches Sonnet 4.5 pricing at $3/$15 per million tokens while internal Claude Code evaluations show it beating the previous frontier model, Opus 4.5, 59% of the time on agentic coding tasks.
Google Permanently Bans AI Pro Users for Accessing Gemini via OpenClaw, Continues Charging $250/Month
A Hacker News post garnering 140 points and 107 comments details how Google terminated Google AI Pro and Ultra accounts without warning after users accessed Gemini through OpenClaw, a third-party client. The incident surfaces deeper issues around prompt caching, subscription economics, and how AI providers enforce terms of service.