Claude Code v2.1.117 Release: Subagent Forking, Plugin Improvements, and Performance Fixes

Claude Code v2.1.117 introduces several practical improvements for developers using AI coding agents, focusing on subagent functionality, plugin management, and performance optimizations.

Key Features and Changes

• Forked subagents can now be enabled on external builds by setting CLAUDE_CODE_FORK_SUBAGENT=1
• Agent frontmatter mcpServers are now loaded for main-thread agent sessions via --agent
• Improved /model command: selections persist across restarts even when projects pin different models, and the startup header shows when the active model comes from project or managed-settings pins
• /resume command now offers to summarize stale, large sessions before re-reading them, matching existing --resume behavior
• Faster startup when both local and claude.ai MCP servers are configured (concurrent connect now default)
• Plugin improvements: plugin install on already-installed plugins now installs missing dependencies; dependency errors show "not installed" with install hints; claude plugin marketplace add auto-resolves missing dependencies
• Managed-settings enforcement: blockedMarketplaces and strictKnownMarketplaces now enforced on plugin install, update, refresh, and autoupdate
• Advisor Tool (experimental): dialog carries "experimental" label with learn-more link; sessions no longer get stuck with "Advisor tool result content could not be processed" errors
• Cleanup expansion: cleanupPeriodDays retention sweep now covers ~/.claude/tasks/, ~/.claude/shell-snapshots/, and ~/.claude/backups/
• OpenTelemetry updates: user_prompt events include command_name and command_source for slash commands; cost.usage, token.usage, api_request, and api_error include effort attribute when models support effort levels
• Native build improvements: macOS and Linux builds replace Glob and Grep tools with embedded bfs and ugrep available through Bash tool (Windows and npm-installed builds unchanged)
• Windows optimization: cached where.exe executable lookups per process for faster subprocess launches
• Default effort change: Pro/Max subscribers on Opus 4.6 and Sonnet 4.6 now default to high effort (was medium)

Notable Fixes

• Fixed Plain-CLI OAuth sessions dying with "Please run /login" when access tokens expire mid-session
• Fixed WebFetch hanging on very large HTML pages by truncating input before HTML-to-markdown conversion
• Fixed crash when proxies return HTTP 204 No Content
• Fixed /login having no effect when launched with CLAUDE_CODE_OAUTH_TOKEN env var and token expires
• Fixed prompt-input undo (Ctrl+_) behavior
• Fixed NO_PROXY not being respected for remote API requests under Bun
• Fixed Opus 4.7 sessions showing inflated /context percentages and autocompacting too early — was computing against 200K context window instead of Opus 4.7's native 1M
• Fixed subagents running different models than main agent incorrectly flagging file reads with malware warnings
• Fixed idle re-render loop reducing memory growth on Linux
• Fixed VSCode "Manage Plugins" panel breaking with multiple large marketplaces

This release primarily benefits developers using Claude Code for complex coding tasks with subagents, those managing multiple plugins, and users experiencing performance issues with large context windows or HTML processing.