2 min read AI-generated

Claude Code 2.1.140 to 2.1.143: Opus 4.7 as Default and New Agent Flags

Copy article as Markdown

Four new Claude Code releases in four days. The biggest change: Fast Mode now defaults to Opus 4.7. Plus new flags for configuring background sessions.

Featured image for "Claude Code 2.1.140 to 2.1.143: Opus 4.7 as Default and New Agent Flags"

Anthropic shipped four new Claude Code versions between May 12 and 15 — from 2.1.140 to 2.1.143. The pace remains relentless, and this batch includes some genuinely useful changes.

The highlights

Fast Mode now defaults to Opus 4.7: This is the biggest change. Fast Mode previously ran on Opus 4.6, now it uses Opus 4.7 by default. If you use Claude Code daily, you’re automatically getting the better model without changing a thing.

New agent flags: claude agents got a whole set of new flags: --add-dir, --settings, --mcp-config, --plugin-dir, --permission-mode, --model, --effort, and --dangerously-skip-permissions. These let you configure background sessions with much more flexibility.

Skills from plugins: Plugins with a SKILL.md in their root directory (without a skills/ subdirectory) are now automatically surfaced as a skill. This makes plugin development considerably simpler.

Bug fixes

The fixes address real-world pain points: a memory leak in HTTP/SSE MCP servers has been plugged — response bodies are now capped at 16 MB per SSE frame. The Skill(name *) wildcard match finally works correctly. Settings hot-reload now detects edits to symlinked ~/.claude/settings.json. And the /model picker correctly reflects ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL overrides.

My take

The switch to Opus 4.7 as the Fast Mode default is the kind of change you barely notice — but that measurably improves your daily workflow. Opus 4.7 is 13% better at coding tasks compared to its predecessor. That adds up over time.

The new agent flags show where things are heading: Claude Code is increasingly becoming an orchestrator for autonomous background tasks. If you’re using Routines and multi-agent workflows, you now have significantly more control over how those sessions are configured.

Sources: GitHub Releases · Releasebot