What's new in Claude

New "Dynamic workflow size" setting in /config (small/medium/large) controls how many subagents Claude spawns in workflows. Added workflow.run_id and workflow.name OpenTelemetry attributes for tracking workflow-spawned agents. Also fixed: inline history search crashes, /rename on background sessions, mTLS handshake failures, Remote Control command routing, image/file handling from mobile, voice dictation retry loops, and session resumption performance. /review <pr> reverts to fast single-pass; use /code-review <level> <pr#> for multi-agent review.

Claude Sonnet 5 sessions no longer use the mid-conversation system role for harness reminders, reducing prompt noise in long-running sessions.

AskUserQuestion dialogs no longer auto-continue by default — Claude now waits for explicit confirmation. The default permission mode is renamed to "Manual" across all interfaces. Multiple background session reliability and MCP configuration fixes also land.

claude mcp login: authenticate a configured MCP server from your shell instead of the interactive /mcp menu; clear stored credentials later with claude mcp logout.

Shell mode responds to command output: running ! npm test now gets an inline explanation without needing a second prompt.

/rewind can resume from before /clear: conversations can now be wound back to a point before the context was cleared, recovering prior work without starting over.

Background subagent permission prompts now surface in the main session instead of auto-denying, so long-running background work no longer silently stalls waiting for approval.

Anthropic restored access to Claude Fable 5 and Claude Mythos 5 following updated classifier deployment. Note: Fable 5 is expected to come off subscriptions after July 7 due to capacity constraints — Anthropic will restore it as soon as capacity allows.

Claude Sonnet 5 (claude-sonnet-5) launched with a 1M token context window and 128k max output at introductory pricing of $2/$10 per MTok through August 31, 2026 ($3/$15 thereafter). Adaptive thinking is on by default; manual extended thinking is removed; new tokenizer produces ~30% more tokens for the same text.

Claude Managed Agents gains event deltas on session streams (preview agent text as it generates), backward pagination for session listing, per-session agent config overrides, vault injection location controls, and lifecycle webhooks for agent, deployment, and run events.

Claude Code weekly limits increased 50% through July 13 for all Pro, Max, Team, and Enterprise users. Artifacts in Claude Code also expanded to Pro and Max plans — Claude writes code, publishes it live to claude.ai, and updates it in real time as you work.

Announced the "Built with Claude: Life Sciences" virtual hackathon — $100k prize pool in credits for building with Claude and Claude Code, in partnership with the Gladstone Institute. Quiet today.

"Reflecting on what engineers love about Claude Code, one thing that jumps out is its customizability: hooks, plugins, LSPs, MCPs, skills, effort, custom agents, status lines, output styles — every engineer uses their tools differently. We built Claude Code from the ground up with great defaults and deep customizability."

"A harness for every task: dynamic workflows in Claude Code" — Thariq wrote about how Claude now writes its own workflow harness custom-built for each task, orchestrating dozens to hundreds of subagents. Also: been asking others at Anthropic how they stay in the loop on a fast-moving codebase.