What's new in Claude

Added Tool(param:value) syntax for permission rules with wildcard support (e.g. Agent(model:opus) to block Opus subagents), improved nested .claude/ directory precedence so the closest directory wins, hardened auto mode by evaluating subagent spawns through the classifier before launch, and enhanced /doctor with a flat tree layout and clearer status icons. Also includes 12 bug fixes covering subagent transcript viewing, background session state, vim undo behavior, and MCP disallowed-tools handling.

/cd command — move the current session to a different working directory mid-conversation without rebuilding the prompt cache; the new directory's CLAUDE.md is appended as a message and the session relocates to the new project's storage so --resume finds it there.

Sub-agents can spawn sub-agents — nested agent trees are now supported with a live tree view in the subagent panel; background chains are capped at five levels deep to prevent runaway concurrency, while foreground chains can go any depth.

Safe mode for troubleshooting — launch with --safe-mode or CLAUDE_CODE_SAFE_MODE to disable all customizations (CLAUDE.md, skills, plugins, hooks, MCPs, custom commands) while keeping auth, model selection, and permissions intact. If a problem disappears in safe mode, one of those surfaces is the cause.

fallbackModel chains — configure up to three fallback models tried in order when the primary model is overloaded or unavailable; --fallback-model now also applies to interactive sessions. Also: plugin search in /plugin marketplace, new disableBundledSkills setting, deny rules now accept glob patterns in the tool-name position, and cross-session SendMessage authority is hardened in auto mode.

Model retirements: claude-sonnet-4-20250514 and claude-opus-4-20250514 have been retired — all requests now return an error. Upgrade to Claude Sonnet 4.6 and Claude Opus 4.8. Researchers can request ongoing access via the External Researcher Access Program.

Programmatic usage credits: Users on paid plans can now claim a dedicated monthly credit for programmatic use, covering the Claude Agent SDK, claude -p CLI, Claude Code GitHub Actions, and third-party Agent SDK apps. This separates programmatic usage from interactive plan limits.

AWS Claude Platform: The GET /v1/environments/{id}/work endpoint (listing pending work for self-hosted sandboxes) is now available on Claude Platform on AWS. See IAM actions docs for the GetEnvironment action that authorizes it.

Announced that starting June 15, paid-plan users can claim a dedicated monthly credit for programmatic usage (Agent SDK, claude -p, GitHub Actions, third-party Agent SDK apps) — separating it from interactive plan limits. Claude Code weekly limits also increased 50% through July 13.

Boris Cherny (Claude Code creator) posted live from "Code with Claude London": "Hello from Code with Claude London!" — the team is on a world tour expanding access. He's also been sharing that he hasn't written a single line of code himself in 2026, noting the team is "living in the future."

Thariq shared a deep-dive on dynamic workflows ("A harness for every task") — Claude Code writing its own task harness on the fly, custom-built per problem. Also shared how Anthropic teammates stay in sync with Claude's work, citing Suzanne's method of using Claude to track and summarize ongoing sessions.