Library
Claude Code Documentation (June 2026) · 6 of 15
Claude Code Documentation (June 2026)
ai HIGH

What's New as of June 2026 (Consolidated Changelog)

Claude Code Documentation (June 2026) Anthropic
changelog models settings slash-commands hooks env-vars

Key Principle

A point-in-time snapshot of durable Claude Code feature/config introductions through the Opus 4.8 release window (June 2026). Release notes change frequently — verify against the live changelog (https://code.claude.com/docs/en/changelog) and claude --version. Identifiers are preserved verbatim.

Why This Matters

Claude Code ships often, and superseded entries get cut from the live notes. This consolidates the durable introductions — new models, settings, slash commands, env vars, hooks, and security hardening — so you know what exists and what to reach for, without trawling raw release notes.

Models & Effort

  • Opus 4.8 — new default model; defaults to high effort. "Now defaults to high effort · /effort xhigh for your hardest tasks." (Section: Release Notes)
  • Opus 4.8 fast mode — "2x the standard rate for 2.5x the speed." (Section: Release Notes)
  • /effort xhigh — highest effort tier for hardest tasks; not offered on models that can't run xhigh. xhigh was introduced on Opus 4.7 (level between high and max; other models fall back to high) (p. 5). On non-xhigh models /effort xhigh is unavailable.
  • Claude Fable 5 (Mythos-class) — "made safe for general use." 1M context by default; [1m] suffix now stripped automatically. Introduced v2.1.170.
  • fallbackModel — configure up to three fallback models, tried in order when the primary is overloaded/unavailable; a turn retries once on the fallback for unexpected non-retryable API errors (auth/rate-limit/request-size/transport still surface immediately). --fallback-model flag now applies to interactive sessions too.
  • enforceAvailableModels (managed) — makes the availableModels allowlist also constrain the Default model; a Default resolving to a disallowed model falls back to the first allowed model. Lower-scope settings can no longer widen a managed availableModels list.
  • Lean system prompt — now default for all models except Haiku, Sonnet, and Opus 4.7 and earlier.

Parallelism

  • ultracode — trigger keyword for dynamic workflows (renamed from workflow); "ask Claude to create a workflow and it orchestrates work across tens to hundreds of agents in the background." Backspace right after the keyword dismisses (same as alt+w). A /config "Workflow keyword trigger" setting can disable the trigger.
  • /workflows — view your workflow runs (running/completed, phase, agents finished).
  • Agent teams (experimental) — CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 gives every session one implicit team; spawn teammates via the Agent tool's name parameter. (TeamCreate/TeamDelete removed; team_name accepted but ignored.)
  • Nested subagents — subagents can spawn their own sub-agents, up to 5 levels deep; foreground subagents respect the same 5-level limit as background ones.
  • CLAUDE_CODE_FORK_SUBAGENT=1 — enable forked subagents (inherit full conversation context); now works in non-interactive sessions (SDK and claude -p).

Slash Commands

  • /code-review (renamed from /simplify) — reports correctness bugs at a chosen effort level, e.g. /code-review high; --comment posts findings as inline GitHub PR comments; --fix applies findings to the working tree after the review.
  • /usage — merges /cost + /stats (both retained as shortcuts) (p. 2); now shows per-category breakdown driving limits: skills, subagents, plugins, per-MCP-server cost.
  • /ultrareview — cloud multi-agent code review; no-arg reviews current branch, /ultrareview <PR> reviews a GitHub PR (p. 5). claude ultrareview [target] runs it non-interactively for CI (--json; exit 0/1).
  • /goal — set a completion condition; Claude keeps working across turns until met. Works in interactive, -p, and Remote Control; shows live elapsed/turns/tokens overlay.
  • /loop — /proactive is now an alias for /loop.
  • /rewind — /undo is now an alias for /rewind.
  • /focus — toggles focus view (split out of Ctrl+O, which now toggles only normal/verbose transcript).
  • /recap — session recap; provides context when returning to a session. Configurable in /config.
  • /cd — move a session to a new working directory without breaking the prompt cache mid-session (reports the correct git branch after the move).
  • /tui fullscreen — flicker-free fullscreen rendering; /tui switches TUI rendering mode in the same conversation.
  • /scroll-speed — tune mouse-wheel scroll speed with live preview.

Settings

  • worktree.bgIsolation: "none" — background sessions edit the working copy directly without EnterWorktree; for repos where worktrees are impractical.
  • worktree.baseRef (fresh | head) — choose whether worktrees branch from origin/ or local HEAD. Default fresh reverts EnterWorktree base to origin/ (was local HEAD since 2.1.128); set "head" to keep unpushed commits.
  • autoMode.hard_deny — auto-mode classifier rules that block unconditionally, regardless of user intent or allow exceptions.
  • parentSettingsBehavior (first-wins | merge) — admin-tier key letting admins opt SDK managedSettings into the policy merge.
  • allowAllClaudeAiMcps — load claude.ai cloud MCP connectors alongside managed-mcp.json.
  • disableBundledSkills (env: CLAUDE_CODE_DISABLE_BUNDLED_SKILLS) — hides bundled skills, workflows, and built-in slash commands from the model.
  • skillOverrides — per-skill visibility: off (hide from model and /), user-invocable-only (hide from model only), name-only (collapse description).
  • sandbox.network.deniedDomains — block specific domains even when a broader allowedDomains wildcard would permit them (p. 4).
  • prUrlTemplate — point footer PR badge at a custom code-review URL instead of github.com (p. 1).
  • tui — backs /tui rendering mode; set to enable flicker-free fullscreen rendering. (Section: Changelog)

Env Vars

  • CLAUDE_CODE_SAFE_MODE / --safe-mode — start with ALL customizations disabled (CLAUDE.md, plugins, skills, hooks, MCP servers) for troubleshooting.
  • ANTHROPIC_BEDROCK_SERVICE_TIER — select Bedrock service tier (default, flex, priority); sent as X-Amzn-Bedrock-Service-Tier.
  • CLAUDE_CODE_SESSION_ID — added to Bash tool subprocess env, matching session_id passed to hooks.
  • CLAUDE_EFFORT / $CLAUDE_EFFORT — active effort level; readable by Bash tool commands and referenceable in skill content.
  • ANTHROPIC_WORKSPACE_ID — workload identity federation; scopes the minted token to a specific workspace.
  • CLAUDE_CODE_HIDE_CWD — hide working directory in startup logo (p. 1).
  • DISABLE_UPDATES — blocks all update paths including manual claude update; stricter than DISABLE_AUTOUPDATER (p. 2).
  • OTEL_LOG_RAW_API_BODIES — emit full API request/response bodies as OTel log events for debugging (p. 5).
  • ENABLE_PROMPT_CACHING_1H — opt into 1-hour prompt cache TTL (API key, Bedrock, Vertex, Foundry); supersedes deprecated-but-honored ENABLE_PROMPT_CACHING_1H_BEDROCK. (Section: Changelog)
  • FORCE_PROMPT_CACHING_5M — force 5-minute prompt cache TTL. (Section: Changelog)

Hooks

  • MessageDisplay (new event) — lets hooks transform or hide assistant message text as displayed.
  • PreCompact — can now block compaction by exiting with code 2 or returning {"decision":"block"}. (Section: Changelog)
  • SessionStart → reloadSkills: true — re-scan skill directories within the same session; hookSpecificOutput.sessionTitle sets the session title on startup and resume.
  • Hook args: string[] (exec form) — spawn command directly without a shell, so path placeholders never need quoting.
  • continueOnBlock: true (for PostToolUse) — feed the hook's rejection reason back to Claude and continue the turn.
  • type: "mcp_tool" — hooks can invoke MCP tools directly (p. 2).

Security / Search

  • Native bfs / ugrep — on native macOS/Linux builds, Glob/Grep tools replaced by embedded bfs and ugrep via the Bash tool (faster, no extra round-trip). Windows and npm builds unchanged (p. 3).
  • Bash deny-rule hardening — deny rules now match commands wrapped in env/sudo/watch/ionice/setsid; Bash(find:*) allow rules no longer auto-approve find -exec/-delete; dangerouslyDisableSandbox no longer runs outside the sandbox without a prompt; glob support added in deny rules ("*" denies all tools) (p. 4).
  • PowerShell -ExecutionPolicy Bypass — PowerShell tool now passes -ExecutionPolicy Bypass by default and is enabled by default on Windows for Bedrock/Vertex/Foundry; opt out via CLAUDE_CODE_POWERSHELL_RESPECT_EXECUTION_POLICY=1 or CLAUDE_CODE_USE_POWERSHELL_TOOL=0.

Key Commands & Config

  • claude --version to check the installed build; the live changelog is generated from CHANGELOG.md.
  • /effort xhigh, /code-review high --comment --fix, /usage, /ultrareview, /goal, /workflows, /tui fullscreen.
  • Managed model control: availableModels + enforceAvailableModels, fallbackModel.

Rules of Thumb

  • Treat this file as a verification checklist, not a source of truth — re-check the live changelog before relying on a recent feature.
  • Renames keep old names as aliases for a while (/undo→/rewind, /proactive→/loop, workflow→ultracode) — but don't assume the alias is permanent.
  • Managed enforceAvailableModels/availableModels and fallbackModel are the levers for controlling which models a team can actually reach.

Related References