Jonathon's AI Wiki

Claude Code Week 16 Release Digest (April 13–17, 2026)

6 min read

Source: Anthropic What’s New — Week 16 Type: Release Notes Releases: v2.1.105 → v2.1.113 Date Range: April 13–17, 2026

Five headline features in one week. Claude Opus 4.7 ships as the default on Max and Team Premium with a new xhigh effort level. Routines land on the web with a full trigger picker (schedule, GitHub events, API). /ultrareview adds an adversarial critique pass and a diffstat in the launch dialog. /usage breakdown surfaces what’s actually driving your limits. Native binaries replace bundled JavaScript across npm and standalone installs. Thirteen smaller wins including PreCompact hook compaction-blocking, ENABLE_PROMPT_CACHING_1H, and hardened Bash deny rules.

Headline features

Claude Opus 4.7 (new model)

Anthropic’s strongest coding model is now the default on Max and Team Premium; available on all other plans via /model. New xhigh effort level sits between high and max — applied as the default on first switch to 4.7. /effort now opens an interactive arrow-key slider when called without arguments.

> /model opus
> /effort xhigh

See Opus 4.7 Best Practices for effort-level tuning and cost implications.

Routines (web)

Templated cloud agents that fire on a schedule, a GitHub event, or an API call — without your machine running. The trigger picker now covers GitHub events with optional filters (author, title, branch, label, draft status). Every routine gets a tokened /fire endpoint for external systems. Create in the web UI or scaffold from the terminal:

> /schedule daily PR review at 9am

See Claude Code Routines for the full setup guide, environment variables, and connector list.

/usage breakdown (CLI)

/usage now shows what’s driving your limits: parallel sessions, subagents, cache misses, and long context — each with a percentage of your last 24 hours and an optimization tip. Press d or w to switch day/week views.

> /usage

/ultrareview (v2.1.111)

Comprehensive cloud code review. Fans your branch across parallel reviewers, runs an adversarial critique pass over each finding, and returns only verified findings while your terminal stays free. The launch dialog now shows a diffstat so you know what’s going up before confirming. Call bare for the current branch or pass a PR number:

> /ultrareview
> /ultrareview 1234

See Ultrareview for pricing, auth constraints, and how it fits vs /review.

Native binaries (v2.1.113)

claude now spawns a native per-platform binary (e.g., @anthropic-ai/claude-code-darwin-arm64) instead of bundled JavaScript. Node is no longer invoked on each run. Install command doesn’t change — npm pulls the right binary via optional dependency.

claude update
claude --version

Other wins

  • Auto mode for Max on Opus 4.7 — --enable-auto-mode flag no longer required.
  • Session recap + /recap — one-line summary of what happened while you were away. Run /recap on demand or disable from /config. See Week 17 where this becomes a headline.
  • /tui command — toggles between classic and flicker-free rendering mid-conversation. /focus is now a dedicated command (moved from Ctrl+O). See CLI Reference.
  • Push notification tool — with Remote Control connected and “Push when Claude decides” enabled, Claude can ping your phone when it needs you.
  • Plugin monitors manifest key — plugins can ship background watchers that auto-arm at session start or skill invoke. See Plugins and Marketplaces.
  • “Auto (match terminal)” option in /theme follows your terminal’s dark/light mode setting.
  • /fewer-permission-prompts — built-in skill that scans transcripts for common read-only Bash and MCP calls and proposes an allowlist for .claude/settings.json.
  • Claude discovers + runs built-in commands (/init, /review, /security-review) via the Skill tool — no user invocation required for agentic flows.
  • PreCompact hooks can block compaction — exit with code 2 or return {"decision":"block"} to prevent the compaction from proceeding. See Hooks.
  • ENABLE_PROMPT_CACHING_1H — opts API key, Bedrock, Vertex, and Foundry users into 1-hour prompt cache TTL. See CLI Reference.
  • sandbox.network.deniedDomains — carves specific domains out of a broader allowedDomains wildcard. See CLI Reference.
  • /undo is now an alias for /rewind; /proactive is now an alias for /loop. See Scheduled Tasks.
  • Hardened Bash deny rules — match through env/sudo/watch wrappers; Bash(find:*) allow rules no longer auto-approve -exec or -delete.

Key Takeaways

  • Opus 4.7 + xhigh effort is the new performance ceiling for Max/Team Premium subscribers. First switch to 4.7 auto-sets xhigh; the interactive /effort slider makes tuning accessible without memorizing level names.
  • Routines is the biggest workflow change — the /fire endpoint and GitHub event triggers make it viable as a real CI/CD integration point, not just a cron replacement. Trigger filtering (author, branch, label) reduces noisy triggers.
  • /ultrareview adversarial critique pass is a significant quality improvement over the original pre-W16 behavior: findings are validated against each other before surfacing, which should meaningfully reduce false positives.
  • /usage breakdown is the first observability tool for subscription limits. If you’ve ever hit an unexplained limit, this is how you diagnose whether it’s long context, subagents, or cache misses.
  • PreCompact blocking is a load-bearing new hook capability. Teams with expensive CLAUDE.md prefixes can now prevent compaction during critical mid-session workflows.
  • Native binaries is a behind-the-scenes improvement that will matter for startup time, especially on macOS Apple Silicon. No user action needed unless on a managed environment blocking arbitrary binaries.
  • ENABLE_PROMPT_CACHING_1H is high-value for API-key users with repeated long prefixes. The 5-minute default TTL has always been the main limitation for API-key prompt caching economics — 1 hour makes the cache hit rate dramatically higher for workloads that pause mid-session.

Try It

  1. Switch to Opus 4.7 with /model opus then run /effort to explore the interactive slider. Test xhigh vs high on a complex refactor to calibrate the tradeoff.
  2. Create your first Routine via the web UI at claude.ai/code. Use a GitHub trigger (e.g., PR opened) with a label filter. Watch the first live run before setting a schedule.
  3. Run /usage after a heavy coding session and identify which driver (parallel sessions, subagents, context) is consuming the most of your limit.
  4. Add a PreCompact hook to your CLAUDE.md for any project where mid-session compaction would lose critical context.
  5. If you’re on API-key / Bedrock / Vertex / Foundry: set ENABLE_PROMPT_CACHING_1H=1 and re-baseline your prompt caching hit rate on a workload that pauses mid-session.
  6. Upgrade to v2.1.113 (claude update) and verify with claude --version — you should see no Node startup delay.
  • Week 15 (April 6–10) — Ultraplan research preview, Monitor tool, /autofix-pr, /team-onboarding, default effort high for non-subscription tiers.
  • Week 17 (April 20–24) — /ultrareview public research preview, session recap as headline, custom themes, web redesign.
  • Opus 4.7 Best Practices — effort-level tuning, cost implications, when to use xhigh vs high vs max.
  • Claude Code Routines — full Routines setup guide including environments, connectors, and prompt best practices.
  • Ultrareview — entity article for /ultrareview, pricing, auth constraints, comparison with /review.
  • Hooks — PreCompact blocking, ENABLE_PROMPT_CACHING_1H interaction with caching behavior.
  • Plugins and Marketplaces — plugin monitors manifest key for background watchers.
  • Scheduled Tasks — /proactive alias for /loop, Monitor tool pairing.
  • CLI Reference — /tui, /focus, /recap, /usage, sandbox.network.deniedDomains, ENABLE_PROMPT_CACHING_1H.

Graph View

Backlinks