Anthropic files for IPO, workflow trigger is now ultracode

TLDR;

Anthropic filed a confidential S-1 with the SEC this week, opening the door to an IPO — but for builders, the June 15 deadline (model retirements + agent credit-pool split, both flagged last week) is now nine days out. Claude Code shipped five releases in the v2.1.160–v2.1.166 series with major new features: fallbackModel chaining, hardened cross-session security, version-range enforcement for platform teams, and a completely redesigned /fork that finally runs background agents with your full context (the old behavior moved to /branch). The dynamic-workflow trigger keyword also changed from workflow to ultracode.

---

---

API & Model Updates

Auto Mode Now Available on Bedrock, Vertex, and Foundry for Opus 4.7 and Opus 4.8

Released in Claude Code v2.1.158 (May 30), auto mode is now opt-in on all three third-party providers — set CLAUDE_CODE_ENABLE_AUTO_MODE=1 to enable it. This brings the same adaptive model-selection behavior available on the Anthropic API to enterprise deployments on cloud provider infrastructure.

---

Claude Code & CLI

Claude Code v2.1.166 — fallbackModel Chaining, Hardened Cross-Session Security, Thinking Controls

The flagship release this week (Jun 6) adds a fallbackModel setting that lets you 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. Cross-session security is significantly hardened: messages relayed via SendMessage from other Claude sessions no longer carry user authority, receivers refuse relayed permission requests, and auto mode blocks them. MAX_THINKING_TOKENS=0, --thinking disabled, and the per-model thinking toggle now disable thinking on models that think by default via the Claude API. Also fixed: a recurring "image could not be processed" error causing extra token usage, remote sessions becoming permanently stuck after brief backend disruptions during startup, and flickering in JetBrains IDE terminals.

---

Claude Code v2.1.163 — Version Range Enforcement, /plugin list, Stop Hook Feedback

The new requiredMinimumVersion and requiredMaximumVersion managed settings let platform teams enforce approved Claude Code version ranges — the client refuses to start if its version falls outside the range. /plugin list lists installed plugins with --enabled/--disabled filters. Stop and SubagentStop hooks can now return hookSpecificOutput.additionalContext to give Claude feedback and keep the turn going without being labeled a hook error. Also fixes claude -p hanging forever when a backgrounded command never exits, claude -p failing on Bedrock/Vertex/Foundry when CI=true and no Anthropic API key is set, and Bash commands failing under bazel and EDR-protected Go workflows due to a $TMPDIR override regression introduced in v2.1.154.

---

Claude Code v2.1.162 — claude agents --json waitingFor, Windsurf→Devin Desktop Rebrand, WebFetch Fix

claude agents --json now includes a waitingFor field showing what a waiting session is blocked on (e.g., a permission prompt) — essential for orchestration scripts that poll agent status. Remote Control now shows as a persistent footer pill instead of a startup message. Windsurf has been renamed to Devin Desktop in the /ide menu, /terminal-setup, and /scroll-speed following the editor's rebrand. Also fixed: WebFetch permission rules not being applied to built-in preapproved domains, and Windows permission rules never matching when spelled with backslashes or case-variant paths.

---

Claude Code v2.1.161 — OTEL Labels on Metrics, Parallel Tool Call Isolation, Clipboard Improvements

OTEL_RESOURCE_ATTRIBUTES values are now included as labels on metric datapoints, so you can slice usage metrics by custom dimensions like team or repo. Parallel tool calls now isolate failures: a failed Bash command no longer cancels other calls in the same batch — each tool returns its own result independently. claude agents rows now show done/total progress when work is fanned out.

---

Claude Code v2.1.160 — Security Prompts for Shell Startup Files, acceptEdits Build-Config Guard

Claude Code now prompts before writing to shell startup files (.zshenv, .zlogin, .bash_login) and ~/.config/git/, preventing unintended command execution. acceptEdits mode now also prompts before writing build-tool config files that grant code execution (.npmrc, .yarnrc*, bunfig.toml, .bazelrc, .pre-commit-config.yaml, .devcontainer/, etc.). Multiple background agent session bugs are fixed: restoring a completed session no longer drops chat history and re-runs the original prompt, and background sessions re-attached after overnight retire no longer lose their conversation.

---

Updated /fork in Claude Code: Background Agent with Full Context; Old Behavior Moves to /branch

/fork now runs a background agent with your exact context — system prompt, tools, history, model, and prompt cache — and returns the result to your current session. The old /fork behavior (copy transcript to a new session you drive yourself) has been renamed to /branch. This is a breaking change for any workflow or documentation that references /fork as a transcript-copy command.

---

Dynamic Workflow Trigger Renamed from "workflow" to "ultracode"

The keyword that kicks off a Claude Code dynamic workflow has changed from workflow to ultracode to prevent accidental triggers when using the word "workflow" in other contexts. You can still say "use a workflow for this," but ultracode is the explicit trigger going forward. Update any scripts, prompts, or documentation that relied on the old keyword.

---

New ant CLI for the Claude Platform: Call Any API Endpoint from Your Terminal

The ant CLI lets you call the Messages API, stand up Claude Managed Agents, pipe results into your shell, and pull full run traces from the terminal. Claude Code knows how to use ant via the built-in /claude-api skill — ask it to list sessions, upload PDFs, or debug a run without any glue code. Install via brew, curl, or go.

---

Rate Limits Reset for Pro and Max; Opus 4.8 Parallel Tool Call Bug Fixed

Anthropic reset 5-hour and weekly rate limits for all Pro and Max users after fixing a bug where Opus 4.8 requests triggered more parallel tool calls than intended, burning through usage faster than expected. The issue was unrelated to dynamic workflows and has been patched.

---

Agent SDK & Managed Agents

claude-agent-sdk-typescript v0.3.166 — MCP Resource Tools Fixed for Runtime-Added Servers

Fixed MCP resource tools not being injected for servers added at runtime via the mcp_set_servers control request — a bug that silently dropped resource tools for dynamically registered servers. Update with npm install @anthropic-ai/claude-agent-sdk@0.3.166.

---

claude-agent-sdk-typescript v0.3.163 — stop_task Idempotency, Stop Hook additionalContext

stop_task control requests now return success when the target task is already gone (not_found or not_running), making it safe to call from cleanup code without error handling for already-stopped tasks. Stop and SubagentStop hook events now support additionalContext in hookSpecificOutput, enabling non-error feedback that continues the turn — matching the Claude Code CLI behavior added in v2.1.163.

---

claude-agent-sdk-typescript v0.3.162 — Detect Refusals via stop_reason: "refusal" Without Text-Matching

Refusal error messages now carry stop_reason: "refusal" and stop_details on the assistant message and in session transcripts. If your code currently text-matches refusal responses, migrate to checking stop_reason — this is the stable, forward-compatible API surface.

---

claude-agent-sdk-typescript v0.3.161 — initialize Now Idempotent, Live Agent Switching via applyFlagSettings

A second initialize control request now returns the same success payload instead of an Already initialized error, making retry-on-startup patterns safe. applyFlagSettings now live-applies agent changes: switching the active agent (or passing null to reset) takes effect on the next turn in a running session without restarting.

---

claude-agent-sdk-python v0.2.88 — Trio Compatibility for Session Stores

Ported session_store code paths (TranscriptMirrorBatcher, session_resume, sessions) from raw asyncio primitives to anyio, fixing a crash (TypeError: trio.run received unrecognized yield message) when passing session_store= to query() or ClaudeSDKClient under trio. Install with pip install claude-agent-sdk==0.2.88.

---

Developer Tools & Community

Automating Business Analytics with Claude — Best Practices Guide

Anthropic published a new blog post covering best practices for building data analysis agents: skills architecture, data foundation design, and evaluation strategies for analytics automation. Published June 3 — direct link to the post was not available in source data; referenced from the @ClaudeDevs thread.

---

Dynamic Workflows Deep Dive — Best Practices and Non-Technical Use Cases

Anthropic's team published a deep-dive on workflow best practices and examples — including use cases beyond code (research, audits, migrations) — also available on the Claude Blog. Workflows are described as the biggest capability upgrade to Claude Code since skills and subagents, enabling Claude to handle large-scale parallel tasks that were previously impractical in a single session.

---

Bun Rust Port Rolling Out Internally to Claude Code

Jarred Sumner (Bun) confirmed the Rust port of Bun is being rolled out internally to Claude Code teams (not yet in public builds as of June 4). No issues reported after the first day of internal use — relevant for builders who care about Claude Code's underlying runtime performance characteristics.

---

Competitor Dev Tools

Cursor 3.7 — Design Mode Multi-Select, Voice Input for Agents

Cursor 3.7 (Jun 5) adds multi-element selection in Design Mode — click two or more elements in the browser, and the agent sees their code, layout, and visual relationships — plus voice input that stays available while an agent is mid-run so you can queue the next change without waiting. The Jun 4 Cursor SDK update adds custom tools (expose your own functions without standing up a separate MCP server), custom stores, and nested subagents.

---

Google Gemini CLI Shutdown Deadline: June 18, 2026

Gemini CLI and Gemini Code Assist IDE extensions stop serving requests for Google AI Pro, AI Ultra, and free-tier users on June 18, 2026; Antigravity CLI is the replacement. Builders whose workflows touch both Claude tooling and Google AI tooling should complete the migration before June 18.

---

Enterprise & Business

Anthropic Confidentially Submits Draft S-1 to the SEC

Anthropic has filed a confidential draft S-1 with the SEC, giving the company the option to pursue an IPO pending SEC review. No immediate platform or pricing changes are announced; noted here because public company obligations will eventually affect API terms, data handling disclosures, and roadmap communication norms.

---

Claude Partner Network: Services Track and Partner Hub Now Open

Anthropic launched a Services Track for consulting and implementation partners alongside a Partner Hub portal for managing partner resources, referrals, and co-selling. Relevant for builders at agencies or system integrators who want a formal channel with Anthropic for enterprise deals.

---

Expanding Project Glasswing — 150 Additional Organizations Across 15+ Countries

Anthropic extended access to Claude Mythos Preview to approximately 150 additional organizations in more than fifteen countries as part of Project Glasswing. If your organization is building in an early-access domain and hasn't applied, the expanded program is the entry point.

---

Research & Safety

Making Claude a Chemist — Opus 4.7 Matches Dedicated NMR Software

Anthropic's science blog details how Opus 4.7 matches — and on some tasks beats — dedicated NMR spectroscopy software for interpreting molecular structure. Directly relevant for builders in biotech, pharma, and materials science considering Claude for scientific instrument data pipelines.

---

What We Learned Mapping a Year's Worth of AI-Enabled Cyber Threats

Anthropic examined 832 malicious accounts and mapped their activity onto the MITRE ATT&CK framework to assess how well existing security community techniques hold up against AI-enabled attacks. Key finding: current defensive taxonomies largely cover AI-assisted threats, but the speed and scale of AI-enabled reconnaissance and initial access are outpacing detection tooling — relevant for any builder shipping Claude-powered security products.

---

Anthropic's Internal Data Shows Claude Accelerating AI Development — Mythos Preview at ~52x Code Speedup

Anthropic published internal benchmark data showing Mythos Preview achieving a ~52x average speedup on an AI training optimization task (compared to ~3x for Claude Opus 4 in May 2025 and 0x speedup for models from May 2024), and improving on human research decisions 64% of the time versus 22% in 2024. Anthropic is transparent that this does not yet confirm recursive self-improvement capability — research judgment (choosing the right problems) remains unsolved — but the trajectory has implications for anyone building long-horizon AI R&D tooling.

---

Action Items

Immediate:

• Update npm install @anthropic-ai/claude-agent-sdk@0.3.166 to fix MCP resource tools silently dropping for runtime-added servers
• Rename any scripts or prompts that use workflow as an explicit dynamic-workflow trigger — the new trigger keyword is ultracode
• Update any code or documentation that references /fork as a transcript-copy command — that behavior is now /branch; /fork now runs a background agent with your full context

By June 15, 2026:

• Migrate all production API calls from claude-sonnet-4-20250514 and claude-opus-4-20250514 to claude-sonnet-4-6 and claude-opus-4-8 — both model IDs return errors after June 15
• Claim your new monthly API credit via your Claude account before June 15 — Agent SDK, claude -p, Claude Code GitHub Actions, and third-party agent usage move to a separate per-user credit pool (no rollover, one-time opt-in required)

By June 18, 2026:

• Complete migration from Gemini CLI to Antigravity CLI if your workflows touch both Claude tooling and Google AI tooling — Gemini CLI stops serving requests on June 18 for Pro, Ultra, and free-tier users

---

Want to catch up? Read every past issue of This week in Claude.

---

Did you like this issue?

One tap is the fastest way to shape next week.

👍 Liked it  ·  👎 Didn't like it

And I read every reply. Hit reply and tell me what you want more of, or less of, and it goes straight into the next issue.