Claude Code Daily Briefing - 2026-05-09

Release Summary

VersionDateKey Changes
v2.1.1375/9Fix Windows extension activation failure
v2.1.1365/8autoMode.hard_deny, MCP OAuth refresh token loss fix, WSL2 image paste, login loop fix
v2.1.1335/7worktree.baseRef, effort level in Hooks, memory optimization

Full release notes

New Features & Practical Usage

v2.1.136 — autoMode.hard_deny, Major MCP Stability Improvements (5/8)

v2.1.136 brings important improvements to auto mode control and MCP stability.

settings.autoMode.hard_deny: You can now set unconditional blocking rules at the auto mode classifier level. Unlike denyRules which prompt for confirmation, hard_deny blocks tool calls outright at the classifier level — a significant safety upgrade for long-running autonomous agents.

MCP OAuth stability fixes: MCP servers configured in .mcp.json, plugins, and claude.ai connectors no longer disappear after /clear. Additionally, OAuth refresh token loss during concurrent server refreshes and login loops from concurrent credential writes have been fixed.

WSL2 image paste: Pasting images from Windows clipboard into WSL2 terminals now works via PowerShell fallback.

Extended thinking 400 error fix: API 400 errors when redacted thinking blocks appeared after tool calls have been resolved.

{
  "autoMode": {
    "hard_deny": [
      { "tool": "Bash", "pattern": "rm -rf /" }
    ]
  }
}

GitHub v2.1.136

Anthropic Signs $1.8B, 7-Year Compute Deal with Akamai (5/8)

Anthropic has signed a $1.8 billion, 7-year cloud computing contract with Akamai Technologies — the largest single deal in Akamai’s history.

CEO Dario Amodei disclosed 80x growth in annualized revenue and usage in Q1, stating the company is “working as quickly as possible” to secure compute resources. Akamai shares surged 28% on the announcement.

This expands Anthropic’s infrastructure portfolio to:

Bloomberg | Yahoo Finance | Boston Globe

Developer Workflow Tips

Five Agentic Workflow Patterns for Claude Code

MindStudio has published a practical breakdown of five Claude Code workflow patterns gaining traction among practitioners:

  1. Sequential Flow: Step-by-step execution. Best for simple, linear tasks
  2. Operator: One orchestrator delegates subtasks to workers, collects results. Best for complex single tasks
  3. Split-and-Merge: Independent tasks split in parallel, results merged. Best for code review, multi-file analysis
  4. Agent Teams: Specialized agents collaborate persistently with distinct roles. Best for long-running projects
  5. Headless: Unattended execution in CI/CD or cron. Best for automated verification and deployment

The key design principle: sub-agents receive only their task prompt, not the parent’s full conversation. This context isolation is what allows Claude Code to handle large codebases without context window overflow.

MindStudio

Building Safety Nets with autoMode.hard_deny

The new autoMode.hard_deny in v2.1.136 complements existing denyRules. While denyRules require user confirmation, hard_deny blocks unconditionally at the classifier level. When running agents autonomously for extended periods, register commands that must never execute here.

{
  "autoMode": {
    "hard_deny": [
      { "tool": "Bash", "pattern": "git push.*--force" },
      { "tool": "Bash", "pattern": "DROP TABLE" }
    ]
  }
}

Security & Limitations

React/Next.js: 12 Security Vulnerabilities Disclosed — Immediate Patching Required (5/8)

React and Vercel simultaneously disclosed 12 security vulnerabilities spanning React Server Components and Next.js.

Severity: 6 High, 4 Moderate, 2 Low

Attack vectors: DoS, middleware bypass, SSRF, XSS, cache poisoning

Patch versions: React 19.0.6/19.1.7/19.2.6, Next.js 15.5.16/16.2.5

Many vulnerabilities cannot be safely blocked at the WAF level, making application code patching essential. If you’re managing Next.js projects with Claude Code, update dependencies immediately.

# Update Next.js
npm install next@latest react@latest react-dom@latest

GeekNews

Ecosystem & Plugins

EPAM Partners with Anthropic — Training 10,000 Claude-Certified Architects

EPAM Systems has signed a multi-year strategic partnership with Anthropic. The centerpiece is a CEO-mandated program to train 10,000+ Claude-certified architects.

Current progress:

The goal is moving enterprises from “AI experimentation” to “safe, large-scale applied AI” using Claude, Claude Code, and Claude Agent SDK.

EPAM | PR Newswire

Anthropic Reports 80x Revenue Growth — Infrastructure Expansion Accelerates

Dario Amodei disclosed 80x growth in annualized revenue and usage during Q1 when announcing the Akamai deal. Signing SpaceX (5/6) and Akamai (5/8) deals within the same week reflects the urgency of meeting explosive demand. The jump from 17x API traffic growth (announced at Code with Claude SF) to 80x suggests post-conference demand surged dramatically.

Community News

Minor Changes

Interesting Projects & Tools

By subscribing, you agree to our Privacy Policy.