Skip to main content
Back

Anthropic Security Plugin , Cursor Model Router , Claude Agents 500

Anthropic ships Claude Code security plugin, Cursor cuts AI costs 60%, tokenizer hits 24GB/s, Microsoft open-sources prompt rewriter

July 23, 2026 6 min read

Hey,

AI coding tools are quietly becoming infrastructure, not just productivity hacks. Anthropic ships a free security plugin that watches your code as you type. Cursor ships a router that picks the right model per request, saving teams 60% on costs.

The pattern? The "write fast, fix later" era is ending. Speed and safety are now table stakes, baked in automatically.

Meanwhile, self-improving agents are learning to fix themselves. We're not just building faster coders. We're building systems that improve on their own.
In Partnership with · partner_image
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Summary

Read time: 6 min 29 sec

Top News

▸ Anthropic ships Claude Code security plugin that scans your code for vulnerabilities

Augment Code

▸ Engineering loops, solved.

Top News

▸ Cursor ships smart model router that cuts AI coding costs by 60%

Airbyte

▸ Airbyte Preps, So Claude Can Cook

Top Repo

▸ Anthropic ships effort controls, 500 skills, and webhooks to Claude Managed Agents

Signals

1. New open-source tokenizer runs 1000x faster than HuggingFace, hits 24GB/s
2. Granola: Granola, the AI notepad, now turns months of meeting notes into a searchable memory with cited answers
3. Outlines open-source library forces LLMs to output valid structured data every time
4. Jürgen Schmidhuber surveys how modern AI agents learn to improve themselves
5. Open-source world model slashes game prototyping from weeks to minutes
6. Microsoft open-sources a tool that rewrites and improves its own AI prompts
Top News
Anthropic ships Claude Code security plugin that scans your code for vulnerabilities
11,838 Likes
Anthropic ships Claude Code security plugin that s

Anthropic just dropped the Claude Security plugin for Claude Code in beta, and it's worth paying attention to.

Here's the problem it solves: AI writes code fast, but fast also means security holes sneak in faster. Think SQL injection, hardcoded secrets, unsafe function calls. They slip through before anyone notices.

This plugin sits inside your terminal and catches those issues before they become your problem. The cool part? No extra commands to remember. It runs automatically as you code.

It works in three layers:

  • Every file edit triggers an instant pattern scan, flagging dangerous stuff like eval(), os.system(), and innerHTML with zero extra cost
  • After each coding turn, a separate Claude model reviews your full git diff for logic-level flaws like auth bypasses and insecure references
  • At commit time, it does a deep cross-file review to catch issues that only show up in context

It's free for all Claude Code users. Install it with /plugins in your terminal, then run /reload-plugins. Done.

Read more
forward →
Presented by Augment Code
Engineering loops, solved.

You've stopped prompting and started designing loops. But every loop on your team runs in isolation: no shared memory, no shared context, nothing compounding.

Augment Code's Agentic SDLC report maps the move from individual agents to a coordinated, team-level system: how humans and agents split work across all six stages of software delivery, and the three traits of systems that compound.

Read the report
partner with us →
Top News
Cursor ships smart model router that cuts AI coding costs by 60%
8,803 Likes
Cursor ships smart model router that cuts AI codin

You know how you always have to pick which AI model to use in Cursor? Too powerful and you burn budget. Too cheap and the output suffers. Cursor Router just killed that tradeoff.

Here is how it works: instead of you picking a model, Cursor analyzes each request and picks for you. Simple autocomplete? Cheap model. Complex refactor? Frontier model. You stop thinking about it.

You get three modes to tune the behavior:

  • Intelligence: best quality, still cheaper than always using top models
  • Balance: solid quality at around 36% lower cost
  • Cost: maximum savings for lighter tasks

In early testing, teams with thousands of users saved 30-50% with no drop in quality versus routing everything to Opus 4.8. Cost per commit dropped too, which is the number that actually matters.

It is live now on Teams and Enterprise plans. Admins can control which models are allowed, set defaults per team, and lock certain modes. Auto mode is now powered by Router out of the box.

Read more
forward →
Presented by Airbyte
Airbyte Preps, So Claude Can Cook

The prep station for your agents

Airbyte pre-indexes your CRM, tickets, and repos into one Context Store, so Claude Code queries once instead of paginating through raw APIs.

50+ connectors, one credential.

Try airbyte
partner with us →
Top Repo
Anthropic ships effort controls, 500 skills, and webhooks to Claude Managed Agents
3,768 Likes
Anthropic ships effort controls, 500 skills, and w

Claude Managed Agents just got a solid set of updates that make building multi-agent systems a lot less painful.

The biggest quality-of-life fix: you used to need separate API calls to create a session and send it events. Now you can seed a session with up to 50 events in a single create call, cutting setup time and complexity.

  • Effort levels per agent -- control how hard each agent thinks. Lower effort means faster, cheaper responses. Higher effort means more thorough work.
  • 500 skills per session -- skills are task-specific instructions you give an agent. You can now stack up to 500 across all agents in one session.
  • Webhooks for environments and memory stores -- get notified when things happen in the background, so you can trigger other actions automatically.
  • Sub-agent event streaming -- watch what your child agents are doing in real time.

Together, these let you build smarter pipelines where different agents handle different tasks at different cost levels. Try it via the claude-cookbooks repo on GitHub.

Read more
forward →
Signals
1
New open-source tokenizer runs 1000x faster than HuggingFace, hits 24GB/s
3,433 Likes
2
Granola, the AI notepad, now turns months of meeting notes into a searchable memory with cited answers
Presented by Granola
3
Outlines open-source library forces LLMs to output valid structured data every time
15,063 Stars
4
Jürgen Schmidhuber surveys how modern AI agents learn to improve themselves
673 Likes
5
Open-source world model slashes game prototyping from weeks to minutes
373 Likes
6
Microsoft open-sources a tool that rewrites and improves its own AI prompts
2,763 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 300,000 developers stay informed and ahead. We're passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We're based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 300,000+ AI developers?
Work with us