天道好轮回！ 前脚刚递交完招股书，Anthropic就收到一份「大礼」。 《华盛顿邮报》报道，半年来靠「Claude Mythos」把全球操作系统和浏览器漏洞翻了个底朝天的Anthropic，自家后院起火—— 被。。。黑。。。了。。。

2026年10月22日，全球AI圈炸开一条‘温柔核爆’——Anthropic官宣：把自家打磨两年的AgentSkills标准，连同名字、Logo、GitHub组织、agentskills.io域名，一股脑捐给了Linux基金会旗下的Agentic AI Foundation（AAIF）。不是授权，不是合作，是真·交钥匙——从此，AI智能体的‘说明书’，正式从一家公司的私藏抽屉，搬进了全人类共建共治 ...

去年年底，团队里一位同事改了一行 SKILL.md 的描述文字，把“优先查询订单状态”改成了“根据用户诉求选择合适工具”。代码没报错，pytest 全绿，PR 顺利合并。三天后线上开始出现跳过订单校验直接退款的事故。 这件事让我意识到一个问题：我们给 Skill 写测试的方式，和给普通代码写测试的方式，几乎完全是两回事。 普通代码改坏了，测试挂掉；Skill 改坏了，Agent 依然“能跑”，只是 ...

As you start delegating work to AI, the number of files containing API keys and passwords (.env) on your computer grows.

Incidents where API keys are leaked due to code written by developers, leading to cloud bankruptcy, are endless. When requesting code reviews from LLMs, cases where confidential credentials or unverif ...

AssemblyAI published a Python tutorial for a real-time voice agent that combines streaming speech recognition, LLM Gateway ...

The root-cause fix will take weeks; protection can’t wait Sweet Security today announced Sweet Safety Net, extending its ...

Look closely at line 11 in the 2027 code block—there is a brutal syntax bug hiding in plain sight. Check the thread breakdown below: regex or LLM API? https://t.co/QIejo3fk6Q ...

Brotli decoder flaw CVE-2026-111497 affects Node.js, Bun, Python packages and Envoy; code execution was demonstrated on one ...

Learn why coordinate clamping differs from global gradient norm clipping, with a complete Python implementation, numerical ...

For AI agents: read AGENTS.md before making any changes. A read-only viewer that shows Pybricks Python (LEGO SPIKE Prime, FLL robots) as Pybricks-style Blockly blocks. Built for teammates who prefer ...