Robat Das Orvi

A2A协议认证安全：代理卡片与签名请求

A2A 协议的身份验证安全性不仅仅依赖于签名的代理卡。有效的签名只能证明是谁签署了该卡，而不能证明您是否应该信任它。

多个会话之间的 Claude Code OAuth 令牌刷新竞争

当多个会话共享一个登录时，Claude Code OAuth 令牌刷新会失败：轮换令牌、过时的覆盖以及 1996 年的 TOCTOU 并发问题。

从 API 响应推断 JSON Schema：optional、nullable、union

从 API 实际响应推断 JSON schema 得到的是样例而非契约。

当接口响应中出现 optional 字段、nullable 值和 union 类型时，drift 检测容易把这些正常变化误判为破坏性变更，从而产生大量误报或漏报。

作者认为，drift 检测的可靠性上限取决于推断出的 schema 是否忠实反映了真实的契约，而基于有限样本推断出的 schema 天然存在偏差。

LLM Structured Output: Validation, Retry or Constrained Decoding?

Constrained decoding fixes JSON syntax, not silent failure.

Why validate-and-retry loops for LLM structured output still lose work, and how to stop it.

用Playwright测试Manifest V3 Chrome扩展：ScreenGuard日志

文章记录了用Playwright对Manifest V3 Chrome扩展做端到端测试的流程：在持久化浏览器上下文中加载扩展，定位其Service Worker（扩展在后台运行的脚本），并模拟屏幕共享行为以测试仅屏幕共享场景下的功能。

该方法针对ScreenGuard这类需要监听屏幕共享状态的扩展。

AI代理的持久化执行：Temporal与纯状态存储的对比

文章对比了AI多代理工作流中两种持久化执行方案：使用Temporal这样的持久化执行引擎，与用普通状态存储（记录每个步骤的输出和中间状态，以便崩溃后从中断处恢复）做检查点。

核心判断是：Temporal在复杂长流程中值得其额外成本，但简单场景下纯状态存储已够用。

最容易被忽视的风险在恢复环节——崩溃后重新执行时，已完成的副作用（如已发送的请求、已写入的数据）可能再次触发，导致重复操作。

JSON API Response Diff: How to Ignore Dynamic Fields Safely

How to ignore dynamic fields in a JSON API response diff: timestamps, request IDs and reordered arrays, without hiding real breaking changes.

Policy as Code for AI Agent Tool Authorization: OPA or Cedar

Policy as code for AI agent tool authorization: why OPA or Cedar should decide every tool call, not the prompt.

Real incidents, one hard lesson.

MCP Tool Poisoning Attacks: Hidden Tool Descriptions and Rug Pulls

MCP tool poisoning hides instructions in a tool description, and rug pulls change them after you approve.

Why one-time approval fails, with real cases.

Feature Flags for AI Agents: Disable One Tool Without a Redeploy

Feature flags for AI agents: how to disable one tool in production without a redeploy, and why the flag itself can hide the failure.

AI Agent Rollback: Why the Saga Pattern Fails for Tool Calls

Compensating transactions can't un-delete a production database.

Why the AI agent rollback saga pattern breaks, with 2025-2026 incident data.

White-Label Client Portal on a Custom Domain: Automating SSL

A white-label client portal on a custom domain needs SSL issued and renewed for every agency.

Here is where that automation fails silently.

Postgres Row-Level Security for Multi-Tenant SaaS: Design and Pitfalls

Postgres row level security for multi-tenant SaaS: why 170 Lovable apps leaked, the owner-bypass trap, tenant policy design, and the test that catches leaks.

Least Privilege for AI Agents: Scoping Tools, Keys and Filesystems

Least privilege for AI agents in practice: how to scope tools, credentials and filesystems, with the Replit and Amazon Q incidents as evidence.

Digital Nomad Visas in 2026: What Actually Works for Developers

Digital nomad visas for developers in 2026: which income tests are real, how the 183-day tax rule catches people, and what to check before applying.

LLM-as-a-Judge Reliability: Check Your Grader Against Human Labels

LLM-as-a-judge reliability depends on more than an 80% agreement score.

Here's what to check before you trust an AI grader.

Multiple AI Coding Agents, Same Repository: The Bugs Git Won't Catch

Running multiple AI coding agents in the same repository breaks in ways git never reports.

Shared ports, databases and lockfiles, and how to guard them.

LLM API Spending Limits: Why Alerts Don't Stop the Bill

An LLM API spending limit that only alerts won't stop a bill.

Here is what 2026 survey data and provider docs show about the gap.

When an Upstream API Changes, Your Agent's Tools Break Silently

API breaking changes detection fails for AI agents because the break returns 200 OK.

Why versioning and contract tests miss it, and what catches it.

Pre-Flight Budget Checks: Blocking an LLM Call Before It Is Billed

Want to limit OpenAI API spending?

Dashboard caps fire after the bill is already running.

The check has to run before the request exists.

How Arne Slot Changed Liverpool's Press Without Destroying What Klopp Built

Arne Slot's Liverpool tactics won a title with a calmer press.

Then one night in Paris, he abandoned them.

Here's the decision that ended his reign.

What Is Human Dignity? Why People Quit Jobs and Lose Money to Protect It

The human dignity meaning you learned is too abstract.

Real data shows people give up pay, jobs and even years of life over it.

Here's the evidence.