KK Blog

Menu

Archives

2026 / 07

Turn the Claude Cookbook into an Agent Regression Suite

2026-07-25

最佳实践 #Claude Cookbook#AI Agent Testing#Regression Testing#LLM Evals#Anthropic#Model Migration#CI/CD#Agent Reliability
Convert Claude Cookbook examples into versioned agent regression tests with deterministic assertions, repeated trials, cost budgets, and release gates
Cursor 重写 SQLite：测试全过为何仍不能上线

2026-07-25

最佳实践 #Cursor#minisqlite#Agent Swarm#SQLite#软件测试#Harness Engineering#AI编程#生产就绪
Cursor Agent swarm 在私有 SQL 逻辑测试中达到 100%。本文拆解它真正证明了什么，以及 AI 生成系统上线前仍需哪些验收。
Cursor minisqlite: Why 100% Tests Are Not Production

2026-07-25

最佳实践 #Cursor#minisqlite#Agent Swarms#SQLite#Software Testing#Harness Engineering#AI Coding#Production Readiness
Cursor's agent swarm reached 100% on held-out SQL logic tests. Learn what that proves, what it omits, and how to assess AI-generated systems for produ
AI 接管飞行：从能力演示到安全论证

2026-07-25

调研 #AI飞行控制#无人机安全#Drone-Bench#Project Pilot#DARPA#F-16 VENOM#AI Agent#Safety Case
Drone-Bench 与 AI 控制 F-16 展示了能力进步，也暴露了自主飞行获得信任前必须补齐的证据链。
AI Flight Control Needs a Safety Case, Not a Demo

2026-07-25

调研 #AI Flight Control#Drone Safety#Drone-Bench#Project Pilot#DARPA#F-16 VENOM#AI Agents#Safety Case
Drone-Bench and AI-controlled F-16 tests show rapid capability progress. They also reveal the evidence required before autonomous flight can be truste
GEPA：当评估器成为编程接口

2026-07-25

最佳实践 #GEPA#optimize_anything#LLM优化#评估器#Prompt优化#AI Agent#Reward Hacking#Harness Engineering
GEPA optimize_anything 把评估器变成 prompt、代码、Agent 和配置的稳定接口。本文给出生产级设计与验证方法。
GEPA optimize_anything: The Evaluator Is the Interface

2026-07-25

最佳实践 #GEPA#optimize_anything#LLM Optimization#Evaluators#Prompt Optimization#AI Agents#Reward Hacking#Harness Engineering
GEPA optimize_anything makes the evaluator the stable interface for prompts, code, agents, and configs. Here is how to design one safely.
ChatGPT Health：连接医疗数据的信任边界

2026-07-25

最佳实践 #ChatGPT Health#医疗记录#健康数据隐私#HIPAA#Apple Health#AI安全#EHR#数据治理
ChatGPT Health 可以在普通对话中读取连接的病历。本文拆解权限、记忆、删除、HIPAA 和端到端验证边界。
ChatGPT Health: Trust Boundaries for Connected Medical Data

2026-07-25

最佳实践 #ChatGPT Health#Medical Records#Health Data Privacy#HIPAA#Apple Health#AI Safety#EHR#Data Governance
ChatGPT Health can read connected medical records across conversations. Evaluate its permissions, memory, deletion, HIPAA, and verification boundaries
OpenAI Presence：企业 Agent 平台的责任

2026-07-24

解读 #OpenAI Presence#企业AI#AI Agent#Agent治理#人工升级#Agent评测#AI运营
OpenAI Presence 把策略、批准动作、仿真、评测、人工升级和受控更新连成生产闭环。本文拆解平台应该负责什么，以及企业不能外包哪些控制权。
Previous Next