KK Blog

Menu

Archives

2026 / 09

Audit Coding Agent Data Flows Before Login: Lessons from ZCode

2026-09-24

最佳实践 #Coding Agents#Repository Security#Data Flow Audit#ZCode#Source Code Privacy#AI Security#Harness Engineering
Audit coding agent data flows before login with canary repositories, auth-state diffs, filesystem traces, egress capture, and deletion evidence.
GPT-6 Prompt 缓存审计：九类失效原因

2026-09-23

最佳实践 #GPT-6#Prompt缓存#OpenAI API#缓存诊断#显式断点#API成本#LLM可观测性
用显式断点、九类失效原因、Usage 证据和包含写入费的成本模型，审计 GPT-6 Prompt 缓存。
Audit GPT-6 Prompt Caching: 9 Ways to Miss

2026-09-23

最佳实践 #GPT-6#Prompt Caching#OpenAI API#Cache Diagnostics#Explicit Breakpoints#API Cost#LLM Observability
Audit GPT-6 prompt caching with explicit breakpoints, nine cache-miss reasons, usage evidence, and a cost model that includes cache writes.
Meta Rebalancer：把策略、求解与调试分开

2026-09-22

最佳实践 #Meta Rebalancer#分配优化#局部搜索#混合整数规划#资源分配#约束求解#系统工程
Meta Rebalancer 展示了怎样用稳定合同连接业务策略、数学表达、求解器和调试工具。
Meta Rebalancer: Separate Policy, Search, and Debugging

2026-09-22

最佳实践 #Meta Rebalancer#Assignment Optimization#Local Search#Mixed Integer Programming#Resource Allocation#Constraint Solving#Systems Engineering
Meta Rebalancer shows how assignment systems improve when policy, expression, solving, and debugging share a stable contract.
AI 数学成果需要独立审查门禁

2026-09-22

最佳实践 #AI数学#独立审查#数学验证#科研治理#形式化验证#OpenAI#科学传播
AI 数学成果需要分别通过正确性、重要性与公开传播三道门，并明确保留最终决策责任。
AI Math Results Need an Independent Review Gate

2026-09-22

最佳实践 #AI Mathematics#Independent Review#Mathematical Verification#Research Governance#Formal Verification#OpenAI#Scientific Communication
AI math results need separate gates for correctness, significance, and public communication, with decision authority kept explicit.
AI Agent 机构记忆需要来源绑定

2026-09-22

最佳实践 #AI Agent记忆#机构记忆#Context Graph#数据溯源#RAG#知识图谱#Agent治理
可信的 Agent 机构记忆需要把每条可行动主张与来源、有效期、依赖关系和失效条件绑定。
AI Agent Memory Needs Source-Linked Claims

2026-09-22

最佳实践 #AI Agent Memory#Institutional Memory#Context Graph#Data Provenance#RAG#Knowledge Graph#Agent Governance
AI agent memory becomes trustworthy when every actionable claim keeps its source, validity, dependencies, and invalidation conditions.
补丁半衰期：AI 编程何时该重写

2026-09-21

最佳实践 #AI编程#软件重写#技术债#测试影响分析#Amdahl定律#Harness Engineering
补丁半衰期把重写争论变成可复算决策：记录同类故障复发时间、修补成本、迁移成本，以及验证和回退能力。
Previous Next