Eye(FR) → Mill(skill) → Library(/app)
正典：已过质量门 / 已成文的 library 知识（has_ka）。人读与 AI 检索默认只看这里。
围绕「为了Claude不被封，把时区语言都改了，如果再被封我就把电脑吃了！」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「突发！OpenAI 大规模重组，总裁 Brockman 夺权挂帅」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Ask HN: How do you catch regressions when you change your AI agent's p」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Ask HN: Does anyone use codex to review Claude's code? What're your ex」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Ask HN: Onboarding LLMs or Students?」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「OpenAI Models in OpenClaw, Done Right」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Kicking the Tyres on Harbor for Agent Evals」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Best Realtime Voice Agent in 2026」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Anthropic investigating claim of unauthorised access to Mythos AI tool」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「ScyllaDB MCP Server: What AI-Native Developer Distribution Looks Like」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Show HN: Setup a box on demand and run your agent on it remotely」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Show HN: Incultnito Studio – Fulfillment MCP for apparel brands」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「We built an agent specialized for CI (using the 3 Claude models)」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Show HN: SwarmWright, structured multi-agent AI defined in markdowns」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「AWS user gets $30K Claude bill after cost alert misses it」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「Run Hermes on a VPS without Leaking your API keys [video]」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
围绕「N8n-MCP: AI assistants can now search and build n8n workflows」：从 agent/MCP/harness/成本与工程化视角抽取可迁移结论，并落到可执行动作。
Grok vs. ChatGPT vs. Gemini Comparison 2026: Complete Guide (Tested)
LocalVibe – Pure-Rust local AI stack with MCP, in one binary (Apple Silicon)
Show HN: AI/ML benchmark for local LLM inference and XGBoost training on GPU/CPU
Show HN: Stoic AgentOS - Open-source operating system for AI agent fleets
Zerostack – Tiny Rust Coding Agent in 8MB of RAM
Show HN: Strava for AI coding – analytics on your Copilot/Claude/Codex usage
Ane: CLI editor that uses LSPs to let agents explore/edit code with fewer tokens