{
  "id": 3099581,
  "title": "My agent.md to improve LLM-assisted code quality",
  "url": "https://urgent.news/2026/08/23/my-agent-md-to-improve-llm-assisted-code-quality-3099581",
  "topic": "ai",
  "section": "AI",
  "published": "2026-08-23T17:59:52.000Z",
  "source": {
    "name": "Hacker News Best",
    "slug": "hacker-news-best",
    "url": "https://fabiensanglard.net/agent.md/index.html"
  },
  "original_language": "en",
  "account": "In January 2026, the author attempted to use large language models (LLMs) to assist with coding, specifically for an mDNS implementation in Rust named libadbmdns. The initial results were unimpressive, as the code produced did not even compile. Upon revisiting LLMs in January 2026, the author found that the models were able to write a complex indexed-binary heap class and even identify an obscure bug in the polling crate due to the Windows IOCP implementation. However, the code quality was poor, with no comments, structure, or production-level standards.\n\nIn March 2026, the author tried using agentic IDEs like Antigravity and VS Code's Claude Code plugin. This allowed for iteration over staged code, with suggestions such as avoiding magic numbers, adding comments, and using short function names. The code quality improved significantly, but the process was still tedious, as the author had to repeat the same suggestions multiple times. To address this, the author created a file named agent.md, which was injected into the prompt at the start of each coding session. This file contained the author's coding style preferences, allowing for more fine-tuned code generation.\n\nThe author also addressed the issue of \"context dilution\" or \"attention dilution\" in LLMs, a phenomenon where models pay less attention to middle instructions as the context grows. To minimize this impact, the author suggested either not opening an editor with every new rule or updating the agent.md file with new instructions. Overall, the author found that while LLMs can be helpful, they still require verification and iteration, with the author focusing more on architecture and design rather than code style.",
  "summary": "Article URL: https://fabiensanglard.net/agent.md/index.html Comments URL: https://news.ycombinator.com/item?id=49410932 Points: 390 # Comments: 169",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Hacker News",
        "title": "My agent.md to improve LLM-assisted code quality",
        "url": "https://urgent.news/2026/08/23/my-agent-md-to-improve-llm-assisted-code-quality",
        "published": "2026-08-23T17:59:52.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}