{
  "id": 15679219,
  "title": "Keep your CLAUDE.md under 40 lines so Claude Code actually follows it",
  "url": "https://urgent.news/2026/10/20/keep-your-claude-md-under-40-lines-so-claude-code-actually-follows-it",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-20T01:10:34.000Z",
  "source": {
    "name": "Dev.to",
    "slug": "dev-to",
    "url": "https://dev.to/vildandenai/keep-your-claudemd-under-40-lines-so-claude-code-actually-follows-it-2hhf"
  },
  "original_language": "en",
  "account": "Many CLAUDE.md files seem to become extensive documents, containing architecture notes, style essays, and multiple pasted READMEs. This can lead to confusion about why agents may ignore the crucial one rule that matters. CLAUDE.md is loaded at the beginning of a session, so each line competes for attention with the actual task at hand. Effective house rules are concise and to the point, offering clear guidelines rather than lengthy explanations. A successful structure involves # House rules, followed by Before you edit instructions and Ask before adding a new dependency reminders. Specific Commands to test, lint, and avoid certain actions like force pushing or deleting files should be listed. The Done criteria include passing tests and lint checks locally, clearly stating changed files and reasons. The Style section emphasizes matching surrounding code without making small, isolated changes. Key rules include keeping commands over prose, using test commands, and enforcing rules through hooks or settings. Reference material should be linked to external documents rather than loading it every session. Regular pruning of the file is recommended, removing rules that have never impacted a diff. Check the memory documentation for your specific Claude Code version, as nested and user-level file loading can vary between releases. This concise house-rules file, along with the allow-list and safety hooks, is distributed with the Vildanden Agent Config. The sample provided is sufficient to start implementing this pattern.",
  "summary": "Most CLAUDE.md files I see grow into a wiki: architecture notes, style essays, a changelog, three pasted READMEs. Then people wonder why the agent ignores the one rule that mattered. CLAUDE.md is loaded into context at the start of a session, so every line competes with your actual task. Short, concrete house rules beat long explanations. A shape that works # House rules ## Before you edit - Read…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 3,
    "also_reported_by": [
      {
        "outlet": "XDA Developers",
        "title": "I tried DeepSeek's agent harness, and it's not the Claude Code clone I expected",
        "url": "https://urgent.news/2026/10/19/i-tried-deepseeks-agent-harness-and-its-not-the-claude-code-clone-i",
        "published": "2026-10-19T14:30:19.000Z"
      },
      {
        "outlet": "The New Stack",
        "title": "One sentence changed Claude Code’s overreach rate from 0% to 17.1%",
        "url": "https://urgent.news/2026/10/19/one-sentence-changed-claude-codes-overreach-rate-from-0-to-17-1",
        "published": "2026-10-19T18:36:02.000Z"
      }
    ]
  },
  "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."
}