Claude Token Efficient - CLAUDE.md

2026-05-14 · source · source ↗ · raw capture

A brief GitHub file containing guidelines for Claude AI agent behavior and interaction protocols.

Key Guidelines

Six core operational principles:

  1. File Reading Strategy: Review existing files before writing new content. Avoid re-reading unless modifications have occurred.
  2. Communication Style: Provide thorough reasoning while keeping output concise and focused.
  3. Large File Handling: Skip files exceeding 100KB in size unless they're essential to the task.
  4. Tone Requirements: Eliminate unnecessary pleasantries and decorative elements from responses.
  5. Formatting Standards: Exclude emojis and em-dashes from all outputs.
  6. Accuracy Mandate: Never make assumptions about APIs, software versions, command flags, commit identifiers, or package names. "Do not guess APIs, versions, flags, commit SHAs, or package names."

These guidelines establish efficiency-focused standards for Claude interactions, emphasizing verification over assumption, directness over elaboration, and practical utility over presentation flourish.