All corrections
X April 2, 2026 at 04:07 PM

x.com/burkov/status/2039588435401302224

1 correction found

1
Claim
your CLAUDE.md file is re-injected on every single **turn** of the conversation.
Correction

Anthropic’s docs describe `CLAUDE.md` as startup/session context, not something injected on every turn. They say it is loaded at the start of each conversation or session, with some scoped files loading only on demand.

Full reasoning

Anthropic’s official Claude Code documentation contradicts the claim that a user's CLAUDE.md is re-injected on every single turn.

What the docs say instead:

  • CLAUDE.md and auto memory are "loaded at the start of every conversation."
  • CLAUDE.md files are read "at the start of every session."
  • Ancestor CLAUDE.md files are "loaded in full at launch."
  • Subdirectory CLAUDE.md files are not loaded every turn; they "load on demand when Claude reads files in those directories."
  • Path-scoped rules also do not trigger constantly; Anthropic says they trigger when Claude reads matching files, "not on every tool use."

Anthropic’s API context-window docs also describe normal multi-turn behavior as the conversation history being carried forward turn by turn: previous user/assistant turns are preserved in the context window as the conversation grows. That is a different description from saying CLAUDE.md is freshly re-injected on every turn.

So the post overstates Claude Code’s behavior. According to Anthropic’s own documentation, CLAUDE.md is startup/session context, with some additional scoped files loaded only when relevant—not a blanket per-turn reinjection of the file.

2 sources
  • How Claude remembers your project - Claude Code Docs

    Claude Code has two complementary memory systems. Both are loaded at the start of every conversation... CLAUDE.md files are markdown files... Claude reads them at the start of every session... CLAUDE.md files in the directory hierarchy above the working directory are loaded in full at launch. CLAUDE.md files in subdirectories load on demand when Claude reads files in those directories.

  • Context windows - Claude API Docs

    As the conversation advances through turns, each user message and assistant response accumulates within the context window. Previous turns are preserved completely... Input phase: Contains all previous conversation history plus the current user message.

Model: OPENAI_GPT_5 Prompt: v1.16.0