Group of 3 You'll notice
Project instruction-file loading now caches unchanged content, discovers nested CLAUDE.md/AGENTS.md files, and 'none' mode keeps org-managed instructions
What
The pipeline that loads CLAUDE.md and related project-instruction files was reworked in several ways:
- Loading now computes the set of instruction files first and compares it to the previous version, reusing cached CLAUDE.md text when the instruction files haven't changed, instead of always recomputing.
- Instruction-file discovery was rewritten to resolve a session root and walk parent directories for nested CLAUDE.md/AGENTS.md files, deduping files by both path and trimmed content across project, local, user, and memory sources.
- The
projectInstructionssetting'snonemode now behaves differently: it drops only the project's own and your personal instruction files, while your organization's managed CLAUDE.md and memory still load. Previously,nonedropped the managed CLAUDE.md too.
Why
These changes make instruction-file loading faster (via caching), more thorough (finding nested files in parent directories), and change what none actually excludes — organizations relying on managed CLAUDE.md should note it now stays loaded even when a user sets projectInstructions to none.
Names in the bundleCLAUDE.mdAGENTS.md
The entry above is what we published on the day. These lines were added later, as Anthropic's own pages caught up, and they sit beside the original rather than replacing it.
Added since
A small documentation edit on Explore the context window touched a line naming AGENTS.md after this was published.
* **Before you type anything**: CLAUDE.md, auto memory, MCP tool names, and skill descriptions all load into context. [AGENTS.md files](/docs/en/memory#agents-md) can load too, on their own or alongside CLAUDE.md. Your own setup may add mo…context-window see the edit
Added since
A small documentation edit on Explore the context window touched a line naming CLAUDE.md after this was published.
* **Before you type anything**: CLAUDE.md, auto memory, MCP tool names, and skill descriptions all load into context. [AGENTS.md files](/docs/en/memory#agents-md) can load too, on their own or alongside CLAUDE.md. Your own setup may add mo…context-window see the edit
Confirmed since
Anthropic's documentation has since written up AGENTS.md, on Explore the .claude directory.
tips: ['Target under 200 lines. Longer files still load in full but may reduce adherence', <>CLAUDE.md loads into every session. If something only matters for specific tasks, move it to a <A href="/docs/en/skills">skill</A> or a path-scope…claude-directory see the edit
Confirmed since
Anthropic's documentation has since written up CLAUDE.md, on Explore the .claude directory.
tips: ['Target under 200 lines. Longer files still load in full but may reduce adherence', <>CLAUDE.md loads into every session. If something only matters for specific tasks, move it to a <A href="/docs/en/skills">skill</A> or a path-scope…claude-directory see the edit
One source agreesOne thing we can check says the same as this entry.
Anthropic's documentation agrees
Anthropic's documentation has since written up CLAUDE.md, on Explore the .claude directory.