What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
CLAUDE.md is the place for instructions your team wants Claude Code to follow repeatedly; Claude Code’s auto memory is where it can retain useful learnings, such as corrections, preferences, and project context it cannot infer from the code. Both are context, not enforcement. Put team conventions and commands in a reviewed instruction file, let auto memory capture relevant learnings, and use settings or hooks for restrictions that must be enforced.
What is the difference between CLAUDE.md and auto memory?
They solve different problems. CLAUDE.md is a human-maintained briefing: it tells Claude Code how a project works and how the team wants recurring work done. Auto memory is Claude Code’s notebook for useful details accumulated over work, including user preferences, corrections, and patterns that are not evident from the repository.
![Claude AI for Beginners Bible: [5 in 1] The Ultimate Guide to Automate Your Work, Save Hours Every Week, and Use AI for Real-World Results](https://m.media-amazon.com/images/I/415+fSJacsL._SL160_.jpg)
Both are loaded as context in Claude Code. The documentation says auto memory loads up to its first 200 lines or 25 KB. That is a loading limit, not a recommendation to fill the file. Anthropic describes the distinction plainly: “Both are loaded at the start of every conversation. Claude treats them as context, not enforced configuration.” Claude Code memory documentation
What belongs in CLAUDE.md?
Write down stable, recurring facts that materially improve work whenever Claude operates in the file’s scope. Think of it as a concise project briefing, not a repository encyclopedia or a transcript of previous sessions.
#1 Best Overall
- Commands: reliable build, test, lint, and development commands, including prerequisites or important flags.
- Architecture: where major components live and how they relate, when that structure is not obvious from the code.
- Conventions: naming, formatting, testing, or review practices that the team consistently expects.
- Constraints: project-specific boundaries and decisions that should guide ordinary work.
- Known gotchas: recurring pitfalls, such as a test that requires a service or a generated file that should not be edited directly.
Keep guidance actionable and current. Avoid duplicating material already maintained in documentation, listing obvious file-tree facts, recording every change in a changelog, or adding aspirational rules with no practical direction. Anthropic’s Help Center suggests keeping the file under roughly 200 lines and periodically removing stale entries; that is a heuristic, not a hard product limit. Anthropic Help Center guidance on Claude Code memory
What should auto memory store?
Auto memory is better suited to observations the agent learns through repeated work than to rules the team should explicitly maintain. Useful examples include a correction after Claude repeatedly makes the same mistaken assumption, a user preference it should remember, or project context that cannot be reliably derived from the repository.
Claude Code’s current documentation says auto memory is per repository and shared across worktrees. Because it accumulates over time, inspect it occasionally: an old preference or a one-off workaround can become misleading. Promote a durable team rule into CLAUDE.md when it should be explicit and reviewed, rather than relying on an automatically accumulated note.
Where should personal, file-specific, and procedural guidance go?
Choose the location by scope and purpose. Claude Code supports managed organization instructions, user-wide instructions, project instructions, and project-local instructions. Files above the working directory load at launch; instructions in subdirectories are loaded when Claude reads in those directories. CLAUDE.local.md is for unshared project preferences and should be gitignored.
| Information | Best fit | Why |
|---|---|---|
| Shared project commands and conventions | Project CLAUDE.md |
Team-wide guidance that should be reviewed and versioned with the project. |
| Preferences across projects | ~/.claude/CLAUDE.md |
User-wide instructions rather than repository policy. |
| Personal preference for one project | Gitignored CLAUDE.local.md |
Project-specific guidance that should not be shared with collaborators. |
| Rules for selected files or directories | .claude/rules/ with paths frontmatter |
Conditional guidance that should apply only to matching paths. |
| Reusable multi-step procedure | Skill | A procedure that need not occupy the always-loaded project briefing. |
| Learned correction or inferred pattern | Auto memory | A useful accumulated learning rather than a team-authored instruction. |
These distinctions help prevent the project briefing from growing into an always-loaded manual. Put a procedure in a skill and a file-specific rule in a path-scoped rule instead of making every conversation carry guidance that applies only sometimes. Claude Code memory documentation
Does Claude Code read AGENTS.md?
It can, but the behavior is conditional. By default, Claude Code reads AGENTS.md when no CLAUDE.md or CLAUDE.local.md exists in the working directory or above it. If both kinds of file exist, Claude Code reads the CLAUDE.md files by default; you can import AGENTS.md from CLAUDE.md or change the Project instructions setting to configure the choice. Directly reading AGENTS.md requires Claude Code v2.1.277 or later, and the documentation notes that support may be unavailable in some sessions. Check the current Claude Code memory documentation for version and setting details.
Rank #4
Are CLAUDE.md and auto memory security controls?
No. They influence behavior but do not guarantee compliance. Do not put a must-follow security or compliance restriction only in CLAUDE.md or auto memory. Use an enforced setting that blocks the relevant tools, commands, or paths, or a PreToolUse hook when an action must be stopped regardless of the model’s choice. Claude Code settings and Claude Code hooks
Writable memory also deserves care. Anthropic warns that malicious or untrusted content can prompt an agent to write harmful material into a Managed Agents memory store, which later sessions may read as trusted context. Use read-only access for reference material that should not change. For a custom API Memory tool implementation, validate paths and prevent traversal beyond the intended memory directory. Managed Agents memory documentation and API Memory tool documentation
Do not confuse Claude Code auto memory with Anthropic platform memory
“Agent memory” can refer to distinct systems. Claude Code auto memory is a Claude Code feature. Anthropic Managed Agents memory stores are a separate platform capability: workspace-scoped collections of text documents attached when a session is created. Agents access them through their tools; stores can be read-write or read-only, and changes create immutable versions. In self-hosted sandboxes, local copies are synchronized rather than mounted live. Anthropic labels this capability beta and documents the agent-memory-2026-07-22 beta header. The documentation states that each memory is capped at 100 kB (about 25,000 tokens) and each store at 10,000 memories. Those limits apply to Managed Agents, not Claude Code auto memory. Managed Agents memory documentation
The API Memory tool is different again: it provides an interface for requested memory operations, while the application supplies and manages the storage. That storage could be local files, a database, cloud storage, or encrypted files; the application is responsible for safeguards such as path validation. Anthropic describes progress, feature-checklist, and session-end update files as one pattern for recovering context in long-running work. API Memory tool documentation
How to decide where a new detail belongs
- Ask who should own it. If the team should specify and review the information, put it in the appropriate instruction file. If it is a useful preference or learning accumulated through work, auto memory may be a better fit.
- Match the scope. Use a shared project file for team guidance, a user-level file for cross-project preferences, a local file for private project preferences, and path-scoped rules for selected files.
- Separate rules from procedures. Keep a recurring instruction in the relevant instruction file; put a longer reusable workflow in a skill.
- Check whether it will stay useful. Add only information that should influence future work in that scope. Remove or revise stale or contradictory guidance.
- Enforce critical restrictions outside context. If a rule must be guaranteed, configure an enforced setting or an appropriate hook rather than relying on instructions or memory.
For memory systems outside Claude Code, compare who authors updates, what they store, their scope, when and how much content enters context, whether changes are versioned, what read/write access exists, and how people can inspect, correct, expire, or delete entries. Those properties vary by product; Managed Agents’ versioning and access controls should not be assumed of other systems.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
