Claude Anchor
for Claude Code command line
A persistent memory, rules, and behavioral context framework for AI-assisted development with Claude Code.
Important
This is open source software maintained by ONE individual in their free time.
While the templates are tested and reviewed before releases, you should ALWAYS have:
- A proper GitHub/Git strategy in place
- Regular backups of your work
- Version control before running ANY script from the internet
DISCLAIMER: This software is provided "as is" without warranty of any kind. Super Basic Studio and the maintainers of Claude Anchor are not responsible for any issues, data loss, or damages that may occur from using this tool. By using Claude Anchor, you acknowledge that you run it at your own risk.
Give Claude persistent memory, enforceable rules, and behavioral consistency across sessions. Claude Anchor is the behavioral brain — it tells Claude how to think and behave in your project.
- Quick Start
- What It Does
- Templates
- Session Load Order
- Template Overview
- Customization Guide
- Naming Conventions
- Companion Project: Claude Conductor
- Contributing
- Privacy & Security
- License
- npm package —
npx claude-anchor - Claude Conductor — Companion codebase documentation framework
- Changelog
- Security Policy
# Copy 3 essential templates into your project
npx claude-anchor
# Copy all 11 templates
npx claude-anchor --full
# Copy into a specific directory
npx claude-anchor ./my-project
# Overwrite existing files
npx claude-anchor --force# Clone the templates
git clone https://github.com/superbasicstudio/claude-anchor.git
# Copy the templates you need into your project
cp claude-anchor/templates/_GOLDEN-RULES.md ./your-project/
cp claude-anchor/templates/_TODOS.md ./your-project/
cp claude-anchor/templates/_LONG-TERM-MEMORY.md ./your-project/- Edit each
.mdfile, replacing[PLACEHOLDERS]with your project-specific content - Customize
_GOLDEN-RULES.mdwith your project's constraints - Claude reads these files automatically at the start of each session
Claude Anchor provides a structured template system that gives Claude:
- Voice & tone — Control Claude's personality, attitude, vocabulary, and communication vibe. Loaded first so it colors everything.
- Persistent memory — Long-term knowledge that survives across sessions (user preferences, project conventions, important decisions)
- Session continuity — RAM for single-session crash recovery, short-term memory for multi-session context across 4-10 sessions
- Enforceable rules — Immutable constraints ("NEVER do X") that Claude must follow, read twice per session to prevent context decay
- Design preferences — Visual design rules: lighter hover colors, WCAG accessibility, typography, flat iconography, UX principles
- Conversation preferences — Standardized output formatting, verbosity levels, and communication style
- Lessons learned — A living record of past mistakes and their fixes, so Claude doesn't repeat them
- Task tracking — Priority-based TODOs with blockers and dependencies
Without persistent context, Claude:
- Loses crucial knowledge between sessions
- Repeats the same mistakes
- Forgets your preferences and conventions
- Can't resume interrupted work or recover from crashes
- Has no enforceable behavioral constraints
- Has no consistent personality or communication style
- Ignores your design system and accessibility requirements
Anchor fixes all of this with a structured set of markdown templates that Claude reads at session start.
your-project/
├── CLAUDE.md # Your existing project docs
├── _GOLDEN-RULES.md # Immutable rules
└── _TODOS.md # Active tasks
your-project/
├── CLAUDE.md # Project context (with load order block)
├── _RAM.md # Single-session volatile memory (crash recovery)
├── _VOICE-AND-TONE.md # Personality, attitude, language style (loaded FIRST)
├── _GOLDEN-RULES.md # Immutable rules (read twice per session)
├── _TODOS.md # Active tasks and priorities
├── _LESSONS-LEARNED.md # Past mistakes and fixes
├── _CONVERSATION-PREFERENCES.md # Output formatting and communication style
├── _DESIGN-PREFERENCES.md # Visual design, accessibility, UX rules
├── _LONG-TERM-MEMORY.md # Persistent memory (NEVER delete)
├── _SHORT-TERM-MEMORY.md # Multi-session temporary context (4-10 sessions)
└── _SYSTEM_ARCHITECTURE.md # Technical diagrams and system design
The CLAUDE.md template includes a mandatory startup block. When Claude enters a project with Anchor files, it reads them in this exact sequence:
0. _RAM.md <- IF EXISTS: recover interrupted session state
1. _SHORT-TERM-MEMORY.md <- IF EXISTS: active multi-session context
2. _VOICE-AND-TONE.md <- Personality and language style
3. _GOLDEN-RULES.md <- BINDING rules — MUST FOLLOW every session
4. _TODOS.md <- Know what's pending
5. _LESSONS-LEARNED.md <- Avoid past mistakes
6. _LONG-TERM-MEMORY.md <- Persistent knowledge and preferences
7. _CONVERSATION-PREFERENCES.md <- Output formatting
8. _DESIGN-PREFERENCES.md <- Visual design and UX rules
9. _GOLDEN-RULES.md <- Re-read (reinforce)
10. CLAUDE.md <- Full project context
11. BEGIN conversation
Why is RAM step 0? If _RAM.md exists, the previous session was interrupted. Reading it first restores the exact working state before anything else loads.
Why is VOICE-AND-TONE first (after recovery)? Claude's personality should be established before it reads anything else. It colors how Claude processes every file after it and how it communicates throughout the session.
Why read GOLDEN-RULES twice? In long contexts, content read early can get "forgotten" as the context window fills. Re-reading critical rules last keeps them fresh and prevents behavioral drift during the session.
| Template | Purpose | Lifecycle | Priority |
|---|---|---|---|
_RAM.md |
Single-session volatile state — crash recovery | Auto-managed. Written continuously, deleted at session end. | RECOVERY |
_VOICE-AND-TONE.md |
Personality, attitude, vocabulary, response style, commit format | Permanent. Adjust to match your working style. | FIRST |
_GOLDEN-RULES.md |
Immutable constraints — things Claude must NEVER/ALWAYS do | Permanent. Updated when new rules are needed. | BINDING |
_TODOS.md |
Active tasks with priorities, blockers, and dependencies | Ongoing. Tasks move from pending to completed. | High |
_LESSONS-LEARNED.md |
Documented mistakes with root cause and prevention | Permanent. Add entries immediately when issues are discovered. | High |
_CONVERSATION-PREFERENCES.md |
Output formatting, colors, progress bars, verbosity | Permanent. Adjusted to match your communication style. | Medium |
_DESIGN-PREFERENCES.md |
Hover states, accessibility, typography, iconography, UX rules | Permanent. Adjust to match your design system. | High |
_LONG-TERM-MEMORY.md |
Persistent knowledge: user identity, system config, code style, decisions | NEVER delete. Accumulates over months/years. | High |
_SHORT-TERM-MEMORY.md |
Multi-session temporary context: active issues, in-progress work | Delete when all items resolved. Persists 4-10 sessions. | Conditional |
_SYSTEM_ARCHITECTURE.md |
ASCII diagrams, data flows, component maps, security model | On-demand reference. Updated when architecture changes. | Reference |
_LONG-TERM-MEMORY.md _SHORT-TERM-MEMORY.md _RAM.md
- NEVER delete - Persists 4-10 sessions - Single session only
- Accumulates over time - Active issues & work - Crash recovery
- User preferences, decisions - In-progress improvements - Written continuously
- System configuration - Delete when items resolve - Deleted at session end
- Always available - "What are we working ON?" - "What am I doing NOW?"
All template files use placeholder syntax:
[PROJECT_NAME] — Your project name
[DESCRIPTION] — Brief description
[DATE] — Current date
[PATH] — File/folder paths
[COMMAND] — CLI commands
- _VOICE-AND-TONE.md — Set this up first. It shapes everything Claude says. Pick your formality, humor, and energy levels.
- _GOLDEN-RULES.md — Be specific. Vague rules get ignored. Include "why" for each rule.
- _TODOS.md — Keep it current. Stale TODOs confuse Claude about priorities.
- _LESSONS-LEARNED.md — Add entries immediately when you discover gotchas. Future you will thank past you.
- _CONVERSATION-PREFERENCES.md — Include visual examples of your preferred output format.
- _DESIGN-PREFERENCES.md — Set your hover, accessibility, and icon rules once. Claude follows them everywhere.
- _LONG-TERM-MEMORY.md — Start with your identity, system environment, and code style. It grows naturally over time.
- _SHORT-TERM-MEMORY.md — Use for context that should persist across several sessions. Delete entries as they're resolved.
- _RAM.md — Claude manages this automatically. If you see one at session start, it means the last session was interrupted.
- _SYSTEM_ARCHITECTURE.md — ASCII diagrams are universally understood. Use them.
| Pattern | Meaning | Example |
|---|---|---|
ALLCAPS.md |
Core documentation | CLAUDE.md, TODOS.md |
_UNDERSCORE.md |
Supporting/reference docs | _SYSTEM_ARCHITECTURE.md |
HYPHENATED.md |
Multi-word names | GOLDEN-RULES.md, LESSONS-LEARNED.md |
The underscore prefix keeps supporting files visually grouped at the top of directory listings.
Two repos, two tools, complementary not competitive.
Anchor is the behavioral brain — it manages Claude's rules, memory, conversation preferences, and session continuity. It tells Claude how to think and behave.
Claude Conductor is the codebase brain — it documents your project's architecture, APIs, build systems, errors, and development history. It tells Claude what your project is.
Together they give Claude full context — what the project is AND how to work on it.
| Conductor | Anchor | |
|---|---|---|
| Focus | Codebase documentation | AI behavior and memory |
| Generates | ARCHITECTURE.md, BUILD.md, API.md, JOURNAL.md | _VOICE-AND-TONE.md, _GOLDEN-RULES.md, _DESIGN-PREFERENCES.md, _RAM.md, _LONG-TERM-MEMORY.md, _TODOS.md |
| Delivery | npx claude-conductor (automated CLI) |
Copy templates into project |
| Answers | "What is this codebase?" | "How should Claude behave?" |
A project can use one or both. They don't overlap — they stack.
We welcome contributions! See CONTRIBUTING.md for full details.
- Fork the repository
- Create a feature branch
- Make your changes and run
npm test - Submit a pull request
All PRs require passing CI checks (tests across Node 16/18/20 + CodeQL security scan) and a code review from @superbasicstudio before merging.
Please keep templates framework-agnostic — they should work with any project in any language.
Claude Anchor is a collection of markdown templates with a lightweight CLI installer. It:
- Templates are pure markdown — no executable code runs in your project
- The CLI (
npx claude-anchor) only copies template files — nothing else - Makes zero network requests (beyond npm install)
- Has no telemetry or analytics
- Runs entirely on your local machine
- Never collects or transmits any data
The _LONG-TERM-MEMORY.md template includes a warning to NEVER store actual passwords or secrets. Follow this guidance.
BSD 2-Clause (c) Super Basic Studio
Made with love by Super Basic Studio
claude, claude code, claude anchor, ai memory, ai context, persistent memory, rules framework, claude code framework, vibe coding, super basic studio
