Claude Memory Kit

Your Claude agent remembers everything. Across sessions. Across projects. Zero setup.

The Problem

Every new Claude session starts from zero. Yesterday's decisions, last week's research, the bug you fixed three days ago — gone. You waste the first 10 minutes re-explaining what Claude already knew.

Claude Memory Kit fixes this in 3 commands. No API cost. Runs on your existing subscription.

Get Started

git clone https://github.com/awrshift/claude-memory-kit.git my-project
cd my-project
claude

That's it. Claude sets everything up and asks a few questions (your name, project name, language).

[!TIP] Type /tour after setup — Claude walks you through the system using your actual files.

---

Before and After

| | Without Memory Kit | With Memory Kit | |---|---|---| | New session | Starts from zero. "What project is this?" | Knows your project, last session, current tasks | | After 10 sessions | Nothing accumulated | Searchable wiki of decisions, patterns, lessons | | Multiple projects | Total chaos | Each project has its own context, automatically | | Context compression | Silently loses everything | Hook blocks Claude until it saves first | | Next day | "Remind me what we did yesterday" | Agent already knows — injected at session start |

---

Your Daily Workflow

Three steps. That's your entire workflow:

1. Open a session

Claude automatically loads your context — project state, recent decisions, knowledge wiki. You don't do anything.

2. Work normally

Talk to Claude. Build features. Fix bugs. Research. Safety hooks run invisibly in the background — they checkpoint your progress every ~50 exchanges and before context compression.

3. Close the day

When you're done, type /close-day. Claude scans everything you changed today and writes a daily summary. Takes 30 seconds.

That's it. Tomorrow's session starts exactly where you left off.

---

Knowledge Compounds Over Time

Every /close-day feeds into a growing knowledge base:

• Day 1–7: Basic patterns, project structure, your preferences
• Week 2–4: Decisions accumulate. Agent starts referencing past reasoning
• Month 2+: Full wiki of concepts, cross-project connections, lessons learned

Run /memory-compile periodically to structure daily logs into searchable wiki articles. Or don't — the daily logs alone give you 80% of the value.

---

What You Get

| What | How it helps | |------|-------------| | Persistent memory | Patterns and decisions survive between sessions | | Multi-project support | Each project has its own backlog and context | | Session handoff | next-session-prompt.md — "pick up exactly here" | | Knowledge wiki | Structured articles with search, built from your daily work | | Safety hooks | Agent can't lose context during compression or long sessions | | Commands | /memory-compile, /memory-lint, /memory-query (in .claude/commands/) | | Skills | /close-day, /tour (in .claude/skills/) |

Everything is plain Markdown files. No database. No external services. git checkout recovers anything.

---

FAQ

No. You talk to Claude in plain language. "Read the marketing plan and draft three emails" works perfectly.

The kit is free and open source. You need a Claude Pro or Max subscription (which you probably already have). No extra API cost.

Yes. Everything stays on your computer in plain text files.

Yes. During setup, tell Claude you have existing code. It analyzes the structure and sets up context around it.

Nothing bad happens. Your in-session saves (patterns, tasks, handoff notes) are already captured by safety hooks. /close-day adds a richer daily summary on top — nice to have, not critical.

Everything is in git. git checkout .claude/memory/ rolls back instantly. Or run /memory-lint to auto-fix structural issues.

---

Project Structure

SKILL.md              ← Skill entry point (for aggregators)
CLAUDE.md             ← Agent brain
README.md             ← You are here
ARCHITECTURE.md       ← Full technical details
skills/               ← Aggregator-facing skill index (symlinks into .claude/skills/)
├── close-day/SKILL.md → ../../.claude/skills/close-day/SKILL.md
└── tour/SKILL.md     → ../../.claude/skills/tour/SKILL.md
knowledge/            ← Wiki articles (grows over time)
projects/             ← Per-project backlogs
context/              ← Session handoff
daily/                ← Daily summaries
.claude/
├── memory/MEMORY.md  ← Hot cache (~200 lines)
├── memory/scripts/   ← Pipeline: compile, lint, query, flush, config
├── hooks/            ← 5 hooks (context injection + safety nets)
├── rules/            ← Domain conventions (see _example.md.disabled)
├── commands/         ← Slash commands (/memory-compile, /memory-lint, /memory-query)
└── skills/           ← Skills runtime source (/close-day, /tour)

Runtime reads skills from .claude/skills/. The root skills/ directory exists for Claude Code skill aggregators that scan repository roots — kept in sync via symlinks so edits to one reflect in the other.

How It Works (for the curious)

You interact with 3 things: your project files, slash commands, and Claude. Everything else is automatic.

Under the hood: a session-start.py hook injects your knowledge wiki index at startup. Safety hooks checkpoint progress automatically. /close-day synthesizes daily articles. /memory-compile transforms them into structured wiki entries.

Full architecture details: ARCHITECTURE.md

---

Provenance

Built on ideas from Andrej Karpathy (LLM knowledge base patterns) and Cole Medin's claude-memory-compiler. Simplified for daily CLI use — auto-flush replaced with deliberate /close-day synthesis after real-world testing showed ~50% failure rate in background automation.

700+ real sessions across 7 projects. This is what survived.

Contributing

Issues and PRs welcome. See CONTRIBUTING.md.

License

MIT — see LICENSE.