magnum-memory

---

😖 The problem

When Claude Code compacts a long conversation, the summary keeps what you're doing — but reliably drops the precise things you figured out: exact file paths, config values, why a decision was made, the alternatives you rejected, the dead-ends you already ruled out.

So you re-explain. You re-decide. You re-discover. Over a long session — or several — that loss compounds into wasted time and repeated mistakes.

✨ Why magnum-memory

• 🎯 It remembers the right things. Claude curates the durable specifics into a tiny file — not a noisy transcript dump.
• ♻️ It comes back automatically. After every compaction, the saved context is re-injected — no manual step.
• 🪶 It's featherweight. Pure Bash + Markdown. No Node, no MCP server, no database — nothing to install or maintain.
• 🔒 It's private by default. One per-project, gitignored file that never leaves your machine.

⚡ Quick start

/plugin marketplace add KashifManzer/magnum-memory
/plugin install magnum-memory@magnum-memory-marketplace

That's it — Claude now keeps a memory and restores it after compaction. Prefer just the skill? Install it in one line:

npx skills add https://github.com/KashifManzer/magnum-memory --skill magnum-memory

🧩 How it works: A + B

The core idea: store only what compaction forgets.

• A — the memory file (.claude/memory/CONTEXT.md): the durable specifics compaction drops. Personal and gitignored (never shared with teammates).
• B — the compaction summary: the live narrative Claude keeps automatically.

A and B are deliberately complementary — near-zero overlap. Rejoined after a compaction, they reconstruct the full working picture. magnum-memory only ever writes what B would lose, so the context it brings back stays small and high-signal.

🚀 Features

🧠 Curated, not captured	Claude writes what matters (decisions + rationale, dead-ends, key facts) — not a transcript dump
♻️ Automatic re-injection	A SessionStart hook restores the Current State after every compaction and on resume
⏰ Checkpoint nudge	A UserPromptSubmit hook reminds Claude to save when memory goes stale
🔎 History recall	/recall <words> searches your Checkpoint Log + Archive
🪶 Zero dependencies	Bash + Markdown; no Node, MCP, or DB
🔒 Local & private	Per-project, gitignored, never transmitted
🛟 Fail-safe	Hooks never block a prompt or compaction

📦 Install

Quick (skill only)

npx skills add https://github.com/KashifManzer/magnum-memory --skill magnum-memory

Installs the magnum-memory skill into .claude/skills/. Claude maintains .claude/memory/CONTEXT.md; you bring it back by reading it (or asking Claude to) when you resume a session or right after a compaction.

Full (plugin — adds automatic re-injection)

/plugin marketplace add KashifManzer/magnum-memory
/plugin install magnum-memory@magnum-memory-marketplace

Adds the SessionStart, PreCompact, and UserPromptSubmit hooks plus the /checkpoint and /recall commands on top of the skill. The hooks automatically re-inject the Current State after every compaction and on new/resumed sessions — no manual step.

	Quick (skill)	Full (plugin)
Maintains the memory file	✅	✅
/checkpoint & /recall commands	—	✅
Automatic re-injection after compaction	manual (read the file)	✅ via hooks
Checkpoint nudge when stale	—	✅

📖 How to use

1. Install once

Full experience (recommended):

/plugin marketplace add KashifManzer/magnum-memory
/plugin install magnum-memory@magnum-memory-marketplace

(Or skill-only: npx skills add https://github.com/KashifManzer/magnum-memory --skill magnum-memory.)

2. Just work — memory builds itself

Use Claude Code normally. At meaningful moments (a decision, a root cause, a dead-end, a key path/config), Claude appends to .claude/memory/CONTEXT.md. You don't have to do anything. A file in flight looks like: