Context Memory — Your Knowledge Memory for Claude

01

Claude forgets everything. This skill changes that.

Normally, Claude starts every session from scratch. Your architecture decisions, bug fixes, research findings — all gone. Context Memory changes this fundamentally.

Just tell Claude what matters. The skill stores it in a Context TreeA hierarchical tree structure that organizes your project knowledge — by type, relevance, and relationships. — and next time, everything is there again.

── Without Context Memory ──
Session 1: "We're using PostgreSQL" → stored in Claude's head
Session 2: "What database do we use?" → ❌ No idea

── With Context Memory ──
Session 1: "Context Memory: Remember — we use PostgreSQL" → ✅ dec-001 saved
Session 2: "Context Memory: What database do we use?" → ✅ PostgreSQL (decision, critical)

How do you invoke the skill?

Just talk normally, Claude detects it automatically

"Context Memory: …" followed by your request

Enter a special slash command

02

Everything that matters — from conversations and files

You can dictate knowledge to Claude — or feed it entire files. The skill extracts the contents and organizes them automatically.

💬

From Conversation

"Context Memory: Remember that we chose PostgreSQL."

📄

From PDFs & Word

Upload a file and say "Add this to the knowledge base."

📊

From Excel & CSV

Columns, data types, and samples are stored automatically.

🎯

From Presentations

PowerPoint slides are extracted individually as knowledge units.

💻

From Code

Claude analyzes your codebase and detects structure, dependencies, and tests.

🌐

Markdown, HTML, JSON…

Over 15 formats. Claude automatically decides how to split.

Learn more All formats and splitting strategies in detail → PDF by pages, Word by sections, Excel by sheets — when to use what

03

Ask questions — get answers

Agentic SearchThree-stage search: First quick match on tags and titles, then full-text search in content, then relation enrichment. searches your entire knowledge — not just titles, but also content and connections.

Your Question

→

Quick Search
Tags & Titles

→

Deep Search
Full-text

→

Results
+ Relations

You: "Context Memory: What do we know about the database?"

Claude: 🔍 3 results:
  1. ⚡ [dec-001] PostgreSQL as database (critical, high confidence)
     🕸️ Required by: REST API
  2. ⚡ [les-001] N+1 query solved (high, medium confidence)
     🕸️ Fixes: User endpoint bug
  3. ⚡ [pat-001] Repository Pattern (medium, high confidence)
     🕸️ Implements: PostgreSQL access

What makes the search special?

It only searches through titles of stored entries

It uses a vector database for semantic similarity

It searches multi-stage — tags, full-text, and shows connections between results

Learn moreThree-stage search in detail →Quick Search, Deep Search, relation enrichment

04

Connect knowledge — the Knowledge Graph

Individual knowledge pieces are good. But the connections between them are what's truly valuable. Context Memory links your knowledge via 22 relation typesDependencies, causality, contradictions, replacements, workflow sequences — each relationship has a direction and a reverse..

You: "Context Memory: The Stanford study confirms the interview findings."
Claude: ✅ Relation: arch-003 validates les-001

You: "Context Memory: The McKinsey report contradicts Stanford."
Claude: ✅ Relation: les-002 contradicts arch-003

You: "Context Memory: How does McKinsey connect to my book structure?"
Claude: 🛤️ Path (3 hops):
McKinsey → contradicts → Stanford → validates → Interview → documents → Book Structure

🔗

Structural

depends on, part of

⚡

Causal

caused by, fixed by

✅

Knowledge

validates, contradicts, documents

🔄

Lifecycle

supersedes, extends, blocks

Learn more All 22 relation types and graph features → Path finding, context traversal, Mermaid diagrams

05

Knowledge has an expiry date — and a certainty level

Not every piece of information is equally certain. And "Revenue 28B" and "Revenue 34B" aren't contradictions — they're a time seriesEach node can carry a source date, a validity window, and temporal confidence. This makes chronological changes visible instead of treating them as errors..

Temporal Dimension:
"AI adoption 12%" ← McKinsey, Q1 2025, confidence: explicit
"AI adoption 23%" ← Accenture, Q4 2025, confidence: explicit
→ Not a contradiction! Time series: +11 points in 9 months

Confidence:
[bug-002] Memory leak in order service
Relevance: critical | Confidence: low
→ Important, but not confirmed yet. Once confirmed: confidence → high

What is the confidence rating for?

It determines how high a result ranks in search

It shows whether information is confirmed or still a hypothesis

It defines how long a knowledge block is stored

Learn moreTime dimension and confidence in detail →Three time layers, time series, confidence vs. relevance

06

Take your knowledge out — 6 export formats

Your knowledge belongs to you — in any format you need. From readable project documentation to machine-readable RAG chunks.

📝

Markdown

Readable documentation with time context and confidence. Ready to share.

💾

JSON Backup

Complete export for importing into other sessions or devices.

🕸️

JSON-LD

Knowledge graph with schema.org context. For Neo4j, Gephi and other graph tools.

📊

Mermaid

Visual diagram. Color-coded by type. Renderable in Obsidian or GitHub.

🤖

RAG Chunks

Clean text blocks without formatting. For LangChain, LlamaIndex, embeddings.

✨

All at once

"Context Memory: Export everything in all formats." — 5 files, one command.

You: "Context Memory: How good is my knowledge base?"

Claude: 🔍 Quality Score: 92/100
✅ 7/8 checks passed
⚠️ 4 nodes without time data — add source dates?

You: "Context Memory: Export everything as Markdown."
Claude: ✅ 34 nodes exported → project-knowledge.md (Quality: 92/100)

Learn moreAll 6 export formats in detail →Markdown, JSON, JSON-LD, Mermaid, RAG chunks, Quality Score

✓

Ready to start?

Just say:

"Context Memory: Create a knowledge project for me."

Install the skill in Claude, and your knowledge stays — forever.

Your Knowledge. Permanent.