Knowledge Vault¶

Curated technical knowledge base across 22 domains. Built for LLM agents - dense, structured references that fit in a context window and give immediately actionable answers.

558 articles | 4.2 MB | 22 domains

GitHub Repository

What is this?¶

A knowledge base designed primarily for AI agents - structured so that RAG retrieval, MCP tools, and context injection return dense, actionable technical content instead of blog-style prose.

Each article is a concentrated extract: code examples, configuration patterns, gotchas, best practices. No filler, no "let me explain why this is important" - just the knowledge an agent needs to solve a real problem.

Also useful for engineers who want quick reference across 22 technical domains without wading through tutorials.

Who it's for:

• LLM agents - structured format optimized for RAG retrieval, ConTree MCP, and context injection
• Engineers - quick lookup of patterns, commands, configurations across 22 domains
• Teams - shared knowledge base accessible via ConTree sandbox or direct file access

How to use¶

Search (top bar) is the fastest way - find specific topics, commands, or patterns across all domains.

Browse the sidebar to explore by domain. Each domain contains 9-85 focused articles.

For agents: upload the repo into a ConTree sandbox and query it via MCP tools - search, read, analyze articles in an isolated environment. Or clone the repo directly - each article is a standalone .md file, easy to retrieve and inject into context.

Key conventions:

• Articles are agent-optimized references - dense, structured, no hand-holding
• [[wiki-links]] connect related concepts across domains (e.g., Kafka article links to Architecture patterns)
• Code examples are copy-paste ready
• Each article covers one focused topic - from a single CLI command to a complete system design pattern

Domains¶

Knowledge Graph¶

Articles aren't isolated - they form a connected knowledge graph with cross-domain references:

• 558 nodes (articles) with 2100+ edges (cross-references)
• 68 topic communities detected via graph analysis
• Queryable via ConTree MCP - upload the repo into a sandbox and search/read/analyze articles with full isolation
• Each article is a standalone .md file - easy to index, retrieve, and inject into LLM context

Freshness Policy¶

Not all knowledge ages equally. Each domain has an update cycle based on how fast the field moves:

Articles include version/date context where relevant (e.g., "PostgreSQL 17", "React 19", "Kubernetes 1.30").

What makes this different¶

Agent-first design. Every article is structured for machine consumption: consistent headers, code blocks with language tags, pattern/anti-pattern sections, explicit gotchas. An LLM agent retrieving a Knowledge Vault article gets immediately actionable context - no parsing needed.

Density over length. A typical article packs the same information as a 2-hour video or a 30-page tutorial into 2-4 pages of pure reference text. Optimized for context window efficiency.

Cross-domain connections. Real engineering problems don't respect domain boundaries. Wiki-links connect Kafka consumer patterns to Architecture decisions, SQL optimization to Data Engineering pipelines, Security practices to DevOps configurations.

Living knowledge base. Continuously updated with new research and domain knowledge. Freshness policy ensures fast-moving fields stay current while stable foundations remain reliable.

Related Projects¶

claude-code-skills

Open-source collection of skills, architectural patterns, and best practices for Claude Code. If you're building with AI agents, this might be useful:

• Multi-agent harness design - Generator-Evaluator pattern, Sprint Contracts, context management
• Proof loop verification - spec freeze -> build -> evidence -> fresh verify cycle
• Autoresearch - iterative self-optimization with mechanical scoring
• Structured reasoning - premises, execution trace, conclusions for complex debugging
• Context engineering - surviving long sessions, post-compaction recovery, JIT retrieval