Skip to content
Primores AI wiki

How This Wiki Is Built

How This Wiki Is Built

TL;DR: This wiki is maintained by Claude Code following a structured methodology. Sources go in, get analyzed, and compound into interconnected knowledge. The AI handles maintenance humans typically abandon — cross-linking, consistency checking, gap identification. The result is a knowledge base that grows stronger over time rather than rotting.

The Core Principle

Traditional wikis die because maintenance is boring. Cross-linking, updating outdated pages, catching contradictions, filling gaps — humans start strong and abandon these tasks within months.

This wiki solves that by making an LLM the maintainer. Claude Code:

  • Ingests new sources and extracts insights across multiple pages
  • Cross-links related concepts automatically
  • Lints for contradictions, orphan pages, and staleness
  • Grows the wiki proactively, not just reactively

The Three-Layer Structure

┌─────────────────────────────────────────┐
│  wiki/                                  │  ← Public knowledge base
│  Interconnected pages, glossary,        │
│  case studies, tools, comparisons       │
├─────────────────────────────────────────┤
│  raw/                                   │  ← Source material (private)
│  Articles, reports, screenshots,        │
│  research that feeds the wiki           │
├─────────────────────────────────────────┤
│  Schema                                 │  ← Conventions & rules
│  Page templates, status system,         │
│  frontmatter requirements               │
└─────────────────────────────────────────┘

Sources feed wiki pages following schema conventions. The LLM handles the translation layer — extracting, connecting, maintaining.

The Three Core Operations

1. INGEST — Adding Knowledge

When a new source is added:

  1. Claude reads it completely
  2. Extracts actionable insights
  3. Updates relevant wiki pages (often 5-15 pages from one source)
  4. Creates new pages if topics don’t exist
  5. Cross-references with existing content
  6. Logs the activity

One source touches many pages. That’s the compounding mechanism.

2. QUERY — Answering Questions

When you ask a question:

  1. Claude searches the wiki for relevant pages
  2. Synthesizes an answer with citations to wiki pages
  3. Offers to save valuable answers as new pages

Good answers become permanent knowledge. They don’t disappear into chat history.

3. LINT — Health Checks

Periodic checks for:

  • Contradictions — Do any pages disagree?
  • Orphans — Pages with no inbound links
  • Staleness — Outdated information
  • Gaps — Concepts mentioned but lacking pages
  • Private path leakage — Links to non-public files

The Status System

Pages mature through three stages:

StatusMeaningCriteria
🌱 SeedlingEarly thoughtsInitial capture, may change
🌿 GrowingSolid foundation2-3 sources, stable structure, validated
🌳 EvergreenComprehensiveWell cross-linked, actively maintained

Seedlings are fine. Publishing early and improving continuously beats waiting for perfection.

Content Philosophy

Voice

  • Practical over theoretical — “here’s what works” beats “here’s what’s possible”
  • Accessible but not dumbed down — respect the reader’s intelligence
  • Honest about limitations — acknowledge when things don’t work

Quality Standards

  • Every claim has a source OR is clearly marked as opinion
  • Tool reviews include hands-on testing
  • Pages update when information becomes outdated
  • “I don’t know yet” is valid

Structure

Every page follows a template:

  • TL;DR — 2-3 sentence quotable summary
  • Main content — Clear H2/H3 hierarchy
  • Key Takeaways — Bullet summary
  • Related — Links to connected pages
  • Sources — Where this came from

Why This Works

For Knowledge Building

  • No knowledge disappears into old conversations
  • Connections between concepts are explicit
  • Gaps become visible and fillable
  • Maintenance happens automatically

For Content Creation

  • Articles draw from accumulated wiki knowledge
  • Named frameworks become citation magnets
  • Case studies build authority over time

For AI Systems

  • Structured pages get cited by AI answer engines
  • TL;DR blocks are quotable summaries
  • Clear attribution builds E-E-A-T signals

The Compounding Effect

Week 1:  10 pages
Week 4:  25 pages + 50 cross-links
Week 8:  45 pages + 150 cross-links + patterns emerging
Week 12: 70 pages + 300 cross-links + named frameworks + case studies

Each new page connects to existing pages. Each connection makes future pages easier. The wiki gets smarter, not just bigger.

Key Takeaways

  • LLM handles the maintenance humans abandon (cross-linking, linting, gap-filling)
  • One source → many wiki pages (compounding)
  • Good answers become permanent pages
  • Status system allows publishing early and improving continuously
  • The wiki gets stronger over time, not weaker