confluence-to-agent-knowledge-base

Confluence to Agent Knowledge Base

Turn any Confluence Cloud instance into a private, searchable AI knowledge base on GitHub.

A Claude Code skill that walks you through converting your company's Confluence spaces into a structured, LLM-powered knowledge base — with access-tiered repos so sensitive docs stay locked down.

What It Does

Your Confluence ──► Private GitHub Repos ──► Searchable AI Knowledge Base
                    (access-tiered)          (search, Q&A, auto-sync)

6-phase interactive walkthrough:

1. Discovery — Finds your Confluence instance and lists all spaces (no API tokens needed)
2. Access Tiers — Groups spaces by who should see them (general, leadership, sales, etc.)
3. Ingest — Pulls all pages into clean markdown with metadata
4. Compile — Builds a categorized wiki with summaries and indexes
5. GitHub — Creates private repos per tier and pushes
6. Auto-Sync — Sets up weekly sync to keep the Knowledge Base current

Install

claude skill add jpmim/confluence-to-agent-knowledge-base

Or manually:

git clone https://github.com/jpmim/confluence-to-agent-knowledge-base.git ~/.claude/skills/confluence-to-agent-knowledge-base

Prerequisites

• Claude Code with Claude Pro or Max subscription
• Atlassian MCP connector connected in Claude Code
• Python 3.9+
• Git with push access to your GitHub org

Usage

In Claude Code, just say:

"Set up a knowledge base from our Confluence"

Or:

"Migrate our Confluence to a private GitHub knowledge base"

Claude will walk you through the entire process interactively.

What You Get

For each access tier, you get a private GitHub repo containing:

your-company-kb/
  raw/              # Original Confluence pages as markdown
  wiki/             # Compiled knowledge base
    _index.md       # Master index
    _summaries.md   # One-line summaries of all pages
    operations/     # Categorized articles
    technical/
    onboarding/
    ...
  confluence_kb/    # Python CLI tool
  CLAUDE.md         # Agent instructions (so your AI agents can use the KB)

CLI commands included:

• ckb search "query" — Full-text search (no API key needed)
• ckb query "question" — LLM-powered Q&A with citations
• ckb interactive — Chat-style Q&A session
• ckb compile --fast — Recompile the wiki
• ckb lint — Health checks and quality audit
• ckb status — Show Knowledge Base stats

Why Access Tiers?

Most companies have sensitive content in Confluence that shouldn't be visible to everyone:

• Management spaces with strategy, financials, board docs
• Sales spaces with pricing, contracts, client data
• HR spaces with compensation, performance reviews

This skill separates those into different private repos with different GitHub collaborator lists. Everyone gets the general Knowledge Base, but restricted content stays restricted.

How It Works Under the Hood

• Uses the Atlassian MCP connector for all Confluence access — no API tokens to manage
• Converts pages to markdown with YAML frontmatter for maximum portability
• Compiles a structured wiki with categories, summaries, and indexes
• Includes a TF-IDF search engine for instant full-text search
• Sets up weekly auto-sync via Claude Code scheduled tasks
• Each repo includes CLAUDE.md so AI agents can read and contribute to the Knowledge Base

Inspired By

Karpathy's LLM Knowledge Base approach — raw data compiled by LLMs into a wiki that agents maintain and query.

License

MIT