Knowledge Graph

Knowledge Graph Skill

Turn any workspace into a queryable knowledge graph with smart memory management. Combines graphify (code/doc graph) with three Obsidian-inspired patterns.

Installation

In pa-skills repo, the skill is at the root: pa-skills/knowledge-graph/ (not under skills/). Copy it to your workspace:

# From your pa-skills clone
cp -r knowledge-graph/ /path/to/your/workspace/skills/knowledge-graph/

Prerequisites

pip install graphifyy
graphify claw install  # ⚠️ This MODIFIES your AGENTS.md — adds graphify query rules

Important: graphify claw install appends a ## graphify section to your AGENTS.md. This is intentional — it tells your agent to check the graph before reading raw files. Review the changes after running it.

.graphifyignore

Create .graphifyignore in your workspace root to exclude noise:

node_modules/
.git/
media/
.tmp/
tmp/
graphify-out/
pa-skills/
skills-external/
*.pdf
*.ogg
*.jpg
*.png

Critical: If you have both pa-skills/knowledge-graph/ and skills/knowledge-graph/ on disk, add pa-skills/ to .graphifyignore to avoid duplicate nodes in the graph. If your repo clone lives under .tmp/pa-skills/, exclude .tmp/ too or the duplicate noise will still leak in.

Components

1. Knowledge Graph (graphify)

Build and query a knowledge graph from code + docs.

# Initial setup (AST-only, free)
graphify update .

# Query the graph
graphify query "how does X connect to Y"
graphify path "ModuleA" "ModuleB"
graphify explain "concept"

# After code changes
graphify update .

Full semantic extraction (with LLM) produces richer cross-doc connections. See graphify claw install output for AGENTS.md integration rules.

2. Auto Cross-Linker

Scans notes and adds [[wikilinks]] for known concepts from the graph.

# Build concept index from graph nodes + skills + projects
python3 scripts/wiki_crosslinker.py --build-index

# Cross-link today's daily note
python3 scripts/wiki_crosslinker.py --daily

# Cross-link all daily notes
python3 scripts/wiki_crosslinker.py --all-daily

Concepts come from: graphify nodes, skill names, project names, contacts. Only links document-type nodes with 4+ character names. Skips code internals.

3. Wiki Compiler (Karpathy Pattern)

"Compile once, query forever" — instead of RAG every time, compile scattered mentions into structured wiki pages per topic.

# See what needs compiling
python3 scripts/wiki_compiler.py --scan

# Compile all topics with 3+ mentions
python3 scripts/wiki_compiler.py --compile

# Compile one specific topic
python3 scripts/wiki_compiler.py --compile "onboarding"

# Check wiki status
python3 scripts/wiki_compiler.py --status

Output: wiki/.md with frontmatter, timeline of mentions, and graph connections. Each page is a self-contained summary — query it directly instead of scanning raw notes.

4. Structured Frontmatter

Adds YAML frontmatter with auto-detected tags to notes.

# Add frontmatter to all daily notes
python3 scripts/note_frontmatter.py --all-daily

# Add frontmatter to project docs
python3 scripts/note_frontmatter.py --projects

# Query by frontmatter
python3 scripts/note_frontmatter.py --query tag=onboarding
python3 scripts/note_frontmatter.py --query type=project

Auto-tags: graphify, crons, ops, whatsapp, calendar, content, monday, onboarding, pa-network, infra, memory, eval, self-improve, skills, owner.

5. Memory Health Checker

Runs on the knowledge graph to detect memory problems.

# Full report
python3 scripts/memory_health.py

# Quick summary
python3 scripts/memory_health.py --quick

Checks: orphan nodes, daily note gaps, stale MEMORY.md entries, weak communities, unreferenced skills, recent vs old activity.

Recommended Crons

# Daily: memory health check (04:00 UTC)
daily-memory-health: python3 scripts/memory_health.py --quick

# Weekly: wiki compilation + cross-linking (Sun 03:00 UTC)
weekly-wiki-compile:
  1. python3 scripts/wiki_crosslinker.py --build-index
  2. python3 scripts/wiki_compiler.py --compile
  3. python3 scripts/note_frontmatter.py --all-daily
  4. python3 scripts/note_frontmatter.py --projects
  5. graphify update .

Token Impact

| Operation | Without | With | Reduction | |-----------|---------|------|-----------| | Topic recall | ~15K tokens (scan daily notes) | ~200 tokens (wiki page) | 75x | | Architecture query | ~411K tokens (read all files) | ~155 tokens (graph query) | 2,655x | | "What happened with X" | grep all notes | frontmatter query + wiki page | ~50x |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: netanel-abergel
• Source: netanel-abergel/pa-skills
• License: MIT

Install and usage instructions live in the source repository linked above.