knowledge-graph

Files (5)

• SKILL.md
• memory_health.py · scripts/
• note_frontmatter.py · scripts/
• wiki_compiler.py · scripts/
• wiki_crosslinker.py · scripts/

SKILL.md (first lines)

# 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:

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

## Prerequisites

```bash
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.

```bash
# 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.

```bash
# 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.

```bash
# 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/<topic>.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.

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

From netanel-abergel/pa-skills, MIT.

More from netanel-abergel/pa-skills