Knowledge Graph Pack Manager

Build, install, and manage domain-expert knowledge packs that give Claude deep expertise in specific technologies, frameworks, and languages.

Each pack is a LadybugDB graph database containing extracted documentation with vector embeddings for semantic search, entity relationships, and facts.

Setup

This skill requires the agent-kgpacks repository. If not present:

# Clone the repo (if not already available)
git clone https://github.com/rysweet/agent-kgpacks.git ~/.wikigr/agent-kgpacks
cd ~/.wikigr/agent-kgpacks && uv sync

Set KGPACKS_ROOT to the repo location (defaults to ~/.wikigr/agent-kgpacks):

export KGPACKS_ROOT=~/.wikigr/agent-kgpacks

Commands

Parse the user's arguments to determine the subcommand:

Execution

list

KGPACKS_ROOT="${KGPACKS_ROOT:-$HOME/.wikigr/agent-kgpacks}"
for db in "$KGPACKS_ROOT"/data/packs/*/pack.db; do
  dir=$(dirname "$db"); name=$(basename "$dir")
  articles=$(python3 -c "import json; print(json.load(open('$dir/manifest.json')).get('graph_stats',{}).get('articles','?'))" 2>/dev/null)
  installed="no"; [ -f ".claude/skills/$name/SKILL.md" ] && installed="yes"
  printf "%-30s %4s articles  installed=%s\n" "$name" "$articles" "$installed"
done | sort

install <name>

1. Verify pack exists at $KGPACKS_ROOT/data/packs/{name}/pack.db
2. Generate and write SKILL.md to .claude/skills/{name}/SKILL.md in the current project
3. The skill tells Claude how to query that pack's KG when the domain comes up

KGPACKS_ROOT="${KGPACKS_ROOT:-$HOME/.wikigr/agent-kgpacks}"
cd "$KGPACKS_ROOT" && uv run python scripts/install_pack_skills.py
# Or install a single pack:
mkdir -p .claude/skills/{name}
# Generate SKILL.md with pack-specific content (see install_pack_skills.py)

build <topic>

1. Normalize topic → pack name (lowercase, hyphens)
2. Use WebSearch to find 30-50 documentation URLs
3. Create $KGPACKS_ROOT/data/packs/{name}/urls.txt
4. Run build:

cd "$KGPACKS_ROOT"
uv run python scripts/build_pack_from_issue.py --issue-json '{
  "pack_name": "{name}",
  "description": "{topic}",
  "search_terms": "{search terms}"
}'

1. Install skill: uv run python scripts/installpackskills.py

query <pack> <question>

import sys
sys.path.insert(0, KGPACKS_ROOT)
from wikigr.agent.kg_agent import KnowledgeGraphAgent

with KnowledgeGraphAgent(
    db_path=f"{KGPACKS_ROOT}/data/packs/{pack_name}/pack.db",
    read_only=True
) as agent:
    result = agent.query(question, max_results=5)
    # Use result["answer"], result["sources"], result["entities"]

info <name>

Read and display $KGPACKS_ROOT/data/packs/{name}/manifest.json.

update <name>

Delete pack.db, re-run build script, re-install skill.

uninstall <name>

rm -rf .claude/skills/{name}

Notes

• Builds require ANTHROPICAPIKEY and take 3-30 minutes
• Pack databases (pack.db) are 2-50 MB, built locally
• Skills activate in the next Claude Code session, not the current one
• Run /kg-pack list to see what's available before installing