skills.rest
IgorGanapolsky
IgorGanapolskyCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

wiki-graph

Validates local graph-based wiki navigation retrieval against vector-only search baselines.

1|Updated Aug 27, 2025
Search Tags:#vector search#neo4j#multi-hop#wiki navigation#retrieval evaluation#graph rag#local graph
One-click install
npx skills add https://github.com/IgorGanapolsky/Random-Timer --skill wiki-graph-igorganapolsky
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: wiki-graph
Source: https://github.com/IgorGanapolsky/Random-Timer/tree/main/.claude/skills/wiki-graph
Command: npx skills add https://github.com/IgorGanapolsky/Random-Timer --skill wiki-graph-igorganapolsky

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Vector-only RAG cannot follow links between documents, so retrieval over wiki-style knowledge bases loses precision when answers require multi-hop navigation across pages. ## Core Features & Use Cases - Local Graph Evaluation: Runs retrieval with the local_graph engine via scripts/wiki_graph.py instead of Neo4j Aura or the ki knowledge-index tool. - Navigation Verification: Uses navigate and evaluate_walk to confirm a retrieval actually walked wiki links before claiming graph-based results. - Fail-Closed Validation: Only accepts a retrieval claim when every JSON ok field is true and the process exits with code 0. - Use Case: When a user references Karpathy's wiki persistence idea or Neo4j's NICD graph navigation results, run the local graph engine over docs like AGENT_TOKEN_SHUNT.md to verify multi-hop retrieval with citations. ## Quick Start Ask the assistant to verify a wiki-graph retrieval over your docs by running the wiki_graph script with the local_graph engine, your page list, query, and citation anchor.

Frequently Asked Questions about wiki-graph

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify a retrieval actually walked wiki links?▼

Run scripts/wiki_graph.py with the local_graph engine, a comma-separated page list, your query, and a --cite anchor. Accept the result only when every JSON ok field is true and the process exits with code 0.

Graph RAG vs vector-only RAG for wiki navigation?▼

Vector search has no hop, so it cannot follow article links. NICD agents that walked wiki links beat vector-only retrieval on precision, truthfulness, and relevancy, which the local_graph engine reproduces without Neo4j.

Does this require Neo4j Aura or the ki knowledge-index tool?▼

No. The skill explicitly avoids installing ki, opening Aura, and adding embeddings. It uses a local graph engine and existing scripts like stair_toc.py for table-of-contents extraction.

When should I not use graph-based wiki navigation?▼

For a flat vector dump or a single file's headings, graph navigation adds nothing. Use the STAIR ToC script for one file's headings and rank by IAP or WQTU rather than token counts.

Why does a wiki-graph retrieval claim get rejected?▼

The skill fails closed: if any JSON ok field is false or the script exits non-zero, the retrieval claim is rejected. This prevents asserting graph-walked results without verified evidence.

Related Skills

Explore More Software Engineering Skills→
SebasArenSebasAren
2

obsidian-wiki-query

Query a personal wiki with full-text search and link traversal for cited answers.

Community
Intermediate
garrytangarrytan
29.4k

citation-graph-ingest

Extracts and classifies typed citation edges between documents in an ingested gbrain corpus.

Community
Advanced
sourcegraphsourcegraph
32

ir-analysis

Compute precision, recall, MRR, nDCG, and MAP for file retrieval comparisons.

Official
Advanced
Renzo-TognellaRenzo-Tognella

Hybrid Storage Sync

Synchronize memory items between Neo4j and Qdrant databases.

Community
Advanced