version: "1.0.0" name: "godmode:mini-context-graph" description: > A persistent, compounding knowledge base combining Karpathy's LLM Wiki pattern with a structured knowledge graph. Ingest documents once — the LLM writes wiki pages, extracts entities and relations into the graph, and stores raw content for evidence retrieval. Knowledge accumulates and cross-references across sessions; it is never re-derived from scratch. Use when building or querying a knowledge base where provenance, structural traversal, and cumulative synthesis matter. requires: [] next: []

Mini Context Graph

Standard RAG re-discovers knowledge from scratch on every query. This skill is different:

1. Wiki layer — Persistent markdown pages (summaries, entity pages, topic syntheses).

Cross-references are already there. The wiki gets richer with every ingest.

1. Graph layer — Entities and relations are extracted once and stored as a navigable

knowledge graph. BFS traversal answers structural queries without re-reading sources.

1. Raw source layer — Original documents stored immutably with chunks. Provenance links

tie every graph node and edge back to the exact text that supports it.

The LLM writes; kgx handles all bookkeeping.

Three Layers

Setup

Install the kgx CLI from the kgx workspace:

cargo install --path ~/dev/kgx/crates/kgx-cli

All commands use --root <dir> to set the data directory (default: .kgx in cwd).

Ingest

When a new document arrives:

1. Read references/ingestion.md — entity/relation extraction rules.
2. Read references/ontology.md — type normalization rules.
3. Extract entities and relations via LLM reasoning.
4. Pipe a JSON payload to kgx ingest — stores raw content, chunks, graph nodes,

provenance.

1. Write a wiki summary page via kgx wiki write.
2. Update entity pages — for each new or updated entity, write or update via

kgx wiki write --category entity.

1. Update topic pages if the document touches an existing synthesis topic.

A single document ingest typically touches 3-10 wiki pages.

# Ingest a document with entities and relations
'{"doc_id":"doc_001","title":"System Crash Analysis","source":"/docs/incident_report.pdf","raw_content":"System crashes due to memory leaks. Memory leaks occur when objects are not released.","entities":[{"name":"memory leak","type":"issue","supporting_text":"memory leaks cause crashes"},{"name":"system crash","type":"issue","supporting_text":"system crashes due to memory leaks"}],"relations":[{"source":"memory leak","target":"system crash","type":"causes","confidence":1.0,"supporting_text":"System crashes due to memory leaks."}]}' | kgx ingest
# => {"doc_id": "doc_001", "chunk_count": 1, "nodes_added": 2, "edges_added": 1}
 
# Write a wiki summary page (content from a file)
# Write the markdown to a temp file, then pipe it:
open /tmp/wiki-page.md | kgx wiki write --category summary --title "System Crash Analysis Summary" \
  --summary "Incident report: memory leaks cause system crashes."

Query

When answering a question:

1. Check the wiki first — kgx wiki search <query> to find relevant pages. Read them.
2. If the wiki has a good answer, synthesize from wiki pages (fast path).
3. If deeper graph traversal is needed, call kgx query <entity-name>.
4. Return the answer with evidence citations from supporting_chunks.
5. If the answer produces new synthesis, file it back as a wiki topic page.

# Fast path — search wiki
kgx wiki search "memory leak"
# => [{"slug": "...", "category": "summary", "path": "...", "snippet": "..."}]
 
# Full path with provenance — BFS from a seed entity
kgx query "memory leak"
# => {"query": "...", "nodes": [...], "edges": [...], "supporting_chunks": [...]}

Lint

Periodically health-check the wiki:

kgx wiki lint
# => {"orphan_pages": [...], "missing_pages": [...], "broken_wikilinks": [...], "isolated_pages": [...]}

Review and fix: broken links, orphan pages, stale claims, missing cross-references. See references/lint.md for the full lint workflow.

Ingestion Rules

• Do NOT add entities not explicitly present in the source text
• Do NOT add relations without direct textual evidence
• Do NOT add edges with confidence below 0.6
• Provide supporting_text for every entity and relation — this enables provenance
• Write a wiki summary page for every ingested document
• Update existing entity pages when new information arrives
• Flag contradictions in wiki pages when new data conflicts with old claims

Retrieval Constraints

• Traversal depth must not exceed 2 (MAX_GRAPH_DEPTH)
• Only edges with confidence >= 0.6 (MIN_CONFIDENCE)
• Maximum 50 nodes returned (MAX_NODES)
• Do NOT fabricate nodes or edges not present in the graph

CLI Reference

All commands output JSON to stdout. Use --root <dir> to override the data directory.

Responsibility Split

The human curates sources and asks questions. The LLM writes the wiki, extracts the graph, and answers with citations. kgx handles all bookkeeping.

Related

• skills/context-map/SKILL.md — pre-implementation codebase mapping (same evidence-first

pattern)

• skills/doublecheck/SKILL.md — verify claims extracted during ingest before committing them
• references/ingestion.md — entity/relation extraction rules (read before every ingest)
• references/ontology.md — type normalization rules
• references/lint.md — wiki health-check workflow
• Source: https://github.com/89jobrien/kgx