# Build Queryable Knowledge Graphs

Large document sets are hard to reason over in one context window. This skill indexes documents into a graph for multi-hop queries with provenance.

## Install

```bash
npx skillstore add cornjebus/recursive-knowledge
```

## Metadata

- Status: approved
- Slug: cornjebus-recursive-knowledge
- Skillstore revision: r1
- Version status: missing
- Tree hash: 88005aebd9edea7ae4455a669c59fdb6b97d49ee9c36ff2a738602bfa1b2eb34
- Author: Cornjebus
- GitHub username: Cornjebus
- License: MIT
- Repository: https://github.com/Cornjebus/rlm-replication-study/tree/main/recursive-knowledge/
- Ref: 02f077c174c5335e2f5d02ca15e77b70d9543e58
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: allowed
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: env\_access, external\_commands
- Quality score: 73
- Public page: https://skillstore.pages.dev/skills/cornjebus-recursive-knowledge
- Manifest: https://skillstore.pages.dev/api/skills/cornjebus-recursive-knowledge/manifest

## Capabilities

- Indexes .txt, .md, and .json documents into entities, relationships, and source metadata.
- Splits large documents into overlapping chunks before extraction.
- Builds deterministic entity, relationship, and document identifiers for deduplication.
- Queries graph entry points with visited-node and visited-edge state tracking.
- Returns answers with confidence, source counts, traversal statistics, and termination reasons.
- Can append new documents to an existing graph file.

## Use Cases

- Research Corpus Mapping: Index a large paper or report collection, then ask questions that require links across documents.
- Knowledge Base Question Answering: Turn internal guides into a graph that returns answers with source counts and confidence.
- Evidence Link Analysis: Trace people, organizations, events, and concepts across many static files.

## Prompt Templates

### Index a Corpus

```
Use recursive-knowledge to index my documents folder into a new graph file. Keep verbose progress enabled.
```

### Ask a Provenance Query

```
Use recursive-knowledge to answer this question from my graph. Include confidence, sources, and the stopping reason.
```

### Append New Documents

```
Use recursive-knowledge to add new documents to my existing graph without rebuilding unchanged documents.
```

### Tune Traversal Rules

```
Use recursive-knowledge with a deeper traversal limit and higher confidence threshold for this cross-document investigation.
```

## Limitations

- Heuristic extraction is basic unless USE\_LLM is enabled.
- Remote LLM extraction can send document text to Anthropic when explicitly enabled.
- The query engine uses simple keyword intent parsing, not full semantic search.
- It handles static corpora and does not process real-time streams.

## Best Practices

- Index only reviewed folders that are safe for local processing.
- Enable remote LLM extraction only after confirming data handling requirements.
- Use higher confidence thresholds for decisions that need stronger evidence.

## Anti Patterns

- Do not use this for small files that fit in the active context.
- Do not treat heuristic extraction as a complete entity recognition system.
- Do not index broad home directories or folders with unrelated secrets.

## Security Audit

- Audited at: 2026-07-05T07:06:04.952\+00:00
- Summary: The system-reconnaissance hits are false positives caused by knowledge-graph terminology and traversal examples. The Markdown backtick detections are documentation formatting, not automatic command execution. The main residual concern is optional remote LLM processing, which can send document chunks to Anthropic when USE\_LLM is enabled.

## Stats

- Views: 167
- Downloads: 26
- Favorites: 1
- Popularity score: 0