Private Knowledge Base

Store, search, and summarize concepts across your PDFs and papers with fast semantic search and cross-document Q&A.

Security Scan

The skill's files and runtime instructions are coherent with a local personal knowledge-base: the scripts operate on user-supplied PDFs, store extracted text and metadata under a configurable KB_ROOT, and use local tools (pdftotext / pypdf); there is no hidden network exfiltration in the provided scripts, but metadata includes absolute source paths and the README suggests calling a local model (ollama).

Purpose & Capability

Name/description match the included scripts and schema: ingestion, text extraction, simple search, and summarization workflows are implemented by the shell scripts and index schema. No unrelated credentials, binaries, or services are requested.

Instruction Scope

Scripts only read user-supplied PDF files and write extracted text, embeddings folder, and index JSON under KB_ROOT (default ~/kb). Two noteworthy items: (1) metadata stores the full source path in index JSON (may reveal filesystem layout or sensitive path names), and (2) summarize.sh prints a suggested command using 'ollama run qwen3.5' — that is an external model invocation the README suggests but is not enforced by the scripts. Otherwise instructions are scoped to the stated purpose.

Install Mechanism

No install spec — instruction-only with local shell scripts. Scripts rely on common local tools (pdftotext, python3, pypdf) but do not download or execute remote code. This is low-risk relative to other install types.

Credentials

No required environment variables or credentials are declared. An optional KB_ROOT env var is used to choose storage location, which is proportionate. No other secrets or unrelated env vars are requested.

Persistence & Privilege

always:false and user-invocable default. The skill does not request permanent system-wide presence, does not modify other skills' config, and only writes files under the configured KB_ROOT.

Assessment

This skill appears to do what it says and works only on local files, but consider the following before installing or running it: - KB metadata stores the absolute source path in index JSON — if you later upload the KB or share it, that may reveal filesystem layout or sensitive directory names. Consider setting KB_ROOT to a dedicated directory and reviewing index JSON files before sharing. - The scripts will read any file you pass to them; only give them PDFs you trust. They write extracted text under KB_ROOT/docs and metadata under KB_ROOT/index. - summarize.sh suggests using 'ollama run qwen3.5' (a local model runtime). That step is optional and not enforced by the scripts; if you run it, verify your ollama setup and understand whether that model is local or configured to call an external service. - The scripts rely on pdftotext or python (pypdf). Installing those packages may be required; install them from well-known sources. - If you plan to back up or share the KB, review contents for sensitive information (full paths, PII in extracted text) first. Overall this is internally consistent and low-risk for local use, but be cautious about storing or sharing the generated index and text files.

Personal document storage and retrieval system for PDFs, papers, and research documents.

Quick Start

Ingest Documents

# Add PDF to knowledge base
./scripts/ingest.sh ~/path/to/document.pdf

# Process entire folder
./scripts/ingest-folder.sh ~/papers/

Query Knowledge Base

# Search for concept across all documents
./scripts/search.sh "transformer architecture"

# Get summary of concept from relevant docs
./scripts/summarize.sh "attention mechanism"

Core Workflows

1. Document Ingestion

When user provides new PDFs or papers:

1. Create document entry in kb/index.json
2. Extract text and metadata
3. Generate embeddings for semantic search
4. Store in kb/docs/ with normalized name

2. Cross-Document Q&A

When user asks "which document mentions X?" or "summarize X from my docs":

1. Search embeddings for relevant passages
2. Retrieve source documents
3. Synthesize answer across documents
4. Cite sources with document names and page numbers

3. Concept Linking

Build associations between documents:

• Shared concepts
• Citation relationships
• Topic clusters

File Structure

private-knowledge-base/
├── SKILL.md
├── scripts/
│   ├── ingest.sh          # Single document ingestion
│   ├── ingest-folder.sh   # Batch ingestion
│   ├── search.sh          # Semantic search
│   └── summarize.sh       # Cross-document summary
├── references/
│   └── schema.md          # KB index schema
└── kb/                    # Created at runtime
    ├── index.json
    ├── embeddings/
    └── docs/

Usage Examples

User: "我之前存的文档里，哪篇提到了 transformer?" → Run ./scripts/search.sh "transformer"

User: "总结一下我文档里关于 attention 的内容" → Run ./scripts/summarize.sh "attention"

User: "把这篇 PDF 加到知识库" → Run ./scripts/ingest.sh <pdf-path>

Configuration

Set knowledge base location:

export KB_ROOT=~/.openclaw/workspace/kb

Default: ~/kb if not set.