About this skill

Problem It Solves

A raw retrieval-augmented answer can mix retrieved snippets, assumptions, and unmarked inferences. LLM Wiki Advanced targets a local knowledge-base workflow where the model must classify the question, retrieve evidence, annotate confidence, and update a persistent wiki before answering.

How It Works

Environment-aware runtime: uses local vector search when bash and Node.js are available, and falls back to keyword matching over wiki/cards/*.md in cloud sessions.
Ingestion: @wiki+ ingest parses PDF, Word, Excel, images, Markdown, and other files, writes L1 summaries to wiki/index.md, creates or updates concept cards, and triggers incremental index updates in local mode.
Evidence-based reasoning: after a hit, the EBR path routes definition, causality, prediction, evaluation, or gap questions to different inference structures.
Confidence labeling: key claims are marked 🟢 confirmed, 🟡 inferred, or 🔴 assumed; if nothing is found, it states the gap instead of hallucinating.
Feedback and diagnostics: @wiki+ remember stores user-confirmed knowledge, check-health scans for broken links, contradictions, and missing citations, and rebuild-index rebuilds the vector index.

Boundaries

Best for maintaining concept libraries, document corpora, research notes, and wiki-based QA. Cloud mode cannot run scripts, so vector updates require a local environment. Feedback writes require a source document or link, and unverified external claims should be confirmed before storage.

Screenshots

Use Cases

Ingest product PDFs, Word files, and images into a local wiki so later answers can cite specific concept cards.
Before answering customer questions in Cursor, retrieve the local knowledge base and label claims as confirmed, inferred, or assumed.
Run periodic health checks to find orphaned links, contradictions, and missing citations, then generate an issue list.
Write user-confirmed facts into the matching knowledge card and update the incremental vector index for later semantic search.

Best For

Engineers maintaining technical docs who want to turn PDFs and Markdown into retrievable concept cards.
Presales reps answering customer questions who need cited knowledge-base evidence and clear assumption labels.
Knowledge managers organizing research notes who want to detect orphaned links, contradictions, and missing citations.
Wiki engineers updating internal knowledge bases who need to store confirmed facts and refresh the index.

Related Skills

Scholar Reasoning Distiller

@user_231765ee/django-plus

Extracts an eight-layer reasoning fingerprint from scholar texts and applies a nine-module constraint to run peer review, supervision, lecturing, or panel discussion with quantitative scores and qualitative comments.

Knowledge Management 2026.08.30 ↓ 608