From wiki-wonka
Ingests source files from raw/ into wiki: reads content, discusses takeaways, creates summary pages for sources/entities/concepts, updates index/log.
How this skill is triggered — by the user, by Claude, or both
Slash command
/wiki-wonka:ingestThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You were invoked by the orchestrator because the user wants to add a new source to the wiki. Follow every step in order. Do not skip steps. Do not proceed to the next step without completing the current one.
You were invoked by the orchestrator because the user wants to add a new source to the wiki. Follow every step in order. Do not skip steps. Do not proceed to the next step without completing the current one.
Locate the file the user indicated in raw/. Read it in full before doing anything else.
raw/, tell the user and stop. Do not ingest from outside raw/.Before touching any wiki file, surface the key takeaways to the user. Be concise: 3–5 bullet points max. Then ask:
Wait for the user's response. Adjust your understanding before proceeding. If the user says "go ahead" with no changes, proceed with your own judgment.
This step exists because the human curates, the LLM executes. Do not skip it even if the source seems straightforward.
Create wiki/sources/<slug>.md where <slug> is a lowercase hyphenated version of the source title.
Use this exact frontmatter:
---
title: "Full title of the source"
slug: source-slug
type: source
date_ingested: YYYY-MM-DD
original_file: raw/filename.ext
authors: []
tags: []
related_concepts: []
related_entities: []
---
Page structure:
## Summary
2–4 paragraphs. What is this source about? What is its main argument or contribution?
Write as if explaining to someone who will never read the original.
## Key claims
- Claim 1 — be specific, include data or quotes where relevant.
- Claim 2
- ...
## Connections to existing wiki
What does this source confirm, challenge, or add nuance to?
Reference existing pages using [[wikilinks]].
If it contradicts an existing page, flag it explicitly:
> [!contradiction] This source disputes [[concepts/existing-page]] on X.
## Open questions
What does this source leave unanswered? What would be worth investigating next?
## Notable quotes
> "Direct quote if relevant." (p. N or timestamp)
Before writing anything, scan wiki/index.md and list every page that this source touches. Think across three categories:
Entities — people, organizations, products, datasets, models named in the source. Check if a page already exists. If yes, update it. If no, create it.
Concepts — themes, techniques, theories, arguments the source engages with. Same rule: update if exists, create if not.
Overview — wiki/overview.md always gets updated when a new source is ingested.
Write this list out before proceeding:
Pages to update:
- entities/author-name (exists)
- concepts/self-attention (exists)
- concepts/cross-attention (new)
- overview (always)
Ask the user if this list looks right. Adjust if needed.
Work through the list from Step 4 one page at a time.
Read the current page fully. Then:
> [!outdated] instead.> [!contradiction] callout with a link to the source page.[[sources/slug]] to the page's related_sources frontmatter field.## From [[sources/slug]] subsection if the source adds substantial new content.Frontmatter:
---
title: "Entity Name"
slug: entity-slug
type: entity
subtype: person | organization | model | dataset | product
tags: []
related_sources: [sources/slug]
related_concepts: []
---
Write a factual summary of who or what this entity is, based only on what your sources say. Do not add external knowledge not present in the wiki.
Frontmatter:
---
title: "Concept Name"
slug: concept-slug
type: concept
tags: []
related_sources: [sources/slug]
related_entities: []
related_concepts: []
---
Page structure:
## Definition
What is this concept? One clear paragraph.
## How it works
Mechanism, process, or explanation.
## Evidence and claims
What do ingested sources say about this? Cite with [[wikilinks]].
## Connections
Links to related concepts and entities.
## Open questions
wiki/overview.md is the synthesis layer. After every ingest:
After all pages are written, update the two navigation files. These are owned by the orchestrator but you write the entries during ingest.
wiki/index.md — add new pages in the correct category. For updated pages, do not add a duplicate entry.
wiki/log.md — append one entry at the top of the log (most recent first):
## [YYYY-MM-DD] ingest | Title of source
Pages touched: sources/slug, entities/x, concepts/y, overview (N total)
New pages created: concepts/cross-attention
Contradictions flagged: 1 (see sources/slug)
Tell the user what was done in plain language. No need to list every file — summarize:
Done. Ingested "Attention Is All You Need" (2017).
Created: sources/attention-is-all-you-need, concepts/cross-attention
Updated: concepts/self-attention, concepts/transformer, entities/vaswani-ashish, overview
Flagged: 1 contradiction with concepts/positional-encoding
Anything you want me to revisit before we continue?
raw/. If the user pastes content directly into chat, ask them to save it to raw/ first, then ingest from there. This keeps the source layer clean.npx claudepluginhub cooperacode/wiki-wonkaIngests new sources (papers, articles, URLs, files, transcripts) into a wiki: reads content, surfaces takeaways, writes summary pages, updates entities/concepts/index/log.
Ingest new sources into the LLM Wiki. Reads unprocessed files from raw/, docs/, and notes/, creates source summaries, updates entity/concept pages, maintains cross-references, and updates the index and log. Use when new files have been added.
Ingests files into a local LLM-powered wiki, creating source pages, extracting entities/concepts, and updating the index. Useful for building a personal knowledge base from documents.