wiki-init

name	wiki-init
description	Use when bootstrapping a new personal wiki for any knowledge domain — research, codebase documentation, reading notes, competitive analysis, or any long-term knowledge accumulation project.

Wiki Init

Bootstrap a new LLM-maintained wiki at a user-specified path.

Pre-flight

Check whether a SCHEMA.md already exists nearby. If yes, ask the user if they want to reinitialize or just continue with the existing wiki.

Process

1. Gather configuration (one question at a time)

Ask:

1. Where should the wiki live? (absolute path, e.g. ~/wikis/ml-research)
2. What is the domain/purpose? (one sentence)
3. What types of sources will you add? (papers, URLs, code files, transcripts, etc.)
4. What categories should index.md use?
   • Research default: Sources | Entities | Concepts | Analyses
   • Codebase default: Modules | APIs | Decisions | Flows — see codebase.md in this skill's directory for detailed codebase guidance
   • Or specify custom

2. Create directory structure

<wiki-root>/
├── SCHEMA.md         ← conventions + absolute path (how other skills find the wiki)
├── raw/              ← immutable source documents (you add these, LLM never modifies)
├── wiki/
│   ├── index.md      ← content catalog: every page, one-line summary, by category
│   ├── log.md        ← append-only operation log
│   ├── overview.md   ← evolving synthesis of everything known
│   └── pages/        ← all wiki pages, flat, slug-named (NO subdirectories)
└── assets/           ← downloaded images, PDFs, attachments

Critical: wiki/pages/ is flat. All pages live here as <slug>.md. No subdirectories. Slugs are lowercase, hyphen-separated.

3. Write SCHEMA.md

# Wiki Schema

## Identity
- **Path:** <absolute path to wiki-root>
- **Domain:** <user's domain description>
- **Source types:** <list>
- **Created:** <YYYY-MM-DD>

## Page Frontmatter
Every wiki page must start with:
---
title: <page title>
tags: [tag1, tag2]
sources: [source-slug1]
updated: YYYY-MM-DD
---

## Cross-References
Use `[[slug]]` where slug = filename without `.md`.
Example: `[[transformer-architecture]]` → `wiki/pages/transformer-architecture.md`

## Citations

Cite every non-common-knowledge factual claim. "Common knowledge" = uncontroversial,
undergraduate-level facts in this wiki's domain. Granularity is paragraph or claim,
never per-sentence. If you cannot produce a citation in one of the forms below,
find one, weaken the claim, or drop it.

Format: Markdown footnotes. Two citation kinds, three valid targets.

**Quote citation** (preferred):

The model uses 8 attention heads.¹

**Synthesis citation** (when no single quote captures the claim):

The architecture is fundamentally an encoder-decoder with attention.²

Three rules for every footnote:

1. **The cited target is one of three forms:**
   - `[[source-slug]]` — a source-type wiki page (preferred for sources you've
     ingested via `wiki-ingest`)
   - `raw/<file>` or `assets/<file>` — a path to a local file (for drive-by
     citations where a synthesis page isn't worth creating)
   - `<URL>` — a live URL, tweet, or ephemeral source (no local copy required)

   Never cite entity, concept, or analysis pages — those are syntheses, not sources.

2. **A locator is present:** `§<section>`, `p.<n>`, `[HH:MM:SS]` for transcripts,
   URL anchor for web, or `(YYYY-MM-DD)` for dated posts.

3. **Either a verbatim quote, or the `[synthesis]` tag plus a description** of
   what the cited range supports. No third option.

**Drive-by citation examples:**

## Log Entry Format
## [YYYY-MM-DD] <operation> | <title>
Operations: init, ingest, query, update, lint, audit

## Index Categories
<one per line, matching the user's chosen taxonomy>

## Conventions
- raw/ is immutable — skills never modify it
- log.md is append-only — never rewritten, only appended
- index.md is updated on every operation that adds or changes pages
- All pages live flat in wiki/pages/ — no subdirectories
- overview.md reflects the current synthesis across all sources
<if codebase domain>
- README boundary: wiki pages must not duplicate README content. Extract structural signals; link to the README for operational content (setup, contributing, running). When ingesting any README, also evaluate it for gaps and suggest edits.
</if>

4. Write wiki/index.md

# Wiki Index — <domain>

<for each category>
### <Category Name>
<!-- entries added by wiki-ingest -->

5. Write wiki/log.md

# Wiki Log

Append-only. Format: `## [YYYY-MM-DD] <operation> | <title>`
Recent entries: `grep "^## \[" log.md | tail -10`

---

## [<today>] init | <domain>

6. Write wiki/overview.md

---
title: Overview
tags: [overview, synthesis]
sources: []
updated: <today>
---

# <Domain> — Overview

> Evolving synthesis of everything in the wiki. Updated by wiki-ingest when sources shift the understanding.

## Current Understanding

*No sources ingested yet.*

## Open Questions

*Add questions here as they arise.*

## Key Entities / Concepts

*Populated as pages are created.*

7. Confirm

Tell the user:

• Wiki initialized at <path>
• Add sources to raw/ manually, or run wiki-ingest directly with a URL or file path
• Run wiki-lint periodically to keep the wiki healthy
• SCHEMA.md is how all other skills locate this wiki — do not move or delete it

Footnotes

1. [[attention-is-all-you-need]] §3.2.2 — "We employ h = 8 parallel attention layers" ↩

2. [[attention-is-all-you-need]] §3.2-3.4 [synthesis] — encoder, decoder, and attention sections together describe the full multi-head architecture ↩