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 CLAUDE.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: Wiki Pages | Concepts Pages | Topic Relations | Quick Navigation
   • Codebase default: Modules | APIs | Decisions | Flows
   • Or specify custom

2. Create directory structure

<wiki-root>/
├── AGENTS.md         ← cross-agent entrypoint and compatibility guidance
├── CLAUDE.md         ← Claude-specific conventions + absolute path
├── .agents/
│   └── skills/       ← compatibility references for other agents/tools
├── .claude/
│   └── skills/       ← canonical skill definitions
├── raw/              ← immutable source documents (you add these, LLM never modifies)
├── wiki/
│   ├── index.md      ← content catalog: table format with page, summary, tags, updated
│   ├── log.md        ← append-only operation log
│   └── pages/        ← all wiki pages, flat, slug-named (NO subdirectories)
└── concepts/         ← generated reports, analyses, answers

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

3. Write CLAUDE.md

Create a project-specific CLAUDE.md with the following structure:

# <Domain> Wiki Schema

## What Is This

LLM Wiki for <user's domain description> — a structured knowledge base continuously maintained by AI.
Raw sources are never modified; the wiki layer is incrementally built and interconnected by AI.

Today's date: {{currentDate}}

---

## Directory Structure

AGENTS.md Cross-agent entrypoint and compatibility guidance

.claude/ Canonical Claude-facing skill definitions .agents/ Compatibility skill references for other agents/tools raw/ Raw sources (read-only, never modify)

wiki/ AI-maintained wiki pages pages/ Topic pages (one .md file per topic) index.md Full index (table format) log.md Operation log (append-only, never modify history)

concepts/ Generated reports, analyses, answer pages

---

## Wiki Page Standards

- One `.md` file per topic, filename in lowercase with hyphens: `machine-learning.md`
- Use `[[topic-name]]` format to link related topics (Obsidian wikilink)
- Add YAML frontmatter at the top of each page:

\`\`\`yaml
---
title: Topic Name
tags: [tag1, tag2]
sources: Number of sources
updated: YYYY-MM-DD
---
\`\`\`

- Suggested page structure: Core Definition → Key Takeaways → Related Topics → Open Questions

---

## Operations

### Ingest (Add New Sources)

1. Read source files from `raw/` (do not modify originals)
2. Discuss key takeaways with user (optional)
3. **Network supplement** (recommended): Call `agent-browser` skill to fetch authoritative sources
4. Write or update related topic pages in `wiki/pages/`, **cite sources**
5. Update `wiki/index.md` (table format)
6. Append a record to `wiki/log.md`

### Query (Ask Questions)

1. First read `wiki/index.md` to find relevant pages
2. Read relevant wiki pages
3. If information is insufficient, call `agent-browser` skill to fetch current sources
4. Synthesize answer with citations
5. If valuable, save to `concepts/`, update `wiki/index.md`
6. Append a record to `wiki/log.md`

### Lint (Health Check)

1. Scan wiki pages for issues
2. Output check report with suggested fixes
3. Append check record to `wiki/log.md`

---

## Index Maintenance Rules

`wiki/index.md` uses table format:

\`\`\`markdown
# Index

## Wiki Pages
| Page | Summary | Tags | Last Updated |
|------|---------|------|--------------|
| [[topic]] | One-line description | tag1, tag2 | YYYY-MM-DD |

## Concepts Pages
| Page | Type | Date |
|------|------|------|
| [[analysis-xxx]] | Analysis Report | YYYY-MM-DD |
\`\`\`

Update index after every ingest or concept page generation.

---

## Log Maintenance Rules

`wiki/log.md` is append-only operation record:

\`\`\`markdown
# Operation Log

Append-only, never modify history.

Format: \`## [YYYY-MM-DD] <operation> | <title>\`

---

## [YYYY-MM-DD] init | Wiki Initialized
\`\`\`

---

## Core Principles

- **raw/ is read-only**: Never modify raw sources
- **wiki/ is AI-written**: Users read, AI maintains
- **Knowledge compounding**: Each ingest enriches the entire wiki
- **Cross-references first**: Integrate new info into existing pages
- **Mark contradictions**: Clearly flag when new sources conflict with old conclusions
- **Preserve good analysis**: Save valuable answers to `concepts/`

4. Write AGENTS.md

Create a concise, tool-agnostic AGENTS.md that:

• explains the repo purpose and the roles of raw/, wiki/, and concepts/
• states that CLAUDE.md contains Claude-specific workflow rules
• states that .claude/skills/ is the canonical skill definition location
• states that .agents/skills/ is a compatibility layer for other agents/tools
• tells maintainers not to duplicate or fork skill content inside .agents/skills/

5. Create .agents/skills/ compatibility references

Create thin compatibility references for each public skill under .agents/skills/<skill>/SKILL.md. Each file should point back to the canonical local .claude/skills/<skill>/SKILL.md file instead of copying the full content.

6. Write wiki/index.md

# Index

## Wiki Pages
| Page | Summary | Tags | Last Updated |
|------|---------|------|--------------|
<!-- entries added by wiki-ingest -->

## Concepts Pages
| Page | Type | Date |
|------|------|------|
| (No saved analysis pages yet) | - | - |

## Topic Relations
(To be populated)

## Quick Navigation

### By Research Area
(To be populated)

7. Write wiki/log.md

# Operation Log

Append-only, never modify history.

Format: `## [YYYY-MM-DD] <operation> | <title>`

---

## [<today>] init | <domain>
- Created CLAUDE.md schema
- Created AGENTS.md compatibility guide
- Initialized log.md

8. Create concepts/ directory

mkdir -p concepts

9. 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
• CLAUDE.md contains Claude-specific rules for this wiki
• AGENTS.md and .agents/skills/ provide cross-agent compatibility entrypoints
• .claude/skills/ remains the canonical skill source of truth