Wiki Ingest

• Pasted text directly
• File path(s) (md, txt, pdf, etc.)
• Directory path (batch processing)

If user doesn't specify a wiki path, use wiki/. Create the directory if it doesn't exist.

2. Check Existing Wiki

Read wiki/index.md (if exists) to understand existing pages and avoid duplicates.

3. Extract Knowledge Entities

Extract from content:

• Concepts (concepts/): Abstract ideas, terminology, theories
• Products (products/): Specific products, tools, services
• Patterns (patterns/): Engineering patterns, design decisions, methodologies
• Comparisons (comparisons/): Cross-product or cross-approach analysis

Extraction threshold: An entity deserves its own page only if it would be referenced by other pages.

4. Create or Update Pages

Each entity maps to one wiki page.

→ Load references/page-templates.md for page templates.

Rules:

• If page already exists → update it, append new information, don't overwrite existing content
• Keep the one-line definition stable unless the new content provides a clearly better one
• Prioritize updating "Sources" and "Related Pages" sections

5. Update Cross-References

Check all existing wiki pages. If new content involves concepts referenced in other pages:

• Update their Related Pages list
• Update their Sources list
• Supplement new information in their detail sections

6. Update index.md

Add new page entries to wiki/index.md, organized by category.

7. Append to log.md

Append to wiki/log.md:

## YYYY-MM-DD: Ingest <source title>

**Source**: <file path or "user input">
**New pages**: list newly created pages
**Updated pages**: list modified pages
**New cross-references**: list newly established links

8. Report Results

Tell user: what pages were created, what pages were updated, what new cross-references were established.

File Naming

• All lowercase, hyphen-separated: agent-loop.md, kv-cache.md
• Name by entity, not by number

Cross-Reference Format

• Use [[category/page-name]] double-link format (Obsidian-compatible), e.g. [[concepts/agent-loop]], [[products/claude-code]]
• Always include the category prefix to avoid ambiguity across directories
• List all related pages at the bottom of each page

Guidelines

• Never create duplicate pages — always check wiki/ first
• Don't extract trivial entities — if a concept appears once and won't be referenced elsewhere, skip it
• For batch processing: extract all entities first, then create pages and update references in one pass to avoid inconsistent intermediate states
• Always update index.md and log.md after every ingest