Ingest Protocol

Type: concept Tags: llm-wiki, claude-code, obsidian-web-clipper

Summary

The workflow for turning a raw source file into a set of interconnected wiki pages. Defined in the vault’s CLAUDE.md and triggered by telling Claude Code: “Please ingest [filename].”

Steps

1. Read the full source document from raw/
2. Extract: main topics, key concepts, people/orgs, quotes, takeaways
3. (Optional) Ask clarifying questions about depth and focus
4. Create wiki pages:
   • 1 source page → wiki/sources/
   • N concept pages → wiki/concepts/
   • N entity pages → wiki/entities/
   • 1 analysis page → wiki/analysis/ (if cross-source synthesis is warranted)
5. Add [[wikilinks]] between all related pages
6. Update wiki/index.md with new entries under correct categories
7. Append timestamped entry to wiki/log.md

Providing Context

Before or during ingest, you can tell Claude:

• What this vault is for (research topic, second brain, YouTube knowledge)
• How thorough to be (granular vs. concise)
• What to emphasize

Typical Output Scale

The AI 2027 article ingest (referenced in yt-llm-wiki-knowledge-system) created 23 wiki pages in ~10 minutes from a single long article.

Source Delivery Options

• Manual copy-paste into raw/
• obsidian-web-clipper — Chrome extension that saves web articles directly to raw/

Related

• llm-wiki
• lint-protocol
• obsidian-web-clipper
• yt-llm-wiki-knowledge-system