kb-ingest

/kb-ingest -- Knowledge Base Ingest

You ingest raw source material into an id8Labs knowledge base for later compilation.

Arguments: $ARGUMENTS

---

Route by Arguments

Parse $ARGUMENTS:

Input	Action
<kb-name> <file-path>	Ingest a local file
<kb-name> <url>	Fetch and ingest a web page (uses defuddle)
<kb-name> --paste	Ingest pasted content from the user's next message
<kb-name> status	Show raw/ inventory (processed vs unprocessed)
(empty)	Show help

---

Setup

1. Read ~/Development/id8/knowledge/manifest.json
2. Locate the target KB
3. Check KB type: if mirror, reject ingest (read-only). If read_only_raw, reject ingest to raw (suggest manual addition to source repo).
4. Read ~/.claude/skills/knowledge/SKILL.md for article templates

---

Ingest Process

For Local Files

1. Read the source file
2. Generate a slug from the content title or filename (kebab-case, max 40 chars)
3. Copy to {kb}/raw/{slug}.md with frontmatter:

---
source_url: "local"
source_path: "{original file path}"
source_type: {article | paper | repo | dataset | image | reference}
ingested: YYYY-MM-DD
processed: false
---

{file contents}

For URLs

1. Use the defuddle skill to fetch and convert the web page to clean markdown
2. Generate a slug from the page title
3. Save to {kb}/raw/{slug}.md with frontmatter:

---
source_url: "{url}"
source_type: article
ingested: YYYY-MM-DD
processed: false
---

{defuddled markdown content}

1. If the page has images, note their URLs in the frontmatter for potential later download

For Paste

1. Prompt user: "Paste the content you want to ingest."
2. Accept the content
3. Ask for a title if not obvious from content
4. Save to {kb}/raw/{slug}.md with frontmatter (source_url: "paste")

---

Post-Ingest

After saving the raw file:

1. Report: "Ingested {filename} to {kb}/raw/. Source type: {type}."
2. Quick scan: extract 3-5 key topics/claims from the source
3. Check against existing wiki/_index.md for overlap: "This relates to existing articles: [[concept-x]], [[concept-y]]"
4. Suggest: "Run /kb-compile {kb-name} to process this into wiki articles."