Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/SamBleed/opencode-obsidian/llms.txt

Use this file to discover all available pages before exploring further.

The wiki-ingest skill is the primary intake valve for Bunker OS. Any file, URL, or image you drop into .raw/ becomes a set of structured wiki pages — one source summary, entity pages for every person/org/product mentioned, concept pages for significant ideas, and updates to the master index, operation log, and hot cache. A single source typically touches 8–15 wiki pages. The .raw/ folder is immutable: sources are never modified, only read.

​
Trigger Phrases

ingest [file]
ingest all of these
process this source
add this to the wiki
read and file this
batch ingest
ingest this url

​
Destination by Source Type

Resource typeFolderExample
Entity (person, org, product)wiki/entities/wiki/entities/OpenRouter.md
Technical conceptwiki/concepts/tech-stack/wiki/concepts/tech-stack/PostgreSQL.md
General concept (idea, pattern, framework)wiki/concepts/wiki/concepts/BM25.md
External sourcewiki/sources/wiki/sources/trivy-security-report.md

​
Preparing Sources

Drop any of the following into .raw/ before running ingest:
  • Markdown files — articles, notes, transcripts, documentation exports
  • URLs — the skill fetches and optionally cleans with defuddle (strips ads and nav, saves 40–60% tokens)
  • Images — .png, .jpg, .webp, .svg, .avif; the skill uses OpenCode’s vision capability to OCR and describe the image before ingesting
  • PDFs — processed via WebFetch or direct read
For URL ingestion, the skill derives a slug from the URL path, saves the fetched content to .raw/articles/[slug]-[YYYY-MM-DD].md with source frontmatter, and proceeds from there. Images are saved to .raw/images/[slug]-[YYYY-MM-DD].md as description files, with the original copied to _attachments/images/.

​
Single Source Ingestion Workflow

1

Read the source completely

The skill reads the full source file without skimming. For large sources it checks the manifest at .raw/.manifest.json first — if the file hash matches a previously ingested entry, it reports “Already ingested (unchanged)” and stops unless you say force ingest.
2

Create source summary in wiki/sources/

A structured summary page is created using the source frontmatter schema. Required fields: type, source_type, author, date_published, url (if applicable), confidence (high / medium / low), and a prose summary of what the source contributes.
3

Create or update entity pages

Every person, organization, product, and repository mentioned in the source gets a page in wiki/entities/. The skill checks the index first — if the entity already exists, the page is updated rather than duplicated.
4

Create or update concept pages

Significant ideas, frameworks, and patterns become pages in wiki/concepts/ or wiki/concepts/tech-stack/. Only concepts with enough standalone substance are created; minor mentions are captured in the source summary instead.
5

Update index, hot cache, and log

  • wiki/index.md — new entries added to the correct sections
  • wiki/hot.md — recent context updated so the next session starts fast
  • wiki/log.md — new entry prepended at the top: 
    ## [YYYY-MM-DD] ingest | Source Title
- Source: .raw/articles/filename.md
- Summary: [[Source Title]]
- Pages created: [[Page 1]], [[Page 2]]
- Pages updated: [[Page 3]], [[Page 4]]
- Key insight: One sentence on what is new.
6

Check for contradictions with existing pages

After all pages are created, the skill scans for conflicts with existing wiki content. When a contradiction is found, a [!contradiction] callout is added to both the new page and the conflicting existing page — the user decides which claim is authoritative. Old claims are never silently overwritten.

​
Required Page Frontmatter

Every wiki page created by the ingest skill uses this frontmatter schema: 
---
type: [entity|concept|source]
title: "Page Title"
domain: [tech-stack|design|ai|security|etc]
created: YYYY-MM-DD
updated: YYYY-MM-DD
tags: []
status: [active|mature|seed]
related: []
sources: []
---

# Title

## Definition
[One-line description]

## How It Works
[Mechanism]

## Example
```language
code here

​
Source

  • URL or [[wikilink to source]]

For source pages, additional fields are required:

```yaml
---
type: source
source_type: [article|paper|documentation|transcript|report]
author: "Author Name"
date_published: YYYY-MM-DD
url: "https://..."
confidence: [high|medium|low]
---

​
Batch Ingestion

When you say ingest all of these, the skill processes multiple files in sequence with a deferred cross-referencing pass:
  1. Lists all files to process and confirms with you before starting
  2. Processes each source following the single-ingest flow, deferring inter-source cross-references
  3. After all sources are processed, runs a cross-reference pass looking for connections between the newly ingested sources
  4. Updates wiki/index.md, wiki/hot.md, and wiki/log.md once at the end (not per-source)
  5. Reports: “Processed N sources. Created X pages, updated Y pages. Here are the key connections I found.”
For 30+ sources, the skill checks in with you after every 10 processed.

​
Delta Tracking

The skill tracks ingested files in .raw/.manifest.json: 
{
  "sources": {
    ".raw/articles/article-slug-2026-04-08.md": {
      "hash": "abc123",
      "ingested_at": "2026-04-08",
      "pages_created": ["wiki/sources/article-slug.md", "wiki/entities/Person.md"],
      "pages_updated": ["wiki/index.md"]
    }
  }
}
Before ingesting any file, a hash (sha256sum on Linux, md5sum otherwise) is computed and checked against the manifest. If the hash matches, the file is skipped. After successful ingestion, the manifest is updated. Use force ingest to bypass delta checking.

​
Cross-Referencing

The ingest skill uses [[wikilinks]] to build a connected knowledge graph:
  • Every entity and concept mentioned in a source summary is linked to its wiki page
  • Related pages are added to each new page’s related frontmatter field
  • Contradiction callouts link bidirectionally between conflicting pages
  • The master wiki/index.md serves as the entry point for the entire graph
The .raw/ directory is immutable. The ingest skill reads from it but never writes to it. Your original source documents always remain in their original state. To re-ingest a modified file, update it in .raw/ and run ingest [file] again — or force ingest to bypass the manifest hash check.

​
Context Window Discipline

The ingest skill is designed to stay within token budget:
  • Reads wiki/hot.md first — if it contains relevant context, existing pages are not re-read
  • Reads wiki/index.md to find existing pages before creating new ones
  • Reads only 3–5 existing pages per ingest session
  • Uses obsidian-vault_patch_note for surgical edits rather than reading and rewriting full files
  • Keeps wiki pages under 100–300 lines; splits pages that grow beyond 300 lines

Build docs developers (and LLMs) love

Get started for freeTalk to us