---
source: raw/wiki-ingest-prompt.md
content_hash: 29f6a0914d428cf62eeb2a832d6b3048edf3e1d16aa7714c09f686377394d128
namespace: work
last_synced: 2026-05-31
confidence: 0.9
tags: [personal-os, wiki, obsidian, agent-prompts]
---

# Wiki Ingest — Process & Prompt

The wiki-ingest agent keeps `~/obsidian/wiki/` synchronised with external
source files exposed as symlinks in `~/obsidian/raw/`. It runs daily at
22:00 via the `personal.os.wiki-ingest` launchd agent, which invokes
`claude -p` as a local subprocess (Hermes can't reach the local
filesystem). See [[personal-os-state-2026-04-27]] Decision 1.

## How sync works

1. For every file in `~/obsidian/raw/`, compute SHA256 via
   `shasum -a 256 <path> | awk '{print $1}'`.
2. Search `~/obsidian/wiki/` for any `.md` with frontmatter
   `source: raw/<filename>`.
3. If a page exists AND `content_hash` in frontmatter matches → skip.
4. Otherwise → queue for synthesis.
5. Broken symlinks → skip silently, do not create a page.

If the queue is empty after the scan, output
"Wiki is up to date. Nothing to ingest." and stop.

## Synthesis rules

- **Never copy source files verbatim.** Extract key facts, decisions,
  and structure. Keep each page under 800 words.
- Name the wiki file **by topic, not by source filename**
  (e.g. `personal-os-schema.md`, not `schema.md`).
- If a topic is large, split into multiple linked pages.
- Cross-reference related pages with `[[double brackets]]`.
- Existing page found → update in place (preserving the file path).

## Frontmatter format

```markdown
---
source: raw/<filename>
content_hash: <sha256>
namespace: work
last_synced: <YYYY-MM-DD>
confidence: 0.8
tags: [tag1, tag2]
---
```

`last_synced` and `content_hash` MUST be updated after every successful
write.

## Confidence scale

| Value | Meaning |
|-------|---------|
| 1.0   | Human-written notes in `work/`, `personal/`, `family/`. **Never edit these.** |
| 0.9   | Highly structured, authoritative source (e.g. schema files) |
| 0.8   | Clear single source (default) |
| 0.6   | Inferred or partial content |

## Hard rules

- Do not edit any wiki page with `confidence: 1.0` — those are
  human-authored.
- Do not process `namespace: family` files unless explicitly instructed.
- Broken `raw/` symlinks are skipped, not removed.
- Always update `content_hash` and `last_synced` after writing.

## Output format

At the end of a run, print:

```
Wiki Ingest — <date>

Processed: N files
Created: N new pages
Updated: N pages
Skipped: N (unchanged)

Pages updated:
- wiki/<filename> (source: raw/<source>)
```

## Why file-based, not DB-based

Original design (v3) used `file_references` + `wiki_pages` Postgres
tables with embeddings. That was dropped in favour of file-frontmatter
hashing once Obsidian became the canonical vault. The vault is now
self-contained, portable, and version-controlled via the NAS git remote
without any DB dependency. See [[personal-os-state-2026-04-27]]
Decision 8 and Decision 2 (`obsidian-mcp` replaced `personal-wiki`
MCP).

## Related
[[personal-os-schema]] [[personal-os-state-2026-04-27]] [[personal-os-catchup-plan-2026-04-27]]