| name | inbox |
| description | Use when the user wants to drop a source into the LLM Wiki inbox WITHOUT parsing it yet. Trigger phrases: "wiki:inbox <url|path>", "inbox this", "drop in inbox", "save to wiki later", "wiki: queue <url>", or when the user pastes a URL/file and says "for later". Stages the input in raw/_inbox/ so the wiki:parse skill (or the cron-ingest launchd job) can process it asynchronously. This skill does NOT call defuddle/yt-dlp/whisper — it only stages. Heavy parsing happens later.
|
wiki:inbox — stage a source for later processing
Lightweight skill. Stages a source in raw/_inbox/ without touching sources/, entities/, or concepts/. The wiki:parse skill (manual) or cron-ingest.sh launchd job (every 10 min) does the actual work later.
Step 1 — Locate the wiki
Find CLAUDE.md with frontmatter type: schema and scope: wiki:
- Project-local: walk up from
pwd
- Vault root (Obsidian):
$VAULT_ROOT/**/wiki/CLAUDE.md
- Memory: single known wiki → use it
- Multiple candidates: ask the user
Confirm by reading frontmatter.
Step 2 — Classify the input
| Input | How to stage |
|---|
| URL (any web link, YouTube, etc.) | Write a .url file inside raw/_inbox/ with the URL on its own line and slug-derived filename |
| Local file path | Copy (do not move — leave original) into raw/_inbox/<basename> |
| Paste of raw text/markdown | Save as raw/_inbox/<slug>-<YYYY-MM-DD>.md with the text body |
| Multiple URLs in one message | One .url file per link |
Filename slug rules
- kebab-case, lowercase
- For URLs: derive from the URL path (e.g.,
youtube.com/watch?v=ABC → abc.url, anthropic.com/news/foo-bar → foo-bar.url)
- For YouTube: use the video ID where possible (
<video-id>.url or <title-slug>.url) — easier for yt-dlp lookup later
- If a slug collision exists, append
-2, -3 etc.
Step 3 — Stage and confirm
- Create
raw/_inbox/<slug>.<ext> with the content
- Print a short confirmation: filename, kind, queue size (
ls raw/_inbox/ | wc -l)
- Tell the user when it will be processed:
- "launchd
com.wiki-parse runs every 10 min — will pick this up on next tick"
- If they want it parsed now: "say
wiki:parse raw/_inbox/ to run immediately"
Step 4 — Log (optional, lightweight)
Do NOT append to log.md here. The wiki:parse skill writes the proper ingest log entry after processing. An inbox-only entry would just create noise.
Hard rules
- Never call defuddle, yt-dlp, whisper, or any heavy network/CPU work in this skill
- Never write to
sources/, entities/, concepts/, comparisons/, or questions/
- Never edit
log.md from this skill
- Never delete or move existing files out of
_inbox/ — only add new ones
- If the same filename already exists in
_inbox/, ask the user before overwriting
Why this exists
Parsing is expensive (LLM context, whisper minutes, defuddle network). The user often wants to capture now, process later — like browser bookmarks or read-later apps. This skill is the wiki equivalent: 5-second drop-off, nothing more. Heavy lifting is async via launchd.
