ideaverse-capture

name	ideaverse-capture
description	Conversational ingestion router. Accepts a YouTube URL, plain HTTP URL, a filename in 40-raw/web-clips/, or a free-text "I learned X" note. Routes to the right handler, refreshes 10-cortex/ via cortex-compile + cortex-connect, and logs to today's daily note. Invoke with /ideaverse-capture <url-or-text>.
user_invocable	true
argument-hint	<youtube-url | http-url | filename-in-40-raw/web-clips/ | 'I learned...'>

ideaverse-capture

The ingestion entrypoint for your vault. One command routes any source into your cortex.

When to use

• Paste a YouTube URL to capture a video as a research note in 40-raw/youtube/
• Paste any web URL to fetch, process, and store it in 40-raw/web-clips/
• Pass a filename after Obsidian Web Clipper drops a raw file into 40-raw/web-clips/
• Type "I learned X" or any free-text insight to synthesize it directly into 10-cortex/

Routing matrix

Input type	Detection	Handler
YouTube URL	matches youtu.be/, youtube.com/watch?v=, or youtube.com/shorts/	/yt-light-research
File in 40-raw/web-clips/	no ://, has .md extension, or filename exists in 40-raw/web-clips/	/web-clip-report
Plain HTTP/S URL (not YouTube)	starts with https?://, not a YouTube domain	fetch + /web-clip-report
Free text	none of the above	direct cortex synthesis

---

Step 0: Load paths

Read 00-agentic-OS/paths.json. Use vaultRoot for all file operations. If paths.json is missing, halt: "paths.json not found. Run /ideaverse-os build --phase=workflow to generate it."

---

Step 1: Detect source type

Inspect $ARGUMENTS (everything after /ideaverse-capture):

1. Matches (?:youtu\.be/|youtube\.com/watch\?v=|youtube\.com/shorts/) -> YT URL. Go to Step 2a.
2. No ://, ends in .md, OR file exists at {vaultRoot}/40-raw/web-clips/<arg> -> Web clip file. Go to Step 2b.
3. Starts with https?:// (not YouTube) -> Plain URL. Go to Step 2c.
4. Anything else -> Free text. Go to Step 2d.

---

Step 2a: YouTube URL

Delegate to /yt-light-research <url>.

That skill fetches the transcript, synthesizes a research note, and writes it to 40-raw/youtube/. After it completes, take the written file path and continue to Step 3.

If /yt-light-research is not available: "yt-light-research skill not found in 60-skills/. Re-run npx ideaverse-os init --force to restore bundled skills."

---

Step 2b: Web clip file

File must exist at {vaultRoot}/40-raw/web-clips/<filename>.

If it doesn't exist: "File not found at 40-raw/web-clips/{arg}. Did you mean to paste the URL directly?"

Delegate to /web-clip-report <filename>. That skill rewrites the file in-place with structured frontmatter, TL;DR, and key takeaways.

After it completes, continue to Step 3 with the file path.

---

Step 2c: Plain URL (not YouTube)

1. Use available browser tools (Playwright MCP browser_navigate + browser_snapshot, or any equivalent fetch tool) to retrieve the page content.
2. Extract the main article text. Discard navigation, ads, footers, cookie banners.
3. Derive a slug from the page title (kebab-case, ASCII, max 60 chars).
4. Write a temporary raw file to {vaultRoot}/40-raw/web-clips/YYYY-MM-DD-<slug>.md with:
   • source_url: the original URL
   • date: today's date
   • The extracted body as plain markdown
5. Delegate to /web-clip-report YYYY-MM-DD-<slug>.md.

If no browser/fetch tool is available: "No fetch tool found. Save the page manually to 40-raw/web-clips/ and run /ideaverse-capture <filename> to process it."

Continue to Step 3 with the processed file path.

---

Step 2d: Free text synthesis

For "I learned X", "Key insight: ...", or any plain-text insight without a source URL:

1. Ask: "What topic tag should this go under? (e.g., ai, career, embedded, productivity, tools)"
2. Ask: "One-line title for this insight?"
3. Derive a slug from the title.
4. Write or update a cortex article at {vaultRoot}/10-cortex/<topic>/<slug>.md using the cortex-compile article format:

---
title: "<title>"
description: "<insight in one sentence>"
topic: "<topic>"
sources: [plain-text capture]
created: YYYY-MM-DD
last_compiled: YYYY-MM-DD
confidence: low
---

# <title>

<insight as 1-2 paragraph article>

## Sources

- Plain-text capture, YYYY-MM-DD

Skip to Step 4 (cortex-connect + daily log). Do not run cortex-compile (the article was written directly).

---

Step 3: Run cortex-compile

After ingestion via Steps 2a-2c, refresh the cortex:

1. Read the written note's topic: frontmatter field.
2. Invoke /cortex-compile <topic> to compile or refresh the relevant cortex articles.
3. Note the articles created or updated.

If /cortex-compile is unavailable, surface as an action item in the report: "Run /cortex-compile to fold this into your cortex."

---

Step 4: Run cortex-connect

Invoke /cortex-connect to scan 10-cortex/ for unlinked mentions and add wikilinks.

If unavailable: "Run /cortex-connect to add wikilinks after reviewing the new article."

---

Step 5: Write to daily log

Append a one-line entry to today's daily note. Determine the path:

• If {vaultRoot}/70-daily/YYYY-MM-DD.md exists, append to it.
• If it doesn't exist, create it with a minimal header and append.

Entry format:

- [capture] <source-type>: [[<relative-path-to-note>]] -- <one-line title or description>

---

Step 6: Report

Captured: <source title>
Source type: <youtube | web-clip | url | plain-text>
Raw note: <path if applicable>
Cortex: <cortex article(s) updated or created>
See also: <related articles if found by cortex-connect>
Daily log: updated 70-daily/YYYY-MM-DD.md

---

Quality rules

• Never invent content. If the transcript or page doesn't say it, don't write it.
• ASCII only. No em dashes, smart quotes, or Unicode bullets.
• Paths always come from paths.json, never hardcoded.
• If a delegated skill is missing, report it clearly. Never silently skip.
• One capture per invocation. Batch capture is not supported in v1.