clip

name	clip
description	Inbox capture that matches Web Clipper output. Clips URLs or inline text as clipping notes with LLM-generated titles. Batch-capable for multiple URLs. No classification, no area/project assignment — triage handles that.
argument-hint	<url1> [url2 url3 ...] or <inline text to capture>
user-invocable	true
context	fork
allowed-tools	mcp__plugin_para-obsidian_para-obsidian__para_create, mcp__plugin_para-obsidian_para-obsidian__para_commit, mcp__plugin_para-obsidian_para-obsidian__para_template_fields, mcp__firecrawl__firecrawl_scrape, mcp__youtube-transcript__get_video_info

Clip

Inbox capture that produces clipping notes matching Obsidian Web Clipper output — real page titles, raw fetched content, <!-- highlights:0 --> marker. No classification, no Layer 1 formatting, no area/project assignment. Everything gets organized later by /para-obsidian:triage.

Design philosophy: ADHD-friendly capture with Web Clipper parity. Save a link, get a real note — not a stub. Capture now, organize later.

---

Step 0 — Discover Template Metadata

Before processing any input, query the clipping template for its current structure:

para_template_fields({ template: "clipping", response_format: "json" })

Extract from response:

• validArgs → which args to pass to para_create (e.g., source, clipped, domain)
• creation_meta.dest → destination folder (e.g., "00 Inbox")
• creation_meta.contentTargets → which section heading accepts content injection (e.g., ["Content"])
• creation_meta.titlePrefix → emoji prefix (e.g., "✂️ ")

Use these discovered values throughout the skill instead of hardcoding them. If the template changes, this skill auto-adapts.

---

Input Detection

Parse $ARGUMENTS to determine mode:

Step 1: Extract all URLs from arguments (regex: https?://\S+).

Result	Mode
1+ URLs found	URL mode — create clipping notes
No URLs found	Inline text mode — create clipping note

Step 2: Deduplicate URLs before processing (URL mode only).

---

Mode 1: URL Capture (Clipping Notes)

For each URL, create a clipping note that triage recognizes as itemType: "clipping".

For each URL:

Step 1 — Extract domain:

Strip www. prefix from hostname:

github.com/anthropics/claude-code  →  domain: "github.com"
www.kentcdodds.com/blog/aha        →  domain: "kentcdodds.com"

Step 2 — Fetch content:

The clip skill should produce notes identical to what the Obsidian Web Clipper creates — including page content in the Content section. Fetch content for every URL.

URL Pattern	Tool	Extract
youtube.com, youtu.be	get_video_info + firecrawl_scrape	Title from video info, description + embed from Firecrawl
x.com, twitter.com	Skip (no tool available yet)	Use domain as title, empty content
Everything else	firecrawl_scrape	metadata.title + markdown body

YouTube URLs:

get_video_info({ url: "<url>" })
→ extract title from response

firecrawl_scrape({ url: "<url>", formats: ["markdown"], onlyMainContent: true })
→ extract markdown as page content (video description, chapters, links)

Combine into content: video embed ![](url) followed by the scraped description.

General URLs (Firecrawl):

firecrawl_scrape({
  url: "<url>",
  formats: ["markdown"],
  onlyMainContent: true
})
→ extract metadata.title as page title
→ extract markdown as page content

X/Twitter URLs: Skip fetching. Firecrawl is blocklisted for x.com. Use domain as title, no content. Future x-twitter plugin will handle this.

If fetch fails: Fall back to domain as title with empty content. Never block capture on fetch failure.

Step 3 — Derive title:

Use the fetched page title as the note title. If fetching was skipped or failed, fall back to domain-based title:

• Single github.com URL → title: github.com
• Two github.com URLs → titles: github.com - anthropics, github.com - modelcontextprotocol

For domain fallback, extract the slug from the first meaningful path segment (skip empty segments).

Step 4 — Create clipping note:

Use discovered values from Step 0 (validArgs for args, creation_meta.dest for dest, creation_meta.contentTargets[0] for content section heading):

para_create({
  template: "clipping",
  title: "<fetched-or-derived-title>",
  dest: "<discovered-dest>",
  args: {
    source: "<full-url>",
    clipped: "<YYYY-MM-DD>",
    domain: "<domain>"
  },
  content: {
    "<discovered-content-target>": "<fetched-content>\n\n<!-- highlights:0 -->"
  },
  response_format: "json"
})

The content parameter injects the fetched page content into the content target section discovered from creation_meta.contentTargets. The <!-- highlights:0 --> marker matches Web Clipper output format.

If no content was fetched (X/Twitter, or fetch failure), omit the content parameter entirely.

Process URLs sequentially — each fetch + para_create call completes before the next. Collect results for the report.

After all URLs created:

Step 5 — Batch commit:

para_commit({
  message: "Clip: <N> URLs to inbox",
  response_format: "json"
})

One commit for all notes. Never commit per-note.

---

Mode 2: Inline Text Capture

Capture free text from the conversation into a clipping note.

Step 1 — Generate title:

Generate a concise, descriptive title (under 60 characters) that captures the essence of the text. Use your judgment as an LLM — don't just truncate, summarize the core idea into a clear title.

Step 2 — Create clipping note:

para_create({
  template: "clipping",
  title: "<generated-title>",
  dest: "<discovered-dest>",
  args: {
    source: "conversation",
    clipped: "<YYYY-MM-DD>"
  },
  content: {
    "<discovered-content-target>": "<full-inline-text>\n\n<!-- highlights:0 -->"
  },
  response_format: "json"
})

The content parameter injects text into the content target section discovered from Step 0.

Defaults: source: "conversation". Zero-friction capture — no enum choices required. Triage or manual review can adjust later.

Step 3 — Commit:

para_commit({
  message: "Clip: <title>",
  response_format: "json"
})

---

Reporting

URL mode (batch):

Clipped N URLs to inbox:

  1. ✂️ AHA Programming             ← kentcdodds.com/blog/aha-programming
  2. ✂️ Claude Code                  ← github.com/anthropics/claude-code
  3. ✂️ x.com                        ← x.com/housecor/status/123456 (no content — X not supported yet)

Committed. Run /para-obsidian:triage to classify.

URL mode (single):

Clipped: ✂️ AHA Programming → 00 Inbox/✂️ AHA Programming.md
Committed. Run /para-obsidian:triage to classify.

Inline mode:

Clipped: <title> → 00 Inbox/<title>.md
Committed. Run /para-obsidian:triage to process.

---

Error Handling

Scenario	Action
Empty arguments	Report: "Nothing to clip. Pass URLs or text to capture."
Invalid URL (parse fails)	Skip that URL, report as warning, continue with remaining
Content fetch fails	Fall back to domain title + empty body. Report warning, continue.
para_create fails for one URL	Log error, continue with remaining URLs
All URLs fail	Report all failures, create nothing
para_commit fails	Report warning — notes still exist uncommitted
Duplicate URL in batch	Deduplicate before processing
Filename collision	para_create handles this automatically (appends number)

Soft failure philosophy: Never block on individual failures in batch mode. Create what you can, report what failed. Fetch failures degrade gracefully to domain-only stubs.

---

What This Does NOT Do

• No classification — triage determines area/project/template
• No Layer 1 formatting — triage handles content distillation
• No user prompts — zero gates, zero decisions
• No area/project assignment — triage handles this
• No X/Twitter content — future x-twitter plugin will handle this

---

Triage Compatibility

Notes created by this skill are fully compatible with /para-obsidian:triage:

Mode	type field	Triage detection	Triage action
URL	clipping	type === "clipping" + source is URL	Routes to analyze-web worker for classification
Inline text	clipping	type === "clipping" + source is NOT URL	Stays in inbox (manual review items)

Clipping notes with fetched content give triage a head start — it can classify from the content already present rather than re-fetching.

---

Examples

Single URL

/para-obsidian:clip https://kentcdodds.com/blog/aha-programming

Creates: 00 Inbox/✂️ AHA Programming.md (with full article content)

Batch URLs

/para-obsidian:clip https://github.com/anthropics/claude-code https://youtube.com/watch?v=abc123 https://kentcdodds.com/blog/aha-programming

Creates 3 notes in 00 Inbox/ with fetched titles and content, commits all at once.

YouTube URL

/para-obsidian:clip https://www.youtube.com/watch?v=dQw4w9WgXcQ

Creates: 00 Inbox/✂️ <Video Title>.md (title only, no transcript — triage fetches that)

Inline capture (mid-conversation insight)

/para-obsidian:clip The key insight is that Layer 1 captures raw content, Layer 2 bolds the most important 10-20%, and Layer 3 highlights the top 10% of bold passages.

Creates: 00 Inbox/Progressive Summarization Layer Strategy.md (LLM-generated title)

Inline capture (quick thought)

/para-obsidian:clip Check if React Server Components solve the waterfall problem for GMS checkout flow

Creates: 00 Inbox/RSC for GMS Checkout Waterfall Problem.md (LLM-generated title)

---

Completion Signal

After reporting results, emit a structured completion signal so the brain orchestrator can parse the outcome:

• URL mode (batch): SKILL_RESULT:{"status":"ok","skill":"clip","summary":"Clipped N URLs to inbox"}
• URL mode (partial failures): SKILL_RESULT:{"status":"partial","skill":"clip","clipped":N,"failed":M}
• Inline mode: SKILL_RESULT:{"status":"ok","skill":"clip","summary":"Clipped inline text to inbox"}
• All failed: SKILL_RESULT:{"status":"error","skill":"clip","error":"All URLs failed to clip"}
• Empty args: SKILL_RESULT:{"status":"error","skill":"clip","error":"Nothing to clip"}