clip-tweet

name	clip-tweet
description	Ingest a single Twitter/X tweet URL into inbox/tweets/, fetch its content via fetch-tweet.sh, and generate an organized version with frontmatter and Markdown body in _organized/. Combines `rill clip` with the tweet branch of /distill Phase 2 in one shot. Use when the user supplies a single tweet URL to capture.

/clip-tweet — Tweet Ingestion

Conduct all conversation with the user — and write all generated output — in the language defined by .claude/rules/personal-language.md (or the user's input language if that file is absent). Follow the language rules in full — exceptions and translation quality are defined in the Language Rules of .claude/rules/rill-core.md and the vault's personal-*.md overrides, never restated per skill. The English instructions below are for skill clarity, not for output style.

Tool references in this skill (shell, WebFetch) describe intent, not Claude-specific tool calls. Each harness should map them to its native equivalent — Claude Code uses its built-in tools as named; Codex CLI uses apply_patch / shell / --search etc. as appropriate.

Ingests a Twitter/X tweet URL into inbox/tweets/, fetches the content via fetch-tweet.sh, and generates an organized version in _organized/. A skill that runs rill clip + /distill Phase 2 (tweet) in one shot.

Arguments

$ARGUMENTS — Tweet URL (e.g. https://x.com/user/status/12345)

Steps

Step 1: URL Validation and Normalization

1. If the argument is empty, display "Please specify a tweet URL" and exit
2. Confirm the URL matches the pattern https://(x.com|twitter.com)/*/status/*
3. Strip tracking parameters (?s=, &t=, &ref_src=, etc.)
4. Extract screen_name and tweet_id from the URL

Step 2: Create File in inbox/tweets/

1. If a file with the same tweet-id already exists in inbox/tweets/, report the duplicate and skip
2. Create the file with rill mkfile (to guarantee timestamp accuracy):

rill mkfile inbox/tweets --slug {screen_name} --field "source-type=tweet" --field 'url="{normalized URL}"' --field 'tweet-id="{tweet_id}"'

3. Use the output path in subsequent steps

Step 3: Fetch Tweet Data (Script Execution)

Run the following in a shell:

bash plugins/twitter/fetch-tweet.sh "{normalized URL}"

The script outputs structured YAML to stdout. Use this output in subsequent steps.

If the script fails (exit code ≠ 0), report the error message from stderr and leave the file created in Step 2 (it can be reprocessed by /distill).

Step 3.5: URL Enrichment (only if urls is non-empty)

Only run this if the Step 3 output has urls: and it is not an empty array [].

1. For each URL in the urls list, run the following in a shell:

   bash plugins/twitter/fetch-url-meta.sh "{url}"
   

2. The script returns title and description as YAML
3. If the description makes sense as a summary (non-empty, not generic) → use it as-is
4. If the description is empty or insufficient → use WebFetch to retrieve the linked page body and create a 3-5 line summary. If WebFetch also fails, record only the title
5. Save the result and use it in Step 4 under "## Linked Content"

Step 3.7: Language and Relevance Detection

Language Rules

• tweet-lang: ja → Both the body and AI-generated text (summary, linked content) are in Japanese
• tweet-lang: en → Record the tweet body in English. AI-generated text (summary, linked content) follows the system output language
• Other / mixed → Preserve the original language; AI-generated text follows the system output language

Engagement Signal

Interpret engagement-save-ratio and engagement-rate from the script output and add engagement-signal to the frontmatter:

• engagement-save-ratio ≥ 1.0 → high-save (bookmarks ≥ likes; practical tools / reference content)
• engagement-rate ≥ 5.0 → viral (engagement rate over 5%; viral content)
• Neither applies → omit engagement-signal

Related Projects

Match the tweet content against projects in projects/*/_project.md (ADR-080: projects moved from the old knowledge/projects/ flat layout), and if related project IDs exist, add relevance-to: [id1, id2] to the frontmatter (max 2). Use the directory names (= IDs) under projects/ and the name field in each projects/{slug}/_project.md as references. Omit if none apply.

Step 4: Generate Organized Version in _organized/

1. Read taxonomy.md to check existing tags
2. Write to inbox/tweets/_organized/{same filename}

frontmatter

---
created: {same as Step 2}
source-type: tweet
original-file: inbox/tweets/{filename}
url: "{normalized URL}"
tweet-id: "{tweet_id}"
tweet-author: {tweet-author from script output}
tweet-author-name: {tweet-author-name from script output}
tweet-date: {tweet-date from script output}
tweet-likes: {number}
tweet-retweets: {number}
tweet-bookmarks: {number}
tweet-views: {number}
tweet-type: {article | tweet}
tags: [{up to 3, selected from taxonomy.md}]
engagement-signal: {high-save | viral — only if applicable}
relevance-to: [{project ID — only if applicable}]
---

Body (for normal tweets)

{the text content as-is}

---

**{tweet-author-name} ({tweet-author})** · {tweet-date}

👍 {tweet-likes} · 🔁 {tweet-retweets} · 🔖 {tweet-bookmarks} · 👁 {tweet-views}

If urls exist, append:

## Links
- [{display}]({url})

If urls exist and Step 3.5 enrichment was retrieved, append after "## Links":

## Linked Content

### [{title}]({url})
{description or WebFetch-based summary}

If media exist, append:

## Media
- 📷 Image ({width}×{height})
- 🎥 Video ({duration}s, {width}×{height})

If a quote exists, append:

## Quoted Tweet
> {quote.text}
>
> — {quote.author-name} ({quote.author}) [original]({quote.url})

Body (for Articles)

# {article.title}

by {tweet-author-name} ({tweet-author}) — {tweet-date}

## Summary

{AI-generated summary: 3-5 key points}

## Body

{full text converted from article.blocks to Markdown}

Article block conversion rules:

• type: "header-two" → ## {text}
• type: "unstyled" → paragraph (separated by blank lines)
• type: "ordered-list-item" → 1. {text}
• type: "unordered-list-item" → - {text}
• type: "blockquote" → > {text}
• type: "atomic" → --- (horizontal rule)
• styles Bold → **{text}**, Italic → *{text}* (applied to the range specified by offset/length)
• entity-map LINK type → convert to [text](url)
• entity-map TWEET type → record as > [Embedded tweet](https://x.com/i/status/{tweet-id})

Step 5: Update .processed

1. Append {filename}:organized to inbox/tweets/.processed
2. Report the result:
   • Path of the created file
   • Tweet type (normal / Article)
   • Number of links and media
   • Body summary (first 1-2 sentences)

Rules

• Do not modify the original file in inbox/tweets/ (create a thin file with frontmatter only, and expand the full text in _organized/)
• Reference taxonomy.md when assigning tags
• Division of responsibility with /distill: /distill batch-processes unprocessed files; /clip-tweet immediately processes a single tweet