agentleFS
kellykampenkellykampen/agent-skills/skills/personal/social-capture/SKILL.md

social-capture

Skill2 starsChanged 15 days agoMITView on GitHub

Capture an X/Twitter post or YouTube video into an Obsidian note with a summary, takeaways, and source metadata. Use when the user asks to save a social URL or analyze a video's visuals.

The file

We split this skill into the rules it gives an agent

8 rules. About 0 are common, 5 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Read the source, synthesize useful notes, and save them through the included scripts.

    line 10
  2. Distinctive:

    Read references/configuration.md before the first save or when migrating an older installation.

    line 16
  3. Very specific:

    Run python3 scripts/extract-youtube.py "URL".

    line 26
  4. Very specific:

    Read the transcript and create summary, key_points, optional how_to_apply, category, and tags.

    line 29
  5. Distinctive:

    If the transcript is unavailable, use available metadata and clearly state that the note is based on metadata, not a watched or transcribed …

    line 31
  6. Very specific:

    Keep only useful frames in the user's attachment folder, with a video-ID prefix on each filename.

    line 44
  7. Distinctive:

    Use an available browser tool to read the full post, especially long-form posts.

    line 53
  8. Distinctive:

    Use the user's selected browser and existing access.

    line 54

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Social capture

Read the source, synthesize useful notes, and save them through the included scripts. Capture on demand when the user asks to save a URL. A pasted URL alone does not authorize saving it or sending messages. Reply with the actual saved path and a short summary.

Local configuration

Read references/configuration.md before the first save or when migrating an older installation. Destinations, category lists, and application headings belong in local configuration, outside the installed skill folder.

The savers require an explicit vault destination. They never guess a vault name. --dry-run --out-dir /path/to/samples renders notes without a vault configuration. Dry runs create sample files in that output folder and do not write to the vault.

YouTube

  1. Run python3 scripts/extract-youtube.py "URL". It returns JSON metadata, a plain-text transcript, and transcript_available. It does not save a note. yt-dlp must be on PATH, or selected with YT_DLP_BIN.
  2. Read the transcript and create summary, key_points, optional how_to_apply, category, and tags. Merge these with the extracted metadata in a JSON file. If the transcript is unavailable, use available metadata and clearly state that the note is based on metadata, not a watched or transcribed video.
  3. Run python3 scripts/save-youtube-note.py --json --in /path/to/record.json.

For visuals, use watch mode only when requested or when visual analysis is part of the user's task:

python3 scripts/watch_youtube.py "URL" --max-frames 30

Watch mode also requires ffmpeg and ffprobe on PATH. Its JSON includes the transcript, PNG frame paths, and frames_dir. Inspect the frames using an available image tool. Keep only useful frames in the user's attachment folder, with a video-ID prefix on each filename. Remove the returned temporary frames folder after inspection or copying. A watch_error means visual analysis failed; report that without claiming to have watched.

Browser cookies are off by default. Set YT_DLP_COOKIES_BROWSER only when needed and authorized. See references/platforms.md for extraction details.

X/Twitter

  1. Use an available browser tool to read the full post, especially long-form posts. Use the user's selected browser and existing access. If an authorized Finch CLI is available, its JSON output can supply metadata or act as a fallback. Check the returned text for truncation rather than assuming every API response is complete.
  2. Synthesize a title, summary, key points, and optional application ideas. Preserve the source text separately. If only partial text is available, set truncated: true and say so in the response. Do not introduce a paid scraping service automatically.
  3. Run python3 scripts/save-x-post.py --json --in /path/to/record.json.

Required X fields are tweet_id and url; the date defaults to today. For example:

{
  "tweet_id": "1234567890",
  "url": "https://x.com/example/status/1234567890",
  "title": "A useful technique",
  "author": "Example",
  "author_handle": "@example",
  "category": "research",
  "tags": ["research"],
  "summary": "What the post explains.",
  "key_points": ["The useful takeaway."],
  "full_text": "The original source text.",
  "how_to_apply": ["A concrete next step."],
  "truncated": false
}

YouTube records require video_id; they may also include title, channel, channel_handle, published, duration, playlist, and the synthesis fields above. The URL is derived from the video ID if absent. The savers default missing titles to Untitled, so provide a useful title when metadata is available.

Saving behavior

Both savers accept --json, --in FILE, --dry-run, and --out-dir DIRECTORY. They emit the actual output path. Frontmatter strings and lists are escaped as YAML. Notes use CATEGORY/DATE-TITLE-SOURCE_ID.md; different source IDs cannot collide just because titles match. A matching existing filename causes an error rather than an overwrite. Review existing notes before choosing a separate update or migration action.

Bulk jobs may call these scripts with the same local configuration and record format. This skill does not install schedules, send Telegram messages, or require external pipeline scripts. Articles and other platforms are outside this skill's current scope.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.