Gobbler

Gobbler

Use Gobbler when the task begins with external content that must become markdown before an agent can reason over it: YouTube videos/playlists, URLs/webpages, PDFs/DOCX/PPTX/XLSX files, audio/video, batches of inputs, local files, or intentionally selected browser tabs.

Do not use this skill for ordinary summarization when the source content is already present in the conversation.

Default behavior

• Prefer the gobbler CLI over ad hoc converters so outputs are consistent and reproducible.
• Prefer writing converted content to an explicit .md output file instead of only printing it in chat.
• Verify that the output file exists, has YAML frontmatter, and has a non-empty markdown body before reporting success.
• Report the source, output path, and command used.
• If processing multiple inputs, use deterministic filenames and summarize successes/failures.
• Run gobbler doctor --json before document, webpage, browser, troubleshooting, or large batch workflows.
• Keep private converted content local unless the user explicitly asks to send or publish it.

Quick routing

User intent/input	Command family	Details
YouTube video, playlist, captions, subtitles	gobbler youtube, gobbler batch youtube-playlist	references/youtube.md, references/batch.md
Audio/video file, podcast, recording, voice memo, interview	gobbler audio	references/audio.md
PDF, DOCX, PPTX, XLSX, document text extraction	gobbler document	references/document.md
URL, webpage, article, web scrape, selector extraction	gobbler webpage	references/webpage.md
Multiple files, URL lists, playlist, bulk conversion	gobbler batch ...	references/batch.md
Authenticated browser tab/session content	gobbler browser ...	Load gobbler-browser skill
Install, service readiness, Docker, ffmpeg, connection refused	gobbler doctor --json	Load gobbler-setup skill if needed

Common commands

# Check CLI and service readiness
gobbler doctor --json

# YouTube transcript, no Docker required
gobbler youtube "https://youtube.com/watch?v=VIDEO_ID" -o ./outputs/video.md

# Deterministic visual overview, interpreted by the calling agent
gobbler youtube "https://youtube.com/watch?v=VIDEO_ID" \
  --frames-only --frames 8 -o ./outputs/overview.md

# Audio/video transcription
gobbler audio ./meeting.mp3 --model small -o ./outputs/meeting.md

# Document conversion
gobbler document ./paper.pdf --no-ocr -o ./outputs/paper.md

# Webpage conversion
gobbler webpage "https://example.com" -o ./outputs/page.md

If working from a source checkout without a global install, use uv --directory:

uv --directory /path/to/gobbler run gobbler doctor --json
uv --directory /path/to/gobbler run gobbler youtube "URL" -o ./outputs/transcript.md

Output verification

After using -o, verify the file exists and has frontmatter plus a non-empty body:

test -s ./outputs/result.md
python - <<'PY'
from pathlib import Path
p = Path('./outputs/result.md')
text = p.read_text()
assert text.startswith('---'), 'missing YAML frontmatter'
assert len(text.split('---', 2)[-1].strip()) > 0, 'empty markdown body'
print('ok')
PY

Safety boundaries

• Ask before browser automation against authenticated/private pages.
• Never extract secrets, cookies, tokens, private keys, unrelated tabs, or unrelated personal content.
• Use gobbler-browser for detailed browser/session automation instructions; it has a different safety model from normal conversion.

Detailed references

Load only the relevant reference when needed:

• references/youtube.md — YouTube videos/playlists, captions, languages, timestamps, and deterministic frame refinement.
• references/audio.md — audio/video files, Whisper models, ffmpeg, timestamps.
• references/document.md — PDFs and office documents, OCR, Docling readiness.
• references/webpage.md — URLs/webpages, selector extraction, Crawl4AI readiness.
• references/batch.md — playlists, directories, URL lists, JSON progress, skip-existing behavior.