Conference Transcribe

What it does

When your agent uses it

• User says transcribe this conference
• Split this livestream into talks
• Transcribe each talk separately
• Provides a YouTube URL of a multi-hour event stream with chapter timestamps

Example prompts

• “transcribe this conference”
• “split this livestream into talks”
• “transcribe each talk separately”
• “/conference-transcribe”

Requirements

• Python 3
• A credential in GROQ_API_KEY
• A credential in ANTHROPIC_API_KEY
• Compatibility (from SKILL.md): Requires macOS with ffmpeg and yt-dlp installed. Needs at least one transcription backend (Groq API recommended for speed). Needs an LLM API key (Anthropic recommended) for cleanup pass.

Workflow steps

8 steps, taken from the step headings in SKILL.md.

1. 1Check Prerequisites
2. 2Get Video Metadata and Captions
3. 3Parse Chapter Timestamps into Talk Manifest
4. 4Build Raw Transcripts (Caption Path)
5. 5(alt): Build Raw Transcripts (Whisper/API Path)
6. 6Download Video Clips (Optional, Parallel)
7. 7LLM Cleanup Pass
8. 8Write Final Output

What it can do on your machine

Read from SKILL.md and the folder at commit 038ef34. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Shell commands in SKILL.md call:

  • yt-dlp
  • ffmpeg
  • python3
  • uv
  • curl
  • pip

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Hosts in commands or code, which the agent is likely to contact:

  • youtube.com
  • api.groq.com

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names these keys or tokens, usually read from environment variables:

  • HF_TOKEN
  • GROQ_API_KEY
  • ANTHROPIC_API_KEY

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

• Compatibility

  Requires macOS with ffmpeg and yt-dlp installed. Needs at least one transcription backend (Groq API recommended for speed). Needs an LLM API key (Anthropic recommended) for cleanup pass.

  From compatibility in the SKILL.md frontmatter.

Context cost

Conference Transcribe loads about 2.4k tokens when it runs. Until then it costs about 134 tokens; SKILL.md has 676 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~134

When it runs · the whole SKILL.md, loaded when a task matches

~2.4k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from swyxio/skills at commit 038ef34, republished under its MIT licence (© swyxio). 676 words, ~2,423 tokens.

name

conference-transcribe

description

Transcribe a multi-talk conference livestream or long YouTube video into separate per-talk transcripts. Parses timestamps from the video description to split talks, downloads audio/video, transcribes each segment, then uses an LLM to clean up and format the transcripts with key takeaways and frequent timestamps. Use when user says "transcribe this conference", "split this livestream into talks", "transcribe each talk separately", or provides a YouTube URL of a multi-hour event stream with chapter timestamps.

compatibility

Requires macOS with ffmpeg and yt-dlp installed. Needs at least one transcription backend (Groq API recommended for speed). Needs an LLM API key (Anthropic recommended) for cleanup pass.

license

MIT

metadata.author

swyxio

metadata.version

1.0

metadata.last-updated

2026-04-10

metadata.primary-tools

yt-dlp, ffmpeg, Groq Whisper API, Claude API

Has YouTube captions? (check with yt-dlp --list-subs)
  YES -> Use YouTube captions (fastest, free, no setup)
  NO  -> Has Groq/OpenAI API key?
    YES -> Use Groq API (whisper-large-v3-turbo, nearly free, very fast)
    NO  -> Has local Whisper installed?
      YES -> Use faster-whisper or mlx-whisper
      NO  -> Install via uv: `uv venv .venv && source .venv/bin/activate && uv pip install faster-whisper`

which yt-dlp || echo "MISSING: brew install yt-dlp"
which ffmpeg || echo "MISSING: brew install ffmpeg"
which jq || echo "MISSING: brew install jq"

# Check for API keys (optional but recommended)
[ -n "$GROQ_API_KEY" ] && echo "Groq: ready" || echo "Groq: not set (needed for Whisper API)"
[ -n "$ANTHROPIC_API_KEY" ] && echo "Anthropic: ready" || echo "Anthropic: not set (needed for cleanup)"

VIDEO_URL="$1"  # YouTube URL from user

# Download metadata + captions (no video)
yt-dlp --write-info-json --skip-download \
  --write-sub --write-auto-sub --sub-lang en \
  -o "media/%(id)s" "$VIDEO_URL"

[
  {
    "index": 1,
    "title": "Speaker Name: Talk Title",
    "speaker": "Speaker Name",
    "slug": "01-speaker-name",
    "source_chapter_start": "00:24:25",
    "source_chapter_end": "00:42:39",
    "start_seconds": 1465,
    "end_seconds": 2559,
    "duration_seconds": 1094
  }
]

# Speaker Name: Talk Title

- Source: https://youtube.com/watch?v=ID&t=1465s
- Source range: 00:24:25 - 00:42:39
- Duration: 00:18:14
- Transcript source: YouTube auto-captions

## Timestamped Transcript

[00:24:26 | +00:00:01] Good morning everyone...

[00:24:54 | +00:00:29] Next paragraph of text...

yt-dlp -f 91 \
  --downloader ffmpeg \
  --downloader-args "ffmpeg_i:-allowed_extensions ALL" \
  --download-sections "*${START}-${END}" \
  -o "clips/${SLUG}.%(ext)s" \
  "$VIDEO_URL"

You are an expert transcript editor. Take this raw auto-caption transcript and produce a clean, readable document.

Rules:
1. KEEP all timestamps in [HH:MM:SS | +MM:SS] format. Include them every 30-60 seconds.
2. Fix transcription errors: proper nouns, technical terms, company names, jargon.
3. Add paragraph breaks at natural topic transitions.
4. Remove filler words (um, uh, like, you know) unless they add meaning.
5. Preserve the speaker's voice -- clean up, don't rewrite.
6. For multi-speaker segments, use [Speaker Name]: format.

Output format:
# Talk Title
**Speaker** -- Role/Company
**Event**: {event name and date}

## Key Points
- 4-8 bullet points of key takeaways

## Timestamped Reading Transcript
[Content with timestamps, paragraphs, and light line-wrapping for readability]

transcripts/
  raw/       -- unedited VTT-parsed or Whisper output
  cleaned/   -- LLM-cleaned markdown
clips/       -- individual talk MP4s (optional)
talks.json   -- manifest with metadata
reports/
  talk-manifest.md  -- summary table

# 1. Grab metadata + captions
yt-dlp --write-info-json --skip-download --write-auto-sub --sub-lang en -o "media/%(id)s" "$URL"

# 2. Build talks.json from chapters in info.json
python3 scripts/build_transcripts.py

# 3. Download individual clips (parallel, optional)
python3 scripts/download_clips.py

# 4. Clean up transcripts with LLM
python3 scripts/cleanup_transcripts.py

© swyxio, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in conference-transcribe of swyxio/skills.

Open the folder on GitHubat commit 038ef34