YouTube to Markdown

Multiple videos: Process one video at a time, sequentially. Do not run parallel extractions. Do not create your own scripts. Channel subfolder naming: Channel Name (channel_id) — always include the channel ID.

Step -1: Detect input type

If input is a channel URL (contains /@, /channel/, /c/, /user/) or a bare channel ID (starts with UC, 24 chars), but NOT watch?v=: Read and follow ./subskills/channel_browse.md.

Otherwise: Continue to Step 0.

Step 0: Check if extracted before

python3 ./scripts/run.py check "<YOUTUBE_URL>" "<output_directory>"

Output JSON contains video_id. Set BASE_NAME = youtube_{video_id} for all subsequent steps.

Tell the user: "Valitse 'Yes, allow all edits during this session'. Kirjoitan warmup-tiedoston Write-oikeuksien varmistamiseksi, jotta agenttien kirjoitusoikeudet toimivat ilman lisäkyselyjä."

Then Write "# Processing {video_id}\n" to <output_directory>/${BASE_NAME}_warmup.tmp. If Write fails because file exists and hasn't been read, Read it first then Write.

If exists: false: Continue to Step 1.

If exists: true: Read and follow ./subskills/update_flow.md.

Step 1: Choose output

AskUserQuestion:

• question: "What do you want to extract from the video?"
• header: "Output"
• multiSelect: false
• options: A. "Summary + Comments (Recommended)" - Summary cross-analyzed with comments. Timestamped transcript stored for reference. B. "Summary + Comments + Formatted Transcript with Watch Guide" - Option A + cleaned and formatted full transcript with synthesized watch guide → double tokens C. "Summary Only" - Summary of video content. Timestamped transcript stored for reference. D. "Formatted Transcript Only" - Cleaned and formatted full transcript

Step 2: Execute modules

Based on user's choice, read and follow each subskill instruction in ./subskills/{file}.

Concurrency: "|" = run concurrently as parallel foreground agents in one message. Never use run_in_background.

Delegation: Subagents cannot launch sub-subagents. Modules that need parallel Agent calls (transcript_polish.md) must be read and executed by the orchestrator directly — do NOT delegate them to an Agent. Only leaf modules (no internal Agent calls) can be delegated.

• A: transcript_extract.md → (transcript_summarize.md | comment_extract.md) → comment_summarize.md
• B: transcript_extract.md → (transcript_summarize.md | transcript_polish.md | comment_extract.md) → comment_summarize.md
• C: transcript_extract.md → transcript_summarize.md
• D: transcript_extract.md → transcript_polish.md

Bold = orchestrator reads and executes directly (not delegated to Agent).

For option B only, create marker file before running modules:

python3 ./scripts/run.py flag "<output_directory>/${BASE_NAME}_watch_guide_requested.flag"

Step 3: Finalize

python3 ./scripts/run.py assemble [flag] "${BASE_NAME}" "<output_directory>"

Flags: A=--summary-comments, B=(none), C=--summary-only, D=--transcript-only

Use --debug to keep intermediate files.