talk-distill

name	talk-distill
description	Meta-orchestrator for distilling a YouTube talk into study material. Use when the user gives a YouTube URL and wants any of — SRT captions, a clean speaker-labelled transcript, an executive summary, or a timestamp-indexed key-takeaways doc. Routes to talk-fetch / talk-transcribe / talk-summarize / talk-takeaways sub-skills based on what's needed.
version	0.1.1
last_reviewed	2026-05-26T00:00:00.000Z

Talk Distill — Meta Skill

Coordinated set of skills for turning a YouTube talk into study artifacts: raw SRT → light-edit speaker-labelled transcript → executive_summary.md (TL;DR + Salient Points + Outline) + key_takeaways.md (timestamped emoji-bullet index).

The 4 sub-skills

Skill	One-liner
talk-fetch	Probe + download the best subtitle track for a YouTube URL. Auto-skips human-authored vs. auto-captioned vs. translated decisions. Outputs captions.SRT.
talk-transcribe	Dedupe YouTube's rolling auto-captions, toggle speakers on >>, emit a paragraph-flowed transcript.txt. Wraps the project's scripts/srt_to_transcript.py.
talk-summarize	Read the transcript, produce executive_summary.md (TL;DR + Salient Points + Outline). First-read entry point.
talk-takeaways	Read the transcript + SRT, produce key_takeaways.md (timestamped emoji-bullet index). Re-navigation entry point.

Typical sequence

For a brand-new talk:

1. talk-fetch — pull the SRT into <topic>_talk/captions.SRT.
2. talk-transcribe — produce <topic>_talk/transcript.txt.
3. talk-summarize — produce <topic>_talk/executive_summary.md.
4. talk-takeaways — produce <topic>_talk/key_takeaways.md.

Each step is independently invokable. If a user already has the SRT and just wants takeaways, skip straight to talk-takeaways.

When to invoke which sub-skill

Situation	Load
User pastes a YouTube URL and says "distill this" / "summarize this" / "get the transcript"	talk-distill (this meta), then route by what's missing
User has an SRT already and wants a transcript	talk-transcribe
User has a transcript and wants the summary	talk-summarize
User wants a jumpable index, not a read-through summary	talk-takeaways
Just the captions, nothing more	talk-fetch

How Claude should use this meta-skill

When a YouTube URL appears in a request along with phrasing like "summarize," "distill," "get the transcript," "make study notes," etc., mentally invoke this meta-skill, check what's already on disk in the target directory, and run only the missing steps. Surface the chosen plan before acting:

• "I see no captions yet — fetching first, then transcribing."
• "Transcript exists; I'll just produce the summary."
• "Both summary and takeaways requested — I'll write them in parallel after the transcript."

If the talk's subdir doesn't exist yet, propose a name based on the video title and uploader, scoped descriptively (e.g. fulcro_statecharts_talk/, not video_1/).

Conventions

• One subdir per talk under the project root; filenames in that subdir are fixed: captions.SRT, transcript.txt, executive_summary.md, key_takeaways.md.
• The executive summary and key takeaways coexist — they serve different purposes (first-read vs. re-navigation). Default is to produce both unless the user asks for only one.
• Speaker labels in the transcript are ~90–95% accurate. The pipeline is deliberately a rough first cut; manual spot-correction is expected for high-stakes citation use.

Source repo

This skill is git-source-controlled in avidrucker/talk-distill-skills. SKILL.md files in ~/.claude/skills/talk-* are symlinks to that repo — edit in the repo, changes are live next session. Each skill has independent semver (VERSION file + CHANGELOG.md).

Deep reference

The design rationale, heuristic details, and cross-video validation results live in workflow_notes.md at the repo root. That doc is the "research/" equivalent for this skill family.

Sister repos this family borrowed conventions from (not "styles" — just the source of the outline shape and file-naming conventions we adopted):

• warmed-skills — first repo I built with the <talk>/captions.SRT + transcript.txt + talk_summary.md shape that talk-distill produces. The executive_summary.md format (TL;DR + Salient Points + Outline) came from there.
• yegor-pm-skills — first repo I built with the meta-skill + per-sub-skill SKILL.md + VERSION + CHANGELOG.md layout that this family follows. The key_takeaways.md format (timestamped emoji-bullet index) came from there.