# Youtube Transcript

> |

- **Type:** Skill
- **Install:** `agentstack add skill-elliotjlt-claude-skill-potions-youtube-transcript`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [ElliotJLT](https://agentstack.voostack.com/s/elliotjlt)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [ElliotJLT](https://github.com/ElliotJLT)
- **Source:** https://github.com/ElliotJLT/Claude-Skill-Potions/tree/main/skills/youtube-transcript

## Install

```sh
agentstack add skill-elliotjlt-claude-skill-potions-youtube-transcript
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# YouTube Transcript

Extract transcripts from YouTube videos and save them as structured, searchable
markdown files. Used for research, interview prep, company analysis, and building
reference material from video content.

## When To Activate

- User shares a YouTube URL (youtube.com/watch, youtu.be, youtube.com/embed)
- User says "transcript", "youtube transcript", or "pull transcript"
- User asks to research a video or extract content from YouTube

## Instructions

### Step 1: Check yt-dlp

yt-dlp must exist at `/tmp/yt-dlp` or on PATH.

If missing, install it:
```bash
curl -L https://github.com/yt-dlp/yt-dlp/releases/latest/download/yt-dlp_macos -o /tmp/yt-dlp && chmod +x /tmp/yt-dlp
```

### Step 2: Extract Transcript

Run the extraction script:
```bash
node /scripts/fetch-transcript.mjs ""
```

The script outputs JSON with:
- `title`, `channel`, `url`, `duration`, `uploadDate`, `description`
- `segments` — individual timestamped lines
- `paragraphs` — segments merged into ~30-second blocks for readability

### Step 3: Save Output

Create a folder and markdown file:

**Folder:** `transcripts/` in the current working directory (create if needed)
**Filename:** `YYYY-MM-DD-kebab-case-title.md`

```markdown
---
type: transcript
source: youtube
video_id: 
channel: 
duration: 
date_published: 
date_extracted: 
url: 
---

# 

**Channel:** 
**Duration:** 
**Published:** 
**URL:** 

---

## Summary

---

## Key Takeaways

### 1. 
- Bullet points summarising the core argument
- Include specific claims, frameworks, or advice given
- Attribute to the speaker where relevant

### 2. 
...repeat for each major theme (typically 3-7 themes for a 30-60 min video)

---

## Full Transcript

[0:00] . First paragraph of merged text...

[4:30] . Second paragraph continues...
```

The transcript section should be the full paragraph-merged output from the script,
with each timestamp block prefixed by a short topic label (2-5 words) for scannability.

### Step 4: Summarise

IMPORTANT: Do NOT just dump raw transcript. The primary value is the AI-generated
summary and key takeaways. Read the full transcript and produce:

1. **Summary** — 2-3 sentences on what the video covers
2. **Key Takeaways** — grouped by theme, with specific claims and quotes
3. **Full Transcript** — paragraph-merged with topic labels (kept as reference/source)

The summary and takeaways should be written in clean, scannable prose. Extract the
actual insights — not just "they discussed X" but what they specifically said about X.
If the video is relevant to a specific context (e.g., interview prep), add a relevance
table mapping insights to actions.

## Integration with Boulot Vault

When used inside the Boulot vault for job research:
- Save to `{user}/active/{company}/research/transcripts/` if the video relates to an active application
- Add a link to the transcript in the company's `research.md`
- Tag with relevant company/role context in frontmatter

## NEVER

- Download the video itself — subtitles only
- Fabricate or hallucinate transcript content
- Skip saving to file — always persist the transcript
- Run without checking yt-dlp exists first

## ALWAYS

- Use kebab-case lowercase filenames
- Include full YAML frontmatter
- Use the paragraph-merged format for readability (not raw segments)
- Include timestamps at paragraph boundaries
- Provide a content summary after extraction

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [ElliotJLT](https://github.com/ElliotJLT)
- **Source:** [ElliotJLT/Claude-Skill-Potions](https://github.com/ElliotJLT/Claude-Skill-Potions)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** yes
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-elliotjlt-claude-skill-potions-youtube-transcript
- Seller: https://agentstack.voostack.com/s/elliotjlt
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.