YouTube Digest Skill

Purpose: Deep video digestion for tactical learning + content repurposing
For: All agents (Lani, Meda, Persey)
Last updated: 2026-03-05

---

Trigger Phrases

• "digest this video"
• "save this video for content"
• "extract from this video"
• "learn from this"
• "what's useful in this video"
• Any YouTube URL paste

Output Standard (Michael)

When asked to digest a video, optimize for time saved:

• Return specifics only (tools, workflows, prompts, skill/plugin names, implementation steps).
• Explicitly map findings to relevant pillar(s) and interests.
• If video is mostly generic, say so fast and label it low value.

Default answer shape must be concise + expandable:

1. Fast verdict (3–6 bullets) — what matters / what doesn't
2. Specifics extracted — concrete tools, steps, frameworks, commands
3. Useful for us? — yes/no + exactly where it applies
4. Deep dive (optional) — only when there is genuinely high-leverage detail

---

Michael's Research Interests (Priority Order)

When digesting videos, prioritize content matching these interests:

1. AI Design for Marketing

What to extract:

• How to generate carousels with AI
• How to generate PDFs with AI
• AI-assisted design workflows
• Branded system creation with AI

Specifics needed:

• Exact prompts for design generation
• Tool names and when to use them
• Step-by-step: prompt → design output
• Design system keywords (tokens, components, etc.)

2. Agent Orchestration / Agent Decoding

What to extract:

• Workflow architectures (Step 1, Step 2, etc.)
• "If X then Y" tool mappings
• MCP usage patterns
• Skill plugin utilization
• Claude Code life hacks and advanced usage

Specifics needed:

• Specific skill names and what they do
• How to chain agents together
• Error handling patterns
• State management between agents

3. Prompt Engineering (All Types)

What to extract:

• UI design prompts
• Marketing content prompts
• Agent control prompts
• Framework names and structures

Specifics needed:

• Copy-pasteable templates with [BLANKS]
• Prompt construction workflows
• Specific keywords that change outputs

4. AI Creatorship (Images/Video)

What to extract:

• Tool-specific workflows
• Exact prompts with all parameters
• Advanced techniques (ControlNet, inpainting, etc.)

Specifics needed:

• "If I need [effect], use [tool]" mappings
• Complete prompt texts
• Settings and parameters

---

Digestion Workflow

Default Engine (MANDATORY)

Always use Supadata first whenever Michael asks to digest a video.

Rule:

1. Use Supadata transcript/metadata endpoints as the primary ingestion path.
2. Do not start with yt-dlp, YouTube web scraping, or random transcript sites.
3. Only fallback away from Supadata if API/auth fails — and clearly state fallback was used.

Supadata Under-the-Hood Reference (embedded)

Auth: x-api-key: $SUPADATA_API_KEY

Primary endpoints for this skill:

• Transcript (default): GET /v1/youtube/transcript?url=<URL>&text=true&lang=en
• Transcript (timestamped): GET /v1/youtube/transcript?url=<URL>&text=false&lang=en
• Video metadata: GET /v1/youtube/video?id=<VIDEO_ID>
• Cross-platform metadata: GET /v1/metadata?url=<URL>
• Search (when needed): GET /v1/youtube/search?query=<QUERY>&type=video&sortBy=views

Fallback rule inside this skill:

• If /youtube/video fails due to ID/param mismatch, derive VIDEO_ID from URL and retry with id.
• If transcript unavailable in lang=en, retry without lang to use first available language.

Step 1: Fetch Video Metadata + Description

curl -s "https://api.supadata.ai/v1/youtube/video?id=<VIDEO_ID>" \
  -H "x-api-key: $SUPADATA_API_KEY"

Check description for:

• Links to prompts, templates, resources
• "Link in description" mentions
• Free downloads, swipe files

If external links found:

• Fetch with web_fetch
• Extract prompts, frameworks, templates
• Save to resources/ folder immediately

Step 2: Fetch Transcript

curl -s "https://api.supadata.ai/v1/youtube/transcript?url=<URL>&text=true" \
  -H "x-api-key: $SUPADATA_API_KEY"

Step 3: Fast Verdict

No-fluff filter: if you cannot extract copy-pasteable tactics, clearly mark video as low tactical value.

Reply to Michael in this format:

**[VIDEO TITLE]** — [Channel] (~Xmin)

**What it is:** [One sentence - what they make/show + tools used]

**Verdict:** USEFUL / PARTIAL / SKIP

**Why:**
• [Specific reason - what's non-obvious]
• [What's genuinely new]
• [What's missing/re-chewed]

**Interest Match:** [Which of Michael's interests this serves]
**Novelty:** High / Medium / Low
**Repurposable:** Yes / No — [why]

**Specific findings preview:**
• [Tool name]: [what it does]
• [Prompt/Workflow]: [brief description]
• [Technique]: [brief description]

Digest and save? (yes / skip / partial)

Step 4: Deep Digestion (if approved)

Extract ONLY:

• Specific tool names and versions
• Exact prompts (copy-paste ready)
• Step-by-step workflows
• "If X then Y" mappings
• Code snippets
• Command examples

Format:

## [Video Title] — Deep Digest

### Tools Mentioned
| Tool | Use Case | When to Use |
|------|----------|-------------|
| [name] | [what it does] | [specific scenario] |

### Prompts / Templates

[Exact prompt text with [BLANKS]]

**Context:** [when to use this]
**Key ingredients:** [what makes it work]

### Workflows
**[Workflow name]:**
1. [Step 1 - specific action]
2. [Step 2 - specific action]
3. [Step 3 - specific action]

### "If X Then Y" Mappings
| If you need... | Then use... | Steps |
|----------------|-------------|-------|
| [specific task] | [specific tool] | [brief steps] |

### Code / Commands
```bash
[exact commands shown]

Key Insights

• [Specific insight 1]
• [Specific insight 2]

Gaps / What's Missing

• [What wasn't covered]
• [What doesn't work]

### Step 5: Save to Learning Repository

**Location:** `~/work/laniameda/laniameda-hq/knowledge-base/sources/youtube/YYYY-MM-DD-slug/`

**Files to create:**

1. **meta.json**
```json
{
  "title": "...",
  "url": "...",
  "channel": "...",
  "date_processed": "YYYY-MM-DD",
  "interests": ["ai-design", "agent-orchestration", "prompt-engineering"],
  "verdict": "useful|partial|skip",
  "tools_mentioned": ["tool1", "tool2"],
  "repurposable": true,
  "marketing_angle": "brief description",
  "tags": []
}

2. digest.md — Full deep digestion
3. prompts.md — Extracted prompts/templates only
4. workflows.md — Step-by-step workflows only
5. tools.md — Tool mappings and comparisons

Step 6: Flag for Repurposing

If content is marketing-ready:

Create: ~/work/laniameda/laniameda-hq/knowledge-base/marketing-ready/YYYY-MM-DD-slug.md

# Marketing Angle: [Video Title]

**Source:** [URL]
**Original Pillar:** [Michael's content pillar]

## Our Angle
[How to reframe this for laniameda's voice/positioning]

## Key Takeaways to Share
- [Point 1]
- [Point 2]

## Content Formats
- [ ] X thread
- [ ] LinkedIn post
- [ ] Carousel
- [ ] PDF guide

## Original vs Our Version
| Element | Original | Our Version |
|---------|----------|-------------|
| [aspect] | [their approach] | [our approach] |

Step 7: Update This Skill

If new tools/workflows discovered:

1. Add tool to "Tools Reference" section below
2. Add workflow pattern to "Common Patterns"
3. Update "Michael's Research Interests" if new category discovered
4. Git commit with message: "skill: add [tool/workflow] from [video]"

---

Tools Reference (Update as Discovered)

For Design/Marketing

Tool	Purpose	Best For
v0.dev	UI generation	React components, landing pages
Tempo	AI design	Marketing materials, brand assets
Lovable	App prototyping	Quick mockups, demos
Claude Artifacts	Component building	Interactive UI elements
Canva AI	Design automation	Social posts, presentations

For Agent Orchestration

Tool	Purpose	Best For
OpenClaw	Agent framework	Multi-agent workflows
Claude Code	Coding agent	Frontend, full-stack
n8n	Workflow automation	Connecting tools
Make.com	Automation	Visual workflow building

For MCPs (Model Context Protocols)

MCP	Purpose	Best For
Browser Use	Web automation	Scraping, form filling
GitHub	Code management	Repo operations
Supadata	Video/data extraction	YouTube, social content
Sequential Thinking	Reasoning	Complex problem solving

For Content/Research

Tool	Purpose	Best For
x-tweet-fetcher	X post extraction	Social research
instagram-extract	IG content extraction	Visual research
supadata	Video transcripts	YouTube digestion
web_fetch	Page extraction	Article research

---

Common Patterns (Update as Discovered)

Pattern: AI Design Workflow

1. Start with prompt in [tool]
2. Iterate on output (2-3 rounds)
3. Export to [format]
4. Refine in [secondary tool]

Pattern: Agent Chaining

1. Agent 1: [task] → outputs [format]
2. Agent 2: takes [format] → does [task]
3. Agent 3: reviews → final output

Pattern: Content Repurposing

1. Extract from video → digest.md
2. Identify angle → marketing-ready/
3. Reframe for brand → content draft
4. Create asset → carousel/PDF/post

---

Repurposing Rules

1. Extract thesis, not creator's products

   • Don't name-drop their tools
   • Use their insight, not their branding

2. Reframe for laniameda voice

   • Art + engineering + systems
   • Taste-driven, not volume-driven
   • Specific, not generic

3. Add our proof points

   • Use our agent examples
   • Reference our workflows
   • Show our results

4. Never copy-paste

   • Digest → understand → re-express
   • Our angle, their insight

---

Checklist Before Delivering

• Specific tool names mentioned (not "AI tools")?
• Exact prompts extracted with [BLANKS]?
• Step-by-step workflows documented?
• "If X then Y" mappings created?
• Saved to KB with proper structure?
• Flagged for repurposing if marketing-ready?
• Skill updated with new tools/patterns?

---

Readaptivity

Log feedback to: ~/work/laniameda/laniameda-hq/knowledge-base/profile/feedback-log.md

Format:

YYYY-MM-DD | [video slug] | [what was wrong] | [how to fix]

Update this skill after 3+ similar feedback patterns.

---

This skill ensures: Every video becomes reusable IP for product and marketing.