Analyze Kids Content

Sub-skill: analyze-kids-content

Part of /youtube-kids-audit. Reads a scraped watch history, filters adult noise, and generates a personalized educational curriculum analysis for a specific child.

Model requirement: Use claude-opus-4-6 (Pro/Opus). This analysis requires expert-level educational judgment — do NOT use Haiku or any flash model.

Usage

/youtube-kids-audit analyze-kids-content [--profile <child-name>] [--history <path>]

Step 0: Load or create kid profile

Check for existing profiles:

ls user_docs/kid_profile/*.md 2>/dev/null | grep -v '_template\|README'

If none exist, run the intake interview — ask these one at a time:

1. "What is your child's name and age?"
2. "Is this a dedicated YouTube account, or shared with an adult?"
   • If shared: "Describe the adult content I should ignore (e.g. finance podcasts, political commentary, foreign-language drama)"
3. "What's their current math level?" (e.g. grade, curriculum name, specific topics)
4. "What's their reading level?" (e.g. reads independently vs with help, example books)
5. "What are their top 3 interests?"
6. "What do they dislike or find boring?"
7. "What do they currently watch or listen to most?"
8. "Any context I should know — learning style, emotional sensitivities, goals?"

Write answers to user_docs/kid_profile/<name>.md using _template.md schema.

Step 1: Load watch history

import glob, json

history_files = sorted(glob.glob("user_docs/watch_history_*.json"), reverse=True)
if not history_files:
    raise FileNotFoundError(
        "No watch_history_*.json found. Run '/youtube-kids-audit' first to scrape."
    )
videos = json.load(open(history_files[0], encoding="utf-8"))
print(f"Loaded {len(videos)} videos from {history_files[0]}")

Step 2: Format history as plain-text table (no URLs)

def format_history_table(videos: list[dict]) -> str:
    lines = ["| # | Title | Channel | Duration |",
             "|---|-------|---------|----------|"]
    for i, v in enumerate(videos, 1):
        lines.append(
            f"| {i} | {v.get('title','')[:70]} | {v.get('channel','')[:40]} | {v.get('durationRaw','')} |"
        )
    return "\n".join(lines)

Do not include URLs or thumbnails — the model doesn't need them and it reduces token cost.

Step 3: Build the analysis prompt

Construct from the kid's profile. The prompt follows the structure of an Educational Consultant analyzing a personalized digital learning curriculum.

def build_prompt(profile: dict, history_table: str) -> str:
    name          = profile["name"]
    age           = profile["age"]
    math_level    = profile.get("math_level", "grade-appropriate")
    reading_level = profile.get("reading_level", "")
    interests     = "\n".join(f"- {i}" for i in profile.get("interests", []))
    dislikes      = "\n".join(f"- {d}" for d in profile.get("dislikes", []))
    ignore        = "\n".join(f"- {p}" for p in profile.get("ignore_patterns", []))
    current_media = profile.get("current_books_shows", "")
    notes         = profile.get("parent_notes", "")
    shared        = bool(profile.get("ignore_patterns"))

    return f"""
You are an expert Educational Consultant specializing in Gifted & Talented Elementary
Education and personalized digital learning curricula.

## Child Profile

- Name: {name}, Age: {age}
- Math level: {math_level}
- Reading level: {reading_level}
- Interests (HIGH value): {interests}
- Dislikes / avoid: {dislikes}
- Currently consuming: {current_media}
- Parent notes: {notes}

## Five Core Educational Goals

When classifying and recommending content, evaluate it against these five goals:

1. **Expand Knowledge Breadth & Depth** — Move beyond surface facts into history,
   biography, global concepts, and cultural literacy.
2. **Improve English Holistic Skills** — Advance listening comprehension, complex
   narrative exposure, writing structure, and articulate speaking. Not phonics drills.
3. **Elevate Math Content** — CRITICAL: DO NOT recommend basic counting, arithmetic
   drills, or phonics-adjacent content. Content must match {name}'s actual level
   ({math_level}), focusing on logic, puzzles, and conceptual thinking.
4. **Broaden Science Horizons** — Spark curiosity across biology, physics, astronomy,
   engineering. Prefer inquiry-based ("How? Why?") over fact-listing.
5. **Reinforce Social-Emotional Learning (SEL)** — Support understanding of emotions,
   resilience, and growth mindset (especially important for advanced learners who
   rarely face challenge and may struggle when they finally do).

## Account Context

{"This account is SHARED with an adult. You MUST IGNORE content matching these patterns:" if shared else "This is a dedicated child account."}
{ignore}

Focus your analysis on: children's educational content, audiobooks, cartoons,
science/math enrichment, English language content, and fantasy/adventure stories.

## Watch History

{history_table}

---

## Your Task

Analyze the watch history above and produce a Strategic Educational Report in this
exact structure:

---

### Part 1: Content Audit — What Is {name} Currently Exploring?

Identify the dominant themes and "systems" {name} is consuming.
For each theme, list 2-3 example titles and rate engagement (High/Medium/Low based on
frequency, repeat views, or channel volume).

**Level check:** Flag any content that appears to be *below* {name}'s level — content
she has likely "graduated from" even though it may still appear in history. Examples:
basic phonics channels for a fluent reader, simple counting for an advanced math learner.
Label these **"Educational Junk Food"** — they feel productive but provide no new
challenge.

---

### Part 2: Three-Way Content Classification

Classify ALL relevant (non-adult) videos into one of these buckets:

**✅ KEEP — The Foundation (already working)**
Content aligned with the 5 goals that should remain core to her curriculum.
For each item: channel name, why it fits, which goal(s) it serves.

**🛑 REMOVE / GRADUATE FROM — Outdated or Misaligned**
Content to deprecate. Two types:
- *Too Easy / Mastered*: she has outgrown this (flag with "Educational Junk Food")
- *Misaligned*: passive entertainment with low vocabulary/complexity for her level

**🟢 ADD — The Growth Areas (gaps to fill)**
This is the most important section. Recommend specific named YouTube channels or
content types to fill gaps identified in the audit. Use these strategies:

- **Elevate Math**: Logic puzzles, proof-based thinking, conceptual geometry —
  no drills. Must match her actual level.
- **The "Trojan Horse"**: History / biography content framed as engineering,
  science, or systems — bypasses any dislike of "dates and names" history.
- **Deepen English**: Fiction featuring competence, survival, and logic.
  Avoid whimsical/nonsense if she dislikes it. Audiobooks at native speed
  beat ESL-simplified videos for vocabulary growth.
- **Science Breadth**: Inquiry-first channels (How? Why? What if?) over
  fact-listing channels.
- **SEL for Gifted Learners**: Growth mindset content specifically addressing
  the "panic when things get hard" pattern common in advanced learners.

---

### Part 3: Curriculum Channel Table

Produce a Markdown table of the recommended final channel lineup:

| Channel / Playlist | Short Description | Why {name} Needs This | Goal Alignment |
|-------------------|-------------------|-----------------------|----------------|
| (keep + add items only, sorted by goal) | | | |

Use these Goal labels: `Knowledge Breadth` | `English Skills` | `Math` | `Science` | `SEL`

---

### Part 4: Implementation Schedule

Suggest a simple weekly rhythm (not a rigid schedule — just framing):

| Time / Context | Content Type | Example |
|----------------|-------------|---------|
| Morning (fresh brain) | Active learning | Math puzzles, writing structure |
| Relaxed / meal time | High-quality entertainment | Engineering shows, narrative history |
| Car / bedtime | Audio-first | Audiobooks, mythology podcasts |

---

### Part 5: 3-5 Immediate Next Steps

A specific, actionable parent checklist:

- [ ] **Subscribe to:** [channel] — [one-line reason]
- [ ] **Remove or stop:** [channel] — [one-line reason]
- [ ] **Buy / borrow:** [book or audiobook] — start with [specific entry point]
- [ ] **Ask {name}:** "[A question that opens a conversation about a gap topic]"
""".strip()

Step 4: Call Claude Opus for analysis

import anthropic

def run_analysis(prompt: str) -> str:
    client = anthropic.Anthropic()
    msg = client.messages.create(
        model="claude-opus-4-6",   # Pro model required — not haiku/flash
        max_tokens=4096,
        messages=[{"role": "user", "content": prompt}]
    )
    return msg.content[0].text

Why Opus and not Haiku: This task requires the model to:

• Detect level mismatches between a child's academic profile and what they're watching
• Identify "educational junk food" vs genuinely enriching content
• Generate curriculum-quality recommendations tailored to a specific learner profile
• Cross-reference five goal dimensions simultaneously

Haiku will produce generic results. Opus produces recommendations that feel like a real educational consultant reviewed the data.

Step 5: Save and display report

import datetime, os

today = datetime.date.today().isoformat()
safe_name = profile["name"].lower().replace(" ", "_")
out_path = f"user_docs/kid_profile/{safe_name}_analysis_{today}.md"

header = f"""# YouTube Watch History Analysis — {profile['name']}

**Date:** {today}
**Videos analyzed:** {len(videos)}
**Model:** claude-opus-4-6

---

"""

with open(out_path, "w", encoding="utf-8") as f:
    f.write(header + analysis_text)

print(f"Report saved → {out_path}")
print("\n" + analysis_text[:3000])
if len(analysis_text) > 3000:
    print(f"\n... [full report in {out_path}]")

Key concepts baked into the prompt

"Educational Junk Food"

Content that feels productive (it's about numbers! it's educational!) but provides no new challenge because the child has already mastered it. Common examples:

• Basic phonics channels for a child who reads chapter books independently
• Simple counting/number cartoons for a child doing advanced math curriculum
• Letter-recognition apps for a fluent reader

These are more harmful than pure entertainment because they consume curriculum time without developmental return, and may actually bore advanced learners into disliking "educational" content.

"Consumer → Creator" gap

Advanced readers often have a large intake-output gap: they consume complex content but haven't been guided to produce at a matching level. The analysis flags this and recommends writing-structure content (not just "write a story") to close it.

The "Trojan Horse" recommendation

For children who dislike a specific subject (e.g. traditional history), recommend content in their preferred domain that secretly teaches the gap subject:

• History of medicine → satisfies anatomy interest + teaches history
• Engineering history (bridges, rockets) → satisfies STEM interest + teaches history
• Biography of a scientist → satisfies science interest + teaches biography genre

Level-matching (critical for gifted learners)

The prompt explicitly tells the model the child's actual academic level and instructs it NOT to recommend content below that level. A child doing 3rd-grade-level math should never see a recommendation for basic counting videos, regardless of how many counting videos appear in the history.

Output files

File map

skills/media/youtube-kids-audit/
├── SKILL.md                              ← Step 1: scrape watch history via CDP
└── analyze-kids-content/
    └── SKILL.md                          ← Step 2: this file — analyze + report

user_docs/
├── watch_history_<date>.json             ← scraped history (gitignored)
└── kid_profile/
    ├── README.md                         ← index
    ├── _template.md                      ← blank profile schema
    ├── <child-name>.md                   ← filled-in kid profile
    └── <child-name>_analysis_<date>.md   ← generated report (gitignored)