curriculum-designer

What this skill does

# Curriculum Designer

Design customized curricula for Apni Pathshala PODs with **real YouTube video links**.

**FEATURES:**
- ✅ Staged execution with checkpointing (recovery from failures)
- ✅ YouTube link verification with **fallback logic** (no blank URLs)
- ✅ Context capping per lesson (reduced token usage)
- ✅ Every topic gets a valid video OR search query fallback

---

## ⚡ Quick Start

### How This Skill Works

When invoked, the agent follows a **5-stage workflow** with checkpointing:

| Stage | What Happens | Checkpoint File |
|--------|--------------|-----------------|
| 1 | Gather requirements | `requirements.json` |
| 2 | Research YouTube videos | `research-results.json` |
| 3 | Verify videos + **fallback logic** | `validated-resources.json` |
| 4 | Design curriculum (one lesson at a time) | `curriculum-structure.json` |
| 5 | Create Google Sheet | `final-sheet-url.txt` |

### Checkpoint Behavior
- Each stage saves its output to a checkpoint file
- If checkpoint exists, stage loads it and **skips processing**
- If checkpoint doesn't exist, stage runs from scratch
- Re-running resumes from first **incomplete stage**

---

## Trigger

User message contains:
- **"Design curriculum"** → Start curriculum creation
- **"Create curriculum for [POD name]"** → Start with POD context
- **"Build learning plan"** → Start curriculum creation
- **"Curriculum for [subject/topic]"** → Start with topic context

---

## Target User

This skill is designed for **Madhur** (Academic Associate) who designs curricula for PODs.

---

## Configuration

- **API Keys:** Stored locally in `~/.openclaw/workspace/skills/curriculum-designer/.env` (NOT in git)
- **Output Folder:** `1upJQu-IVmZRJQsNGmJNRzq9IwL67MVL9` (Curriculum Designer)
- **Checkpoint Directory:** `~/.openclaw/workspace/curriculum-designer-checkpoints/`

**YouTube API Key:**
```
YOUTUBE_API_KEY=your_key_here
```

Get from: https://console.cloud.google.com/apis/credentials

---

## Agent Workflow Instructions
## Model Allocation for Stages

**Action:** Use different LLM models for different stages to optimize cost and performance.

### Stage-Specific Model Assignment

| Stage | Recommended Model | Reason |
|--------|------------------|--------|
| Stage 1: Requirements Collection | glm-4.7 | Quick reasoning, sufficient for form filling |
| Stage 2: YouTube Research | **glm-5** | Fast research, needs latest web knowledge |
| Stage 3: Video Validation | glm-4.7 | Pattern matching, simple logic |
| Stage 4: Curriculum Design | **glm-4.7** | Structured generation, cost-effective for lessons |
| Stage 5: Sheet Creation | glm-4.7 | JSON formatting, simple transformations |

### How to Set Models

**Option 1: Specify model when calling agent**
```bash
# Use glm-5 for research stage
agent.chat --model glm-5 --message "Research YouTube videos for..."

# Use glm-4.7 for design stage
agent.chat --model glm-4.7 --message "Generate lesson structure..."
```

**Option 2: Configure in SKILL.md**
Each stage should include model recommendation in its instructions:

```
### Stage 2: Research YouTube Resources

**Action:** Search YouTube for videos based on requirements

**Recommended Model:** glm-5 (fast research, latest web knowledge)

**Why:** Research needs up-to-date information and fast processing.
```

---

## Agent Workflow Instructions

### Stage 1: Gather Requirements

**Action:** Ask the user these questions (from SOP):

#### Basic Information
1. **POD Name** - Which POD is this curriculum for?
2. **Target Audience** - Grade level or age group of students?
3. **Subject Areas** - What subjects/topics should be covered?
4. **Duration** - How long is the program? (e.g., 1 month, 3 months, 6 months)
5. **Frequency** - How many classes per week?
6. **Daily Lab Hours** - How many hours will the lab operate?
7. **Previous Exposure** - Have students done digital learning before?

#### Teacher Context
8. **Teacher Capability** - Can teachers operate computers independently?
9. **Teacher Training Needed** - Do teachers need any training?

#### Learning Outcomes
10. **Learning Area Focus** - Which area(s) to prioritize?
    - Digital Literacy
    - Academic Empowerment
    - Skill Development
    - Employment Readiness
11. **Specific Skills** - What specific skills should students acquire?
12. **Assessment Method** - How will learning be measured?

**Output:** Save to checkpoint as JSON:
```json
{
  "pod_name": "Example POD",
  "target_audience": "Grade 8-10",
  "subject_areas": ["Digital Literacy", "Computer Basics"],
  "duration": "1 month",
  "frequency": "3 days/week",
  "daily_lab_hours": 2,
  "previous_exposure": "None",
  "teacher_capability": "Basic",
  "teacher_training_needed": true,
  "learning_area_focus": ["Digital Literacy"],
  "specific_skills": ["Basic computer operations", "Internet safety"],
  "assessment_method": "Practical exercises and quizzes"
}
```

**Checkpoint:** `~/.openclaw/workspace/curriculum-designer-checkpoints/<timestamp>-<session-id>/requirements.json`

---

### Stage 2: Research YouTube Resources

**Action:** Search YouTube for videos based on requirements

**API:** Use YouTube Data API v3 with key from `.env`

**Search Queries (Default):**
```python
search_queries = [
    "computer basics tutorial hindi beginners",
    "typing practice hindi tutorial",
    "internet browser basics hindi",
    "gmail email tutorial hindi beginners",
    "google docs tutorial hindi",
    "google sheets tutorial hindi",
    "chatgpt tutorial hindi beginners 2024",
    "ai tools for students hindi"
]
```

**Search Parameters:**
- `part=snippet`
- `q=<query>`
- `type=video`
- `maxResults=5`
- `videoDuration=medium` (5-10 minutes preferred)
- `relevanceLanguage=hi` (Hindi preference)

**Output Structure:**
```json
{
  "resources": [
    {
      "topic": "computer basics",
      "videos": [
        {
          "title": "Computer Basics for Beginners in Hindi",
          "channel": "TechGuruji",
          "url": "https://youtube.com/watch?v=ABC123",
          "video_id": "ABC123"
        }
      ]
    }
  ]
}
```

#### Research Summary (Before Validation)

After completing all searches, **summarize the research results** before passing to validation stage.

**Why summarize?**
- Reduces token usage when passing to Stage 3 (validation)
- Provides cleaner input for validation logic
- Allows easy review of what was researched

**Summary Structure:**
```json
{
  "research_summary": {
    "total_searches": 8,
    "topics_researched": [
      "computer basics",
      "typing practice",
      "internet browser basics",
      "gmail email tutorial",
      "google docs tutorial",
      "google sheets tutorial",
      "chatgpt tutorial",
      "ai tools for students"
    ],
    "total_videos_found": 24,
    "video_channels": ["TechGuruji", "LearnWithMe", "DigitalSkills", "HindiTechTutorials"],
    "search_language": "Hindi preference",
    "video_duration_preference": "5-10 minutes",
    "notes": "Most videos from 2023-2024. Good variety of channels. Some topics have fewer results, may need fallback search."
  }
}
```

**Save summary:**
- Append `research_summary` to `research-results.json`
- Validation stage uses summary for context, not raw results

**Checkpoint:** `~/.openclaw/workspace/curriculum-designer-checkpoints/<timestamp>-<session-id>/research-results.json`

---

### Stage 3: Verify Videos + Fallback Logic

**Action:** Verify each video via YouTube oEmbed API. If invalid, retry with alternative search terms.

#### Verification Method

Use oEmbed endpoint (fast, lightweight):
```
https://www.youtube.com/oembed?url=https://youtube.com/watch?v=VIDEO_ID
```

- HTTP 200 = Valid video
- HTTP 404 = Invalid/deleted video
- HTTP 4xx/5xx = Try again (rate limit or temporary error)

#### Fallback Logic (CRITICAL)

For **each topic**, follow this logic:

```
For each video in topic:
  1. Verify via oEmbed
  2. If valid → Add to validated list, done with topic
  3. If invalid → Try next video in topic

If NO valid videos fou