# Creator Screening

> Screen and evaluate social media creators/influencers using configurable quality frameworks. Analyzes Instagram, TikTok, YouTube creators using Memories.ai V2 Video Understanding API — fetches profile metadata, runs MAI visual+audio AI analysis on videos, and scores against production quality, audio, delivery, and positioning criteria. Use when asked to vet creators, screen influencers, evaluate content quality, or generate creator reports.

- Skill: `modbender/creator-screening` (Agent Skill, multi-file: 7 files)
- Install (CLI): `npx skillmds@latest add modbender/creator-screening`
- Raw SKILL.md: https://api.skillmd.com/api/skills/modbender/creator-screening/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: modbender (https://skillmd.com/u/modbender)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/modbender/creator-screening

---


# Creator Screening Skill

Automated creator/influencer screening powered by **Memories.ai V2 Video Understanding API**.

## Parameters

| Parameter | Default | Description |
|-----------|---------|-------------|
| `videos_per_creator` | 5 | Number of top videos to analyze per creator |
| `video_seconds` | 30 | First N seconds of each video to analyze |
| `platforms` | instagram | Supported: instagram, tiktok, youtube |
| `analysis_mode` | `mai` | `mai` (visual+audio AI analysis) or `transcript` (audio-only fallback) |
| `framework` | default | Screening framework to apply. See `references/frameworks/` |
| `output_format` | discord | `discord`, `pdf` (Google Doc→PDF), or `json` |
| `batch_size` | 10 | Max creators per batch run |

## Quick Start

```
Screen these creators: @anshmehra.in, @nishkarshsharmaa
Parameters: videos_per_creator=3, framework=cac-crusher
```

## Workflow

### Step 1: Parse Input
Accept creator URLs in any format:
- Profile: `https://www.instagram.com/username/`
- Individual reel: `https://www.instagram.com/reel/SHORTCODE/`
- YouTube: `https://www.youtube.com/@channel` or `/shorts/ID`

### Step 2: Get Profile & Video Metadata

**Memories.ai V2**: `POST /instagram/video/metadata`

```bash
python3 scripts/scrape_profiles.py --urls "reel_url1,reel_url2" --channel rapid
```

Returns per video ($0.01/video):
- **Owner profile**: username, full_name, followers, verified, profile_pic
- **Video stats**: views, play_count, duration, caption, comments, dimensions, audio info

### Step 3: Video Understanding (MAI)

**Memories.ai V2 MAI**: `POST /instagram/video/mai/transcript`

This is the core analysis step. MAI provides:
- **Visual scene descriptions**: lighting quality, framing, environment, clothing, production value
- **Audio transcription**: speech-to-text with timestamps
- **Content understanding**: topic classification, delivery style, structure

```bash
python3 scripts/analyze_videos.py --mode mai --videos_per_creator 5 --urls "url1,url2"
```

Each video returns visual + audio AI analysis. Use this to evaluate:
- Section 2.1: Look & Feel (lighting, environment, framing) — from **visual scenes**
- Section 2.2: Audio Quality (clarity, echo, consistency) — from **audio analysis**
- Section 3.x: Delivery & Content (structure, fluency, maturity) — from **transcript text**
- Section 4: Positioning (tone, energy, brand safety) — from **combined analysis**

**Fallback**: If MAI is unavailable, use `--mode transcript` for audio-only analysis:
```bash
python3 scripts/analyze_videos.py --mode transcript --videos_per_creator 5 --urls "url1,url2"
```

### Step 4: Apply Framework

Score against the selected screening framework:

```bash
python3 scripts/score_creator.py --framework cac-crusher --profile profile.json --transcripts transcripts.json
```

Frameworks live in `references/frameworks/`:
- `cac-crusher.md` — CAC Crusher Creator Screening Framework (Talking Head + Skit categories)
- `default.md` — Generic quality screening (5 dimensions, weighted scoring)
- `template.md` — Template for creating custom frameworks

### Step 5: Generate Report

Output per-creator screening cards with:
- Profile stats (followers, verified, engagement)
- Per-section scores (PASS/FAIL/FLAG)
- Transcript excerpts as evidence
- Visual quality notes from MAI
- Final verdict (APPROVED / REJECTED / CONDITIONAL)

## Memories.ai V2 API Reference

All endpoints use: `Authorization: <API_KEY>` header (no Bearer prefix).
Base URL: `https://mavi-backend.memories.ai/serve/api/v2`

| Endpoint | Method | Use | Cost |
|----------|--------|-----|------|
| `/{platform}/video/metadata` | POST | Profile + video stats | $0.01/video |
| `/{platform}/video/mai/transcript` | POST | Visual + audio AI analysis | ~$0.11/video |
| `/{platform}/video/transcript` | POST | Audio transcription only | ~$0.01/video |

Platforms: `instagram`, `tiktok`, `youtube`, `twitter`

**Request body**: `{"video_url": "...", "channel": "rapid"}`
**MAI response**: `{"data": {"task_id": "..."}}` (async, results via webhook)

**URL format**: Instagram must use `/reel/SHORTCODE/` (not `/p/` or `/reels/`)

## Error Handling

- Add 0.5s delay between API calls
- Retry failed requests once, then skip
- If MAI webhook not received, fall back to transcript mode
- Always normalize Instagram URLs: `/p/` → `/reel/`, `/reels/` → `/reel/`