---
name: sandcastles-shortform-playbook
description: Use when researching viral short-form content with Sandcastles AI, turning outlier videos into TikTok/Reels/YouTube Shorts scripts, building content calendars, creating recording/editor Google Docs, generating three-image visual plans, or applying the AI Business Boomer/Kallaway/cowboy short-form playbook from content discovery through final production handoff.
---

# Sandcastles Shortform Playbook

Use this skill to turn market-proven short-form videos into original, editor-ready content systems.

Primary reference: `references/playbook.md`. Read it before creating scripts, calendars, Sandcastles research packs, Google recording docs, editor briefs, or user-facing playbooks.

## Operating rules

- Use Sandcastles as the evidence layer, not as a copying machine.
- Model repeatable structures: hook, proof, pacing, tension, CTA, and comment loop.
- Do not copy a creator's wording, claims, likeness, or proprietary assets.
- Translate each outlier into the user's niche, customer pain, and business offer.
- Prefer proof-first hooks over generic tips.
- Every final output should be executable by a creator or low-skill editor.
- If a native Sandcastles export is unavailable, say so and use visible page data or manual notes instead.

## OpenClaw / Codex Sandcastles MCP workflow

When Sam asks to use the Kallaway/Sandcastles Hook Machine without Claude Desktop, run it through OpenClaw/Codex using the Sandcastles MCP.

Current local bridge:

- MCP server name: `sandcastles`
- OpenClaw config already contains `https://mcp.sandcastles.ai/`
- Codex can call it through `mcporter` from the workspace config at `config/mcporter.json`
- Verify before use:
  - `npx -y mcporter call sandcastles.ping`
  - `npx -y mcporter call sandcastles.current_org_usage`

Hook Machine operating pattern in Codex:

1. Check connection and credits with `ping` and `current_org_usage`.
2. If Sam provides channels, use `add_channels_to_watchlist` first. If he does not provide channels, use existing watchlist research with `top_hooks`, `top_formats`, `top_topics`, or `search_my_videos`.
3. Do not spend analysis credits casually. `analyze_video` costs one credit per video; use it only when Sam explicitly asks for deep analysis or the requested artifact requires it.
4. Screen out likely paid/boosted videos by default: engagement under 2% on a high performer is suspicious unless Sam asks to include brand deals.
5. Keep channels segmented. Extract hook patterns per creator/channel before doing any cross-channel synthesis.
6. Build three outputs:
   - hook format library with original hook, formula, creator/channel, and performance
   - custom grading rubric based on universal hook principles plus Sandcastles evidence
   - hook generator for Sam's topic with graded/ranked hooks
7. For production handoff, turn the best hooks into recording docs using the required output structure below.

Useful MCP calls:

- `npx -y mcporter call sandcastles.top_hooks --args '{"scope":"watchlist","filters":"AI tools, AI agents, Claude, Codex, small business AI, automation","lookback_days":60,"limit":30,"min_views":10000,"min_engagement":0.02,"min_outlier_score":1}'`
- `npx -y mcporter call sandcastles.get_video_details --args '{"video_uuid":"VIDEO_UUID"}'`
- `npx -y mcporter call sandcastles.search_my_videos --args '{"query":"AI agents automation small business","lookback_days":60,"limit":20,"min_engagement":0.02}'`

## Default workflow

1. **Define the niche and outcome**
   - target audience
   - product/service being promoted
   - desired comment, lead, sale, or follow action

2. **Find outliers in Sandcastles**
   - build or update a watchlist of relevant creators
   - filter recent videos by niche and high outlier score
   - collect creator, title/hook, platform, views, engagement, age, outlier score, and URL when available

3. **Extract viral formats**
   - identify the opening belief correction
   - identify the proof or curiosity gap
   - classify the format using the reference
   - capture the CTA/comment mechanic

4. **Translate into original angles**
   - map the viral format to the user's market
   - make the hook specific to the user's buyer pain
   - add a real business implication: revenue, leads, speed, admin, trust, retention, or cost

5. **Build the production asset**
   - script under 60 seconds unless asked otherwise
   - 3-image visual plan or actual image references
   - exact image-vs-face timeline
   - caption and pinned comment
   - editor brief with captions, zooms, pacing, and export settings

6. **Create the Google Doc handoff when requested**
   - title the doc clearly
   - include Sandcastles evidence, format, script, visuals, timeline, CTA, and editor checklist
   - share with the requested collaborators
   - return the doc link and any row/calendar reference updated

## Required output structure for recording docs

```text
Title:
Viral format:
Sandcastles evidence:
Business translation:
Why the hook works:
First 5 seconds:
Script:
3 visuals / image prompts:
Editor timeline:
Caption:
Pinned comment:
Low-skill editor checklist:
```

## Quality bar

- The first two seconds create curiosity, tension, proof, or a correction.
- A human face appears by roughly second 2–3 unless the user asks for slideshow-only.
- The video explains only what is on screen while the image is visible.
- Opinions, stakes, and CTA usually happen face-to-camera.
- The viewer leaves with a useful mental model, not just hype.
- The CTA creates useful comments or leads.