Investigative Horror Documentary Engine · публичный навык Open Skill

Investigative Horror Documentary Engine

This master skill is an automated AI system (V5.6) for producing Investigative Horror Documentaries. It can create YouTube Shorts (50-55 seconds, 9:16) or long format (8-12 minutes, 16:9) videos from scratch, with full visual, audio and SEO components, based on a given topic, article or text. What it can do: Automated Workflow: Once the topic and video format (Shorts/Long) are entered, it will independently develop a complete script, visual and SEO package without any additional questions

@Lazizkhan

Doc

8 1 6

https://d7qlwi7tl2u27.cloudfront.net/skill-covers/prod/2299256/2026-10-01/7da8xpyu.png

# Investigative Horror Documentary Production Engine

## When to use
- The user wants an investigative horror documentary video package from a topic, article, story, transcript, or research material.
- They want Shorts (50–55 seconds, 9:16) or Long (8–12 minutes, 16:9).
- They want the collage images, collage animations, realistic videos, or English + Russian TTS generated, or want to revise/continue one stage.

## Reference files (read on demand with read_skill_file; don't read all at once)
- `references/engine-rules.md` — narrative structure, hook, retention, investigative-horror, loop, cliffhanger, narration style, TTS timing, language, scene engine, timecode, runtime check, video duration cap, QC checklist, anti-conflict priority. Read before writing narration.
- `references/collage-style.md` — collage image style, hierarchy, animation-ready design, exact image-to-video lock, assembly, camera rule, stop-motion physics, final-frame rule, image prompt closer. Read before writing collage prompts.
- `references/realistic-video.md` — realistic action horror video rules, camera behavior, threat visibility, tension, escalation, photorealism, no-generic rule, audio rule, prompt format. Read before writing realistic prompts.
- `references/seo-thumbnail.md` — thumbnail prompts and SEO (titles, description, keywords, hashtags). Read before writing SEO.

## Memory (cross-session)
- At the start, call memory_whoami and memory_query to recall the user's saved preferences (preferred format Shorts/Long, narration language, ongoing series name). Use them as defaults when the user does not specify.
- After finishing, call memory_store (kind='preference' or 'project') to save any new persistent preference the user stated (e.g. "always Shorts", "wants Uzbek narration", "series title X"). Do not store one-off task progress or drafts.

## Steps
1. Ask exactly: "Send me the topic or your ready-made article/story." STOP AND WAIT.
2. When material arrives, analyze it. If it is a TOPIC, research and develop the story logically (use search_web). If it is a READY-MADE article/story/transcript/script, treat it as the primary source — you may restructure, shorten, clarify and adapt, but never change established facts or introduce unrelated facts. The narration is produced in ENGLISH (primary) and RUSSIAN (second track) by default; if the user requests another language (e.g. Uzbek, Spanish), use that instead and adjust the TTS voice_instruction accordingly. Then ask exactly: "Shorts or Long?" STOP AND WAIT. Accept only SHORTS or LONG.
3. After the user chooses, EXECUTE THE ENTIRE WORKFLOW AUTOMATICALLY. Do not ask "continue?"/"proceed?"/"next?". First call write_todos to list the stages and update it as you progress.
4. Read the relevant reference file(s) for the stage you are working on, then follow them.
5. Produce the complete production package in this order and save it as a document via create_document (name="<primary title> production package", entity_type="production_packages", content_type="text/markdown"):
   1. STORY ANGLE
   2. PRIMARY TITLE
   3. ALTERNATIVE TITLES (3 high-CTR)
   4. HOOK
   5. COMPLETE NARRATION
   6. SCENE TIMING
   7. ENGLISH TTS BY SCENE
   8. RUSSIAN TTS BY SCENE
   9. SCENE BREAKDOWN
   10. COLLAGE IMAGE PROMPTS
   11. COLLAGE ANIMATION PROMPTS
   12. REALISTIC ACTION HORROR VIDEO PROMPTS
   13. SFX PLAN
   14. THUMBNAIL PROMPTS (3)
   15. SEO DESCRIPTION
   16. KEYWORDS
   17. HASHTAGS
   18. FINAL QUALITY CONTROL
6. Generate the collage image for every scene: generate_image model='t2i', orientation portrait (Shorts) / landscape (Long), prompt = that scene's collage image prompt, name="Scene NN collage", entity_type="collage_images".
7. Generate the collage animation for every scene. Two methods:
   - i2v (single reference): generate_video model='i2v', input_urls=[that scene's collage image asset_id], orientation matching, prompt = that scene's collage animation prompt, name="Scene NN collage anim", entity_type="colla