Edits & Video Skills Index

10 files standalone · local

Edits and video skills — complete operator map

Standalone high-performance guide for the folder designs-content/vip/x content/edits and video skills. Every file is mapped by section, purpose, and how an AI agent (or human editor) should use it. Includes a merge harness for combining files/parts with tools, and for merging two files or parts into one production stack.

10markdown skill files
2subfolders
~8ktotal lines of craft
2merge harness types
File list Apex master Promedic multi-format Effects craft Merge 2 parts Full pipeline

1. File names in this folder

Exact inventory of …/edits and video skills/ (excluding .DS_Store).

# File name Location ~Lines Role
1 Qwen-expert-Ai-editor.md ./ (root) 4013 MASTER Apex AI-Editor v6.2 full system
2 AI-Agent Video Production Guide.md ./ (root) 259 CORE Multi-format Promedic family production
3 Effects and edits.md ./ (root) 925 CORE Craft + CLI img2vid toolkit
4 Effects and videos optimization edits hybrid.md ./ (root) 555 HYBRID Effects × retention × 4 apps
5 Video optimization-virality-hybrid.md ./ (root) 658 HYBRID Agent-ready virality pipeline
6 screening and conversion.md ./ (root) 65 OPS Screen-record → 9:16 short pipeline
7 follow screening and editing output.md ./follow up/ 502 Paths, prompts, server, chrome crop
8 opus-screening.md ./extra to review/ 262 EXTRA 2K H.265 server screening batch
9 models-video virality-answers.md ./extra to review/ 869 EXTRA Multi-model virality research dump
10 python-Video-optimization-scripts and limitations.md ./extra to review/ 418 EXTRA Master Python optimizer + limits

2. Role map — which file owns which job

Use this as the routing table before opening any file. Prefer the thinnest file that answers the task; escalate to Apex only when you need a full cognitive pipeline.

JobPrimary fileSupport file(s)
Full autonomous edit pipeline (perceive→QA) Qwen-expert-Ai-editor.md Effects and edits (shell funcs), verification layer
Promedic app brand grades + format templates AI-Agent Video Production Guide.md Effects + opt hybrid (per-app FILE 1–4)
Pick color grade / transition / cut rhythm Effects and edits.md Effects hybrid Part 1 table
Ken Burns / captions / xfade code for agents Effects and edits.md (CLI section) Qwen Pillar 5–6, python scripts
Short-form retention engineering Video optimization-virality-hybrid.md Effects hybrid Part 2, models-virality
Long-form / LinkedIn / YouTube pacing AI-Agent Video Production Guide Part 2 Effects hybrid Part 3
Screen capture → encode → QA checklist screening and conversion.md follow up (chrome/dock crop), opus-screening
Server batch 2K H.265 screening opus-screening.md follow up SSH + paths
Batch FFmpeg without LLM creativity python-Video-optimization-scripts… virality hybrid AGENTS.md steps
Research / debate virality claims models-video virality-answers.md virality hybrid “inconsistencies” section

3. Detailed file sections

Each file below lists its internal sections, what each section is for, and concrete “use this part when…” rules.

01

Qwen-expert-Ai-editor.md

MASTER · 4013 lines Apex AI-Editor System v6.2 Unified Master Edition

What the whole file is: Production-grade cognitive video editor for CLI AI agents. 5-stage loop (Perceive → Interpret → Compose → Realize → Critique) plus 16 pillars covering install, vision, motion, render, QA, skills, orchestrator, system prompt, visual judge, brain, theme seeds, VLM critique, loudness, and a power-tips vault.

Sections (map)

Header / merge provenance meta

Status, architecture summary, merge sources (v5–v6.2), fix tables. Use when auditing what was absorbed from sibling tip files.

Expert Philosophy + Operating Charter rules

Six Deadly Sins (VFR, A/V drift, context decay, false positive, blind acceptance, face leak) and free/local/light/stable charter. Use as non-negotiable agent preamble.

Pillar 1 — Hardened Environment install

install_apex.sh: brew, ffmpeg, venv, Python deps, Playwright optional, Whisper, verification. Use for first-time setup on Mac/Linux.

Pillar 2 — Defensive Core (core.py) code

StateEngine, VFR-safe runner, FFmpeg escaping, platform delivery budgets. Use whenever writing or invoking render helpers.

Pillar 3 — Perceive vision

vision_bridge.py multi-key multi-model cloud vision; perceive.py local OpenCV (real colorfulness) + optional cloud. Use for image folder analysis before edit decisions.

Pillar 4 — Interpret beats

Beat-to-asset meaning match; dual templates ad (5 beats) vs narrative (7 beats). Use to assign images to story beats by meaning, not aesthetics alone.

Pillar 5 — Compose motion

Motion + transitions per beat. Use to choose Ken Burns / zoom / xfade type per beat energy.

Pillar 6 — Realize render

VFR-safe render, xfade math, audio drift fix (apad + -shortest, -ar 48000). Use for actual export commands.

Pillar 7 — Critique & 5-Tier QA gate

critique.py, qa_gate.py, verification_engine.py — PASS/FAIL/INCONCLUSIVE, blocks false “done”. Use before claiming a render is finished.

Pillar 8 — Skills tools

Web research, registry, PDF→images (CMYK-safe), task tracker, image style prompt library (Styles A–C + animation). Use for research, PDF ingest, style prompting.

Pillar 9 — Pipeline Orchestrator cli

End-to-end CLI assembly of the 5-stage loop. Use as the main entry script shape.

Pillar 10 — Master System Prompt prompt

Identity, 5-stage loop, absolute hard rules, verification protocol, hybrid perception (eyes vs scripts), editorial decision framework (color intent, transitions, pacing, effect WHY), pro grade recipes, deliverables, deployment. This is the agent system prompt block to paste into AGENTS.md / session preamble.

Pillar 11 — Lightweight Tool Catalog tools

Free/local tools list + CapCut i2v primary path. Use when choosing tooling without paid SaaS.

Pillars 12–15 — Judge · Brain · Theme Seeds · VLM · Loudnorm expert

visual_judge 3-tier verdict; multi-agent brain; save EXCELLENT theme seeds; API semantic critique; EBU R128 loudnorm. Use for expert-tier quality and memory of what works.

Pillar 16 — Power Tips Vault tips

Folder-wide adopted tips (faceless force, stills→motion craft, HDR→SDR, hook math, kits). Use as secondary craft reference after core pillars.

How to use this file
  • You can use Pillar 10 system prompt as the permanent agent identity for any video edit task.
  • You can use Pillars 1–9 code blocks to scaffold a local Apex install + pipeline.
  • You can use Pillar 7 + final_gate as the mandatory exit condition before saying “done”.
  • You can use Pillar 16 when you need craft tips without leaving the master file.
  • Do not use this as the only source for Promedic per-app color numbers — merge with AI-Agent Video Production Guide Section 1.
02

AI-Agent Video Production Guide.md

CORE · 259 lines Multi-format system for Promedic app family

What the whole file is: Locked production bible for Dose Calculator, Female ProMedic, Coach ProMedic, Dentist Pro across short-form 9:16, long-form 16:9, and feed 4:5/1:1. Source values vs proposed values are marked.

Sections

§0 Universal Rules always-on

Faceless absolute rule; readability beats style; non-destructive capture; hard cuts default; negative/mistake hook framing; brand-family caption consistency; Dose↔Dentist differentiation notes.

§1 Per-App Variable Table grades

WB, exposure, blacks, highlights, sat, grain, vignette, motion blur, accent, zoom ceiling, hook window, interrupt cadence, long-form visual-change cadence — one column per app. This is the single source of truth for brand look numbers.

PART 1 — Short-form 9:16 template

Copy-paste instruction template (Record → Grade → Motion → Safe zone → Captions → Interrupts → Audio → Loop → Export → QA); filled Coach example; short-form pro-tips per app.

PART 2A — Long-form 16:9 youtube

Same grade, slower holds, kinetic only in hook, chapters, title-safe margins, export 1920×1080.

PART 2B — Feed 4:5 & 1:1 instagram

Grid-thumbnail crop protection; 1080×1350 primary / 1080×1080 alt.

Changelog meta

Explains resolved conflicts (zoom ceiling vs hook punch, universal grade → per-app table, etc.).

How to use this file
  • You can use §1 table to fill every {VARIABLE} in the short/long/feed templates for a specific app.
  • You can use PART 1 template as the agent brief for TikTok/Reels/Shorts app demos.
  • You can use PART 2A when expanding a short cut into YouTube without re-deriving brand look.
  • You can use §0 faceless + readability rules as hard constraints injected into any image-gen or edit prompt.
03

Effects and edits.md

CORE · 925 lines Craft theory + ready kits + CLI image-to-video code

What the whole file is: Two layers: (A) editorial craft for grades/transitions/effects/rhythm + content kits; (B) copy-paste shell + Python pipeline so agents can execute those looks with FFmpeg only.

Sections

Opening craft brief theory

Best techniques for video/images/animation; post-effects as seasoning; Script Density secret; Fake BTS caution (trust vs evasion).

Color Grading looks

Teal & orange, bleach bypass, warm nostalgic, clean high-key, moody teal, vibrant pop, B&W — each with use/avoid.

Transitions cuts

Hard cut, J/L-cut, whip pan, match cut, zoom transition, glitch, light leak, dissolve, speed-ramp.

Effects matched to tone table

Grain, leaks, blur, CA, vignette, flare, shake — when to reach / skip.

Cut rhythm by situation pacing

Hook, motivational, storytime, listicle, true crime, comedy, tutorials — different clocks.

Ready-to-use kits presets

Content-type → grade + transitions + effects + rhythm in one row.

Hook attention math + Parts 1–5 review review

1.3s TikTok / 2.1s Reels hook windows; shot list consistency; 9:16 sharp-core/blurred-fill; tool notes; expert tips (textless master, no hallucinated medical numbers).

CLI-Agent Expert Editor code code

Hard rules (framerate 30, ar 48000, apad+shortest); shell: probe, Ken Burns, zoom punch, fit_916, grade presets, grain/vignette/leak/blur, xfade, burn captions, mux audio, speed ramp; Python full pipeline stages 1–5 + QA; decision matrix; self-check script; integration notes.

How to use this file
  • You can use Ready-to-use kits when picking a full look for motivational vs listicle vs tech in one decision.
  • You can use Clean high-key row as the default grade for app UI screens.
  • You can use kb_zoom_in / fit_916 / grade_* shell functions as drop-in agent tools.
  • You can use Python Stage 1–5 assembler for batch stills → vertical video with transitions + captions + QA.
  • You can use self-check script after any pipeline run (exists, has video, 1080×1920, ~30fps, duration > 0).
04

Effects and videos optimization edits hybrid.md

HYBRID · 555 lines Effects craft × virality retention × 4-app production packs

What the whole file is: Explicit hybrid answer: Part 1 reviews effects uses; Part 2 merges effects with short-form retention; Part 3 merges for long-form; then four full per-app production files (Dose, Female, Coach, Dentist) each with 8 subsections.

Sections

Part 1 — Review + effect uses table reference

When each effect is best; default recommendation for app screens = clean high-key + Ken Burns + kinetic captions.

How to use Parts 2 & 3 routing

Part 2 = default short system; Part 3 = long adjustments; start short, relax pacing for long; keep grade/captions unified.

Part 2 — Short hybrid stack tiktok

Core retention rules (hook 1–2.5s, interrupt 1.8–3.2s, kinetic captions, no dead air, loop, 9:16); full hybrid stack for app screens (visual base, hook, pacing, captions, audio, transitions, export).

Part 3 — Long hybrid youtube

Adjustment table short→long; recommended long-form hybrid for app screens.

Quick app summaries (1–4) + universal rules apps

Condensed identity per app before the deep packs.

FILE 1–4 deep packs (Dose / Female / Coach / Dentist) production

Each: (1) Brand positioning (2) Visual identity (3) Screen recording + post workflow (4) Short retention hybrid (5) Long adaptation (6) Platform growth notes (7) Ready agent prompts (8) Common mistakes.

How to use this file
  • You can use Part 2 hybrid stack as the default short-form operating procedure for any app screen video.
  • You can use FILE 3 Coach §7 agent prompts as paste-ready briefs for fitness app content.
  • You can use FILE 1 Dose §8 mistakes to hard-code “never do X” constraints for clinical demos.
  • You can use Part 3 adjustment table when the same project must ship YouTube + Reels without brand drift.
05

Video optimization-virality-hybrid.md

HYBRID · 658 lines CLI-agent virality synthesis + AGENTS.md + code snippets

What the whole file is: Turns multi-source virality research into ordered, measurable agent operations. Emphasizes retention curve over vanity metrics.

Sections

Hybrid answer / review for CLI agents context

Strengths, inconsistencies (exact % soft), adaptation for multimodal agents, 8-step agent-friendly pipeline, high-priority principles, what to de-emphasize.

Ranked ROI order (1–7) priority

Hook → dead air/interrupts → kinetic captions → loop → audio → export/variants → secondary polish only last.

Complete sequential agent pipeline steps

Steps 0–7 with detection rules and actions; platform overrides; master prompt skeleton; how to use it.

Copy-paste System Prompt + AGENTS.md prompt

Ready system prompt + step checklist for short-form viral optimizer agent.

Supporting code examples code

Jump-cut + zoom helper; safe-zone kinetic captions (pysubs2); image sequence Ken Burns; final export template; quick-start examples.

How to use this file
  • You can use Ranked ROI order to sequence agent work so polish never steals time from hooks/captions.
  • You can use System Prompt + AGENTS.md block as a dedicated short-form optimizer subagent.
  • You can use Step 1 hook reconstruction as a standalone pass on any existing screen recording.
  • You can use failure modes section as hard-coded guards (logo open, dead air, no captions, watermark, etc.).
06

screening and conversion.md

OPS · 65 lines Minimal 3-phase screen → short-form instructions

What the whole file is: Compact universal pipeline for turning a clean app screen recording into a publishable 9:16 short. Generic [APP NAME] placeholder.

Sections

PHASE 1 — RECORD capture

High-res native capture; no effects while recording; HDR→SDR first; edge margin 10%; archive raw forever.

PHASE 2 — PROCESS (ordered) edit

(1) Color grade locked preset (2) Ken Burns punch-ins (3) Safe zone (4) Captions (5) Numbers & pattern interrupts (6) Audio LUFS (7) Export H.264 1080×1920.

PHASE 3 — QA gate

Muted playback, real phone preview, grade drift check, illegible numbers check.

How to use this file
  • You can use this as the shortest agent brief when you do not need per-app branding tables.
  • You can use PHASE 1 alone as capture SOP before any editing stack.
  • You can use PHASE 2 order as the fixed processing sequence (never grade after captions burn, etc.).
  • Merge with AI-Agent Guide §1 to replace the generic grade with app-specific numbers.
07

follow up/follow screening and editing output.md

Ops notebook: SSH, paths, prompts, chrome crop, merge notes

What the whole file is: Working notes that connect screening/editing skills to real paths, servers, episode styles, faceless editor prompt, reverse-style workflow, merge instructions, and chrome/dock removal for clean UI captures.

Sections (logical groups)

SSH roster + VIP memory paths infra

hostinger / contabo / hetzner / ai-developer; permanent memory file locations.

Episode styles registry styles

Type-C, type-4/5/6/7, Dentist episode style numbers — for comic/editorial asset generation.

Screening encode settings + FFmpeg OS notes encode

2K/native, CRF 18–21, hvc1 tag, yuv420p, no-sao for screen content; avfoundation / x11grab / gdigrab.

Editor prompt (faceless force) prompt

Master creative editor identity + absolute no-eyes/no-face features rule.

Video-similar animation / reverse-engineer path style

How to reverse screened videos into style .md files; free light tools; path to rec-screened-videos.

Merging + HTML template notes meta

Instructions for valuable merges and building standalone HTML indexes (this page pattern).

Chrome & dock removal capture

Crop/zoom rules so recordings show clean app UI only.

Paths inventory (media, screens, apps, content) paths

Where screened videos, images, episode prompts, app stacks live on disk.

How to use this file
  • You can use Paths to locate raw screens / outputs without guessing folders.
  • You can use Chrome and dock removal as a pre-flight rule on every Mac screen recording.
  • You can use Editor prompt as the creative identity when generating faceless assets.
  • You can use Merging section as the policy for how agents should combine skill files.
08

extra to review/opus-screening.md

EXTRA · 262 lines 2K H.265 9:16 server screening batch (25 videos)

What the whole file is: Spec + remote layout for Opus-driven 2K HEVC screening on Hetzner (/root/opus-server-videos/), 50s × 25 videos across 5 apps, EN/AR scripts, master record script.

Sections

Summary of specs & configuration spec

1440×2560, libx265 crf 18, hvc1, 50s duration, human-like mouse trajectories.

Remote output structure tree

Per-app folders (clinical, drugs, coach, ielts-fast, the-ielts-fast) + root SEO docs + record_all_2k_h265.py.

Audio script & SEO docs copy

EN/AR description files with voiceover + platform SEO.

Useful server commands ops

SSH commands to inspect/run the batch on the remote host.

How to use this file
  • You can use this as the remote batch screening contract when generating many app feature videos off-Mac.
  • You can use the folder tree as the expected artifact layout after a batch run.
  • Merge with screening and conversion.md for post-process (grade/captions) after 2K capture.
09

extra to review/models-video virality-answers.md

EXTRA · 869 lines Multi-model deep dive on viral factors (research vault)

What the whole file is: Aggregated answers from multiple models on algorithm signals, hooks, editing, effects, platform specs, strategy, tools, diagnostics, and a long “ultimate deep-dive” + second-pass sections. Treat numbers as directional; operationalize via virality-hybrid.md.

Major section groups

Retention / engagement / 1.3s rule algo

Watch-time signals, engagement hierarchy, first-second rules.

Hook factors + editing + effects craft

Patterns that work/kill; pacing; movement; transitions; captions; color/SFX; trending effects.

Platform-specific (TT / Reels / Shorts / LinkedIn) platform

Per-platform nuances and B2B notes.

Strategy · tools · diagnostic · 2026 checklist ops

Hook-body-CTA, mute test, posting mechanics, tool stack, “why not viral” framework, checklist.

Ultimate deep-dive (11 chapters) + anatomy of 30s viral deep

Technical specs through thumbnails; second synthesis on algorithmic drivers and structural pacing.

How to use this file
  • You can use this as a research library, not as hard agent law.
  • You can use the 2026 viral checklist as a human review list before publish.
  • You can use platform-specific sections when one platform underperforms the same cut.
  • Always re-ground soft numbers through Video optimization-virality-hybrid.md before coding them as thresholds.
10

extra to review/python-Video-optimization-scripts and limitations.md

EXTRA · 418 lines Master Python viral optimizer + when not to use pure code

What the whole file is: Comparison of Agent form vs single Master Python script; full ready script (images or video → silence cut, captions, 9:16, variants, punch-in); usage; honest limitations.

Sections

Agent vs Master Script comparison table decide

Intelligence vs reliability; when to pick which.

Master Python Script code

Images→video Ken Burns, silence removal, kinetic captions, export, variants; optional faster-whisper / pysubs2 / opencv.

How to use the Master Script cli

FFmpeg required; CLI examples for video and image folder modes.

Limitations 1–4 + Summary caveats

Complexity explosion, pure code vs intelligence, FFmpeg/Python practical limits, design goal of the script.

How to use this file
  • You can use the Master Script for reliable batch processing of many similar clips.
  • You can use the Agent form (from virality-hybrid) when hooks/claims need creative judgment.
  • You can use the limitations section to stop over-engineering a pure-code “full editor”.
  • Best hybrid: Python for mechanical steps + agent for hook rewrite / claim language.

4. Merging harness

Two merge types. Type 1 pairs a skill file (or a part of it) with an external tool/runtime. Type 2 merges two skill files or two parts (or a full file + a part) into one agent brief or production stack — this is the high-value path and is expanded below.

Formula: skill_file[section] + tool → capability

T1-01 Screening pipeline + FFmpeg

screening and conversion.md [PHASE 1–3] + ffmpeg / ffprobe → deterministic record→grade→caption→export pipeline
Outcome: Clean 9:16 H.264 short from a raw screen .mov without inventing brand rules.
  1. Follow PHASE 1 capture rules (archive raw first).
  2. Implement PHASE 2 steps with FFmpeg filters (eq/colorbalance, zoompan, subtitles, loudnorm).
  3. Run PHASE 3 checklist; do not publish on partial pass.

T1-02 Effects CLI toolkit + shell agent

Effects and edits.md [CLI-Agent Expert Editor · shell functions] + zsh/bash + ffmpeg on PATH → agent can call kb_zoom_in, fit_916, grade_*, burn_captions, xfade_transition
Outcome: Image folder becomes vertical clips with motion, grade, captions — no Python deps.

T1-03 Apex master + local venv + optional cloud vision

Qwen-expert-Ai-editor.md [Pillars 1–9 + 7 QA] + install_apex.sh environment + (optional) free vision APIs → full 5-stage cognitive pipeline with final_gate
Outcome: Autonomous edit with disk checkpoints and three-state verification.

T1-04 Virality AGENTS.md + multimodal CLI agent

Video optimization-virality-hybrid.md [System Prompt + AGENTS.md] + Grok Build / AGY / Claude agent runtime → retention-first rewrite of existing videos (hook, interrupts, captions, loop)
Outcome: Creative optimization layer that pure FFmpeg cannot invent (claims, framing).

T1-05 Master Python script + Whisper

python-Video-optimization-scripts… [Master Script] + FFmpeg + faster-whisper + pysubs2 → batch mechanical optimizer (silence cut, captions, 9:16, variants)
Outcome: Cheap reliable batch; not creative hooks — pair with agent for language.

T1-06 Opus screening + remote Hetzner

opus-screening.md [tree + record_all_2k_h265.py] + ssh hetzner + server GPU/CPU → 25× 2K H.265 feature videos off local Mac
Outcome: Bulk raw/demo captures for later local or Apex post-edit.

Follow-up chrome crop + screen capture tool

follow screening… [Chrome and dock removal] + browser window bounds / crop filter / OBS region → UI-only recordings (no chrome tabs, no Mac dock)
Outcome: Source footage that does not need heavy crop in post.

T1-08 Production Guide templates + CapCut / Resolve

AI-Agent Video Production Guide.md [PART 1 or 2 templates] + CapCut or DaVinci Resolve (manual or semi-auto) → human-operated brand-correct cut when agent render is unavailable
Outcome: Same brand numbers and safe zones in a GUI NLE.

Formula variants:

A) file_A + file_B → unified stack B) file_A[part] + file_B[part] → focused brief C) file_A + file_B[part] → full system + specialized slice D) file_A[part] + file_B → specialized slice + full companion E) three-way: capture + brand + retention (recommended production path)

Type 2A — Full file + full file

T2-A01 Effects craft + Virality pipeline

Effects and edits.md ⊕ Video optimization-virality-hybrid.md → “what look” + “what retention sequence” in one agent
Get: Agent that can both pick a kit (grade/transition/rhythm) and execute ROI-ordered retention steps (hook → captions → loop).
  1. Load Effects kits for visual language.
  2. Load virality ranked ROI as execution order.
  3. Forbid fancy transitions until steps 1–5 of ROI are done.

T2-A02 Production Guide + Effects hybrid (per-app packs)

AI-Agent Video Production Guide.md ⊕ Effects and videos optimization edits hybrid.md → numeric brand table + narrative app packs + short/long hybrid stacks
Get: Complete Promedic multi-app production system (numbers + prompts + mistakes + long/short).

T2-A03 Apex master + Production Guide

Qwen-expert-Ai-editor.md ⊕ AI-Agent Video Production Guide.md → cognitive pipeline constrained by faceless + per-app grade/zoom ceilings
Get: Autonomous editor that never invents brand grades and always respects faceless + safe zones.

T2-A04 Screening mini-SOP + Virality hybrid

screening and conversion.md ⊕ Video optimization-virality-hybrid.md → capture order + retention reconstruction for any app
Get: From raw screen to viral-ready short without per-app branding (fast path).

T2-A05 Virality hybrid + Python master script

Video optimization-virality-hybrid.md ⊕ python-Video-optimization-scripts… → agent creativity for hooks + script reliability for batch mechanics
Get: Hybrid of intelligence and throughput (agent designs; script executes batch).

T2-A06 Opus screening + Screening conversion

opus-screening.md ⊕ screening and conversion.md → bulk 2K capture then standard post-process phases
Get: Server raws → local/social delivery pipeline.

Type 2B — Part + part (focused merges) — extend heavily

T2-B01 Brand numbers ⊕ Short template

AI-Agent Guide [§1 Per-App Variable Table] ⊕ AI-Agent Guide [PART 1 Short-form template] → filled short-form brief for one app
Get: Zero free variables left — agent only executes.

T2-B02 Clean high-key craft ⊕ Short hybrid stack

Effects and edits.md [Color Grading · Clean high-key] ⊕ Effects hybrid [Part 2 Recommended Hybrid Stack] → app-screen visual + retention recipe
Get: Default look for SaaS/app demos with algorithm-safe pacing.

T2-B03 Ken Burns code ⊕ Hook reconstruction

Effects and edits.md [kb_zoom_in / zoom_punch shell] ⊕ Virality hybrid [Step 1 Multi-channel hook 0–3s] → mechanical punch-in + creative hook text/audio rules
Get: First 3 seconds that both move and claim value.

T2-B04 Apex hard rules ⊕ Production faceless rule

Qwen [Six Deadly Sins + Absolute Hard Rules] ⊕ AI-Agent Guide [§0 Universal Rules] → technical + brand non-negotiables in one preamble
Get: Single agent system block that fails closed on VFR, drift, faces, and unreadable UI.

T2-B05 Apex QA gate ⊕ Screening QA checklist

Qwen [Pillar 7 5-Tier QA + final_gate] ⊕ screening and conversion.md [PHASE 3 QA] → technical gate + product/publish gate
Get: File integrity AND muted/phone/grade checks before publish.

T2-B06 Ready-to-use kits ⊕ Cut rhythm

Effects and edits.md [Ready-to-Use Kits table] ⊕ Effects and edits.md [Cut Rhythm by Situation] → full kit with situation-specific timing
Get: “Listicle kit” that also enforces metronomic fact cuts.

T2-B07 Coach deep pack ⊕ Coach short template

Effects hybrid [FILE 3 Coach ProMedic §§1–8] ⊕ AI-Agent Guide [PART 1 filled with Coach variables] → maximum-depth Coach production brief
Get: Brand feeling + agent prompts + mistakes + numeric template in one.

T2-B08 Dose clinical pack ⊕ Long-form 16:9

Effects hybrid [FILE 1 Dose] ⊕ AI-Agent Guide [PART 2A Long-form] → clinical LinkedIn/YouTube walkthrough system
Get: Trust-preserving long-form for medical tools (restrained motion, chapters).

T2-B09 Encode settings ⊕ Chrome crop

follow up [Screening encode CRF/hvc1 notes] ⊕ follow up [Chrome and dock removal] → clean UI capture at web-compatible HEVC/H.264 settings
Get: Source masters ready for browser/YouTube without chrome chrome.

T2-B10 Ranked ROI ⊕ Part 2 short hybrid

Virality hybrid [Ranked ROI Order] ⊕ Effects hybrid [Part 2 Core Retention Rules] → single ordered short-form law (no contradictions)
Get: One checklist: hook → interrupts → captions → loop → audio → export → polish.

T2-B11 Models platform notes ⊕ Virality platform overrides

models-video virality-answers.md [Platform-Specific] ⊕ Virality hybrid [Platform Overrides] → research context + agent-executable overrides
Get: Soft research grounded into hard agent branches (TT vs Reels vs Shorts).

T2-B12 Editorial decision framework ⊕ Color grading section

Qwen [Pillar 10 · Pro Color Grading Recipes + Color Intent Map] ⊕ Effects and edits.md [Color Grading] → WHY-first grade selection with FFmpeg recipes
Get: Agent picks grade by mood then applies tested filter string.

T2-B13 Compose pillar ⊕ Shell xfade/Ken Burns

Qwen [Pillar 5 Compose] ⊕ Effects and edits.md [xfade_transition + Ken Burns functions] → decision layer + executable motion primitives
Get: Beat-aware motion without re-deriving FFmpeg filtergraphs each time.

Faceless editor prompt ⊕ Apex face-leak sin

follow up [editor prompt · NO EYES…] ⊕ Qwen [Deadly Sin 6 Face Leakage + force-faceless] → creative prompt language + technical sanitize (OpenCV blur) + QA override
Get: End-to-end faceless enforcement (prompt → render → fail if eyes appear).

T2-B15 PHASE 2 process order ⊕ Apex Realize pillar

screening and conversion.md [PHASE 2 steps 1–7] ⊕ Qwen [Pillar 6 Realize · VFR/A-V rules] → ordered human SOP enforced with production render invariants
Get: Grade/motion/captions order that never produces VFR or drift bugs.

Type 2C — File + part (extend)

T2-C01 Full Apex ⊕ only §1 brand table

Qwen-expert-Ai-editor.md (full) ⊕ AI-Agent Guide [§1 table only] → general cognitive editor specialized to Promedic numbers
Get: Apex does not invent grades; injects {APP} column at compose time.

T2-C02 Full virality hybrid ⊕ Dose FILE 1 only

Video optimization-virality-hybrid.md (full) ⊕ Effects hybrid [FILE 1 Dose Calculator] → viral short optimizer specialized to clinical tone/mistakes
Get: Retention engineering that will not apply light leaks / energetic grades to clinical UI.

T2-C03 Full Effects and edits ⊕ Part 2 short hybrid only

Effects and edits.md (full craft + code) ⊕ Effects hybrid [Part 2 short stack] → rich toolkit constrained by short-form retention law
Get: Agents can use every shell function but must still hit hook/interrupt/caption rules.

T2-C04 Full screening SOP ⊕ follow-up paths part

screening and conversion.md (full) ⊕ follow up [Paths · Ai-screen / khaled screening / images trial] → process + correct disk locations for I/O
Get: Agent knows where to read raws and where to write outputs.

T2-C05 Full Python script ⊕ Virality Step 1 only

python scripts (full master) ⊕ Virality hybrid [Step 1 Hook reconstruction only] → batch engine with human/agent-written hook pass first
Get: Mechanical pipeline that refuses to start without a designed hook.

Type 2E — Recommended three-way production merges

T2-E01 Capture · Brand · Retention (default production path)

screening and conversion.md [PHASE 1–3] ⊕ AI-Agent Guide [§0–§1 + PART for format] ⊕ Virality hybrid [ROI order + AGENTS steps] → complete short/long production path for any Promedic app
Get: From empty project to publish-ready cut with brand fidelity and algorithm structure.

T2-E02 Apex · Brand · Effects code

Qwen [Pillars 4–7 orchestrator+QA] ⊕ AI-Agent Guide [§1 table] ⊕ Effects and edits.md [shell functions] → autonomous pipeline with brand numbers and proven FFmpeg primitives
Get: Best “ship quality” autonomous stack in this folder.

T2-E03 Server capture · Post SOP · App pack

opus-screening.md ⊕ screening and conversion.md ⊕ Effects hybrid [FILE n for target app] → bulk remote capture refined with app-specific hybrid
Get: Scale screening without losing brand/retention quality downstream.

Merge policy (from follow-up notes)

Rules for any Type 2 merge

  1. Pick winners, don’t average losers. Prefer AI-Agent §1 numbers over generic grades in older hybrid text when they conflict.
  2. Order is law. Capture → grade → motion → captions → audio → export → QA. Never reverse.
  3. Retention before polish. Virality ROI outranks light leaks / heavy grain for app UI.
  4. Faceless + readability always win. If a merge introduces faces or unreadable doses, the merge is invalid.
  5. Hard cuts default ~80–90%. Fancy transitions only after ROI steps pass.
  6. Three-state QA. INCONCLUSIVE is not PASS. Block “done”.
  7. Output shape. Merged deliverable should be one .md or one agent prompt with: rules, ordered steps, code blocks, QA checklist — no orphaned contradictions.

Pick two sources (file or file+part). The builder writes a merge recipe you can paste into an agent brief.

Generated merge brief
Select sources and click Build…

Quick reference: merge left with right to get the stated capability.

Merge this… …with this …to get
Effects and edits [kits] Virality hybrid [ROI] Look + retention order
AI-Agent Guide [§1] screening [PHASE 2] App-branded processing
AI-Agent Guide [PART 1] Effects hybrid [FILE n] Deep per-app short system
Qwen [Pillar 10 prompt] AI-Agent Guide [§0] Editor identity + faceless law
Qwen [Pillar 6 Realize] Effects [shell toolkit] Safe render + primitives
Qwen [Pillar 7 QA] screening [PHASE 3] Double gate (file + product)
Virality [AGENTS.md] python master script Creative + batch hybrid
opus-screening screening conversion Bulk capture → post
follow [chrome crop] screening [PHASE 1] Clean UI capture SOP
Effects hybrid [Part 2] Effects hybrid [Part 3] Short↔long dual export system
models [platform] Virality [overrides] Research-backed platform branch
Effects [CLI code] FFmpeg binary Executable stills→video toolkit
Qwen [full] AI-Agent Guide [§1] Autonomous Promedic editor
Virality [Step 3 captions] Effects [burn_captions] Kinetic caption rules + burn tool
AI-Agent [PART 2A] Effects hybrid [Part 3] Long-form brand + pacing hybrid
follow [editor prompt] Qwen [face leak sin] Faceless creative + sanitize + QA
screening [full] Virality [full] Capture-to-viral generic path
Effects [self-check] Qwen [final_gate] Lightweight + full verification stack

5. When to use what (decision playbook)

Start at the first row that matches. Prefer fewer files open at once.

SituationOpenMerge pattern
I only need a clean screen short today screening and conversion.md T1-01 (+ FFmpeg)
Promedic app, correct brand look AI-Agent Video Production Guide.md T2-B01 (§1 ⊕ PART template)
Need retention / virality structure Video optimization-virality-hybrid.md T2-B10 or T2-A04
Need motion code for stills Effects and edits.md (CLI section) T1-02
Full autonomous pipeline + QA Qwen-expert-Ai-editor.md T2-E02 (Apex ⊕ brand ⊕ shell)
Deep per-app prompts & mistakes Effects … hybrid.md FILE 1–4 T2-B07 / T2-C02
25 videos on remote server opus-screening.md T2-E03
Batch without LLM python-Video-optimization-scripts… T2-A05
Where are the files on disk? follow screening… Paths T2-C04
Research / debate claims models-video virality-answers.md T2-B11 (do not hardcode soft %)

6. End-to-end recommended pipeline

Default path for Promedic-style app screen content (short-form first, then expand).

Ordered stages

  1. Locate I/O — follow up Paths (raw screens, output folders).
  2. Capture — screening PHASE 1 + chrome/dock removal; archive raw.
  3. Optional bulk — opus-screening on Hetzner for many features.
  4. Brand lock — AI-Agent Guide §0 + §1 for the target app.
  5. Retention structure — Virality ROI order / Effects hybrid Part 2.
  6. Execute motion/grade/captions — Effects shell toolkit or Apex Realize.
  7. Long/feed variants — AI-Agent PART 2A/2B or Effects hybrid Part 3.
  8. QA — screening PHASE 3 + Apex Pillar 7 / final_gate (three-state).
  9. Publish — only when all gates are PASS (not INCONCLUSIVE).
Canonical merge for this path: T2-E01 (capture ⊕ brand ⊕ retention). Escalate to T2-E02 when the agent should own the full cognitive loop.