Vertical Video Producer

Shortform production

by ericosiu · MIT license · ★ 3,615 on GitHub

Produce readable vertical shorts with evidence-based B-roll, three opening treatments, editorial review, and authorized Metricool API delivery. Use for shortform edits, captions, hook experiments, or scheduling approved clips.

---
name: shortform-production
description: Produce readable vertical shorts with evidence-based B-roll, three opening treatments, editorial review, and authorized Metricool API delivery. Use for shortform edits, captions, hook experiments, or scheduling approved clips.
---

# Shortform Production

Keep the visual style consistent. Change the storytelling to suit the source. An experiment is a hypothesis, not a proven retention gain.

## Preamble

When available, use the repository's version check and telemetry initializer:

```bash
python3 telemetry/version_check.py 2>/dev/null || true
python3 telemetry/telemetry_init.py 2>/dev/null || true
```

Overview

Keep the visual style consistent. Change the storytelling to suit the source. An experiment is a hypothesis, not a proven retention gain.

Manual install (npx, ZIP)

Claude Codenpx skills add ericosiu/ai-marketing-skills --skill shortform-production -g -a claude-code

Claude Code (degit)npx degit ericosiu/ai-marketing-skills/shortform-production#main ~/.claude/skills/shortform-production

Claude.ai / desktopZip the folder and upload it under Settings → Capabilities → Skills.

Codex, Cursor, other agentsPoint them at SKILL.md. The other files are plain text.

Connects with Agent Alley MCP · agentalley.io/mcp

Files included

SKILL.md2.7 KB

Read SKILL.md here

ericosiu/main1 file shown

SKILL.md2.7 KB

Browse every file on GitHub↗

SKILL.md

Show the full text38 lines

shortform-production/SKILL.md38 lines · 2.7 KB

PreviewCode

Outline

RawView on GitHub

Shortform Production

Keep the visual style consistent. Change the storytelling to suit the source. An experiment is a hypothesis, not a proven retention gain.

Preamble

When available, use the repository's version check and telemetry initializer:

python3 telemetry/version_check.py 2>/dev/null || true
python3 telemetry/telemetry_init.py 2>/dev/null || true

Remote telemetry requires opt-in. Never log content, paths, account details, or credentials.

Choose the work

Edit: Read V5 style and creative formats. Use the named source and timestamped transcript. Preserve credentials, claim qualifiers, natural speech, and the CTA. Inventory real assets before designing inserts. Use an available renderer; this package supplies editorial rules and delivery tools.
Experiment: Produce three opening treatments for one clip. Choose one before rendering the full edit unless complete variants were requested. Do not automatically publish near-duplicates.
Review: Read the bundled rubric, judge prompt, reference requirements, and scorecard schema. Apply review requirements. Report missing evidence; never invent a score.
Caption or delivery: Read API delivery. Inspect the actual final video and CTA. Do not re-render an approved upload to match production defaults. Use the API for Metricool.
Results: Read review and learning. Record missing metrics as null. One post cannot establish a winning style.

Installation does not authorize publishing, new accounts, automations, purchases, or public releases. Existing explicit task authorization is sufficient; do not ask twice.

Deliver

Include artifacts relevant to the requested mode: source hash and transcript; claims and asset provenance; hook hypothesis and shot list; versioned master and SRT; cover, title, caption and CTA; review evidence; verified delivery receipt; and supported performance observations.

Keep credentials and private receipts outside the repository. Preserve raw media and previous renders. Report review candidate, scheduled, publishing, or published only with matching evidence.

Use the next-batch plan to turn these rules into a production cycle.

1	---
2	name: shortform-production
3	description: Produce readable vertical shorts with evidence-based B-roll, three opening treatments, editorial review, and authorized Metricool API delivery. Use for shortform edits, captions, hook experiments, or scheduling approved clips.
4	---
5
6	# Shortform Production
7
8	Keep the visual style consistent. Change the storytelling to suit the source. An experiment is a hypothesis, not a proven retention gain.
9
10	## Preamble
11
12	When available, use the repository's version check and telemetry initializer:
13
14	```bash
15	python3 telemetry/version_check.py 2>/dev/null \|\| true
16	python3 telemetry/telemetry_init.py 2>/dev/null \|\| true
17	```
18
19	Remote telemetry requires opt-in. Never log content, paths, account details, or credentials.
20
21	## Choose the work
22
23	1. Edit: Read [V5 style](references/v5-style.md) and [creative formats](references/creative-formats.md). Use the named source and timestamped transcript. Preserve credentials, claim qualifiers, natural speech, and the CTA. Inventory real assets before designing inserts. Use an available renderer; this package supplies editorial rules and delivery tools.
24	2. Experiment: Produce three opening treatments for one clip. Choose one before rendering the full edit unless complete variants were requested. Do not automatically publish near-duplicates.
25	3. Review: Read the bundled [rubric](references/eval/rubric.json), [judge prompt](references/eval/judge-prompt.md), [reference requirements](references/eval/references.json), and [scorecard schema](references/eval/scorecard.schema.json). Apply [review requirements](references/review-and-learning.md). Report missing evidence; never invent a score.
26	4. Caption or delivery: Read [API delivery](references/metricool.md). Inspect the actual final video and CTA. Do not re-render an approved upload to match production defaults. Use the API for Metricool.
27	5. Results: Read [review and learning](references/review-and-learning.md). Record missing metrics as null. One post cannot establish a winning style.
28
29	Installation does not authorize publishing, new accounts, automations, purchases, or public releases. Existing explicit task authorization is sufficient; do not ask twice.
30
31	## Deliver
32
33	Include artifacts relevant to the requested mode: source hash and transcript; claims and asset provenance; hook hypothesis and shot list; versioned master and SRT; cover, title, caption and CTA; review evidence; verified delivery receipt; and supported performance observations.
34
35	Keep credentials and private receipts outside the repository. Preserve raw media and previous renders. Report `review candidate`, `scheduled`, `publishing`, or `published` only with matching evidence.
36
37	Use [the next-batch plan](references/next-batch.md) to turn these rules into a production cycle.
38