1 FRONTMATTER
---
name: youtube-script
description: >
Write beat-by-beat YouTube video scripts with
hook selection, pacing control, and delivery
notes. Use this skill whenever the user wants
to write a YouTube script, plan a video...
---
2 PURPOSE STATEMENT
# YouTube Script Skill
Write high-retention YouTube scripts grounded
in patterns extracted from 1.4M+ views across
top-performing creators. This skill produces a
beat-by-beat script with delivery notes, not a
generic outline.
3 STEP-BY-STEP WORKFLOW
## Step 1: Gather Context
## Step 2: Select Hook Archetype
## Step 3: Write the Script
## Step 4: Apply Delivery Techniques
## Step 5: Review Against Anti-Patterns
4 REFERENCE POINTERS
Read `references/hook_archetypes.md` for
detailed examples and line-by-line breakdowns.
Read `references/pacing_models.md` for
detailed pacing guidance and transitions.
5 OUTPUT FORMAT
## Output Format
Deliver the script as a single markdown
document following this template:
# [Video Title]
## Video Description
## INTRO/HOOK
## MAIN CONTENT
## RESOLUTION & INSPIRATION
6 ANTI-PATTERN CHECKLIST
## Step 5: Review Against Anti-Patterns
| Anti-pattern | Fix |
| Opens with "Hey guys" | Jump to substance |
| Generic examples | Use real scenarios |
| Monotone energy | Vary by section |
| Mid-roll CTA | End only |
The Frontmatter: Name + Description
This is the trigger. Cowork reads the description to decide whether to use this skill. It never reads the rest of the file until it decides the skill is relevant.
Your description should be specific and slightly "pushy." Include the phrases people actually say when they need this skill: "write me a script," "plan a video," "YouTube script about X."
Pro tip: Use the YAML block scalar > for descriptions with colons or special characters. It prevents parsing errors when your skill is packaged.
The Purpose Statement
One paragraph that tells Cowork what this skill produces and what makes it different. This is not marketing copy. It is a calibration statement that shapes every decision Cowork makes while using the skill.
"Beat-by-beat script with delivery notes, not a generic outline" tells Cowork the expected quality bar. Without this, you get generic output.
Pro tip: Mention where the knowledge came from. "Patterns extracted from 1.4M+ views" gives Cowork confidence in the instructions. It treats them as tested, not speculative.
The Step-by-Step Workflow
This is the spine of your skill. Each step is a phase Cowork follows in sequence. It gathers inputs, makes choices, does the work, then reviews.
The key principle: write instructions, not a prompt. You are telling Cowork how to approach a type of task, not what to do once. The same skill runs for any video topic, any audience, any length.
Pro tip: Give Cowork decision points. "Present the user with a selection table. Recommend based on their topic, but let them choose." This creates interaction instead of a monologue.
Reference Pointers (Progressive Disclosure)
This is the architecture trick that keeps your skill under 500 lines. Detailed knowledge lives in separate files. The main SKILL.md just tells Cowork when to read them.
Cowork only loads a reference file when it reaches that instruction. So if the user picks Hook A, Cowork reads the hook archetypes file. If they pick Pacing P3, it reads the pacing file. No wasted context.
Pro tip: The rule: if guidance applies to every run, it stays in SKILL.md. If it applies to some runs, it goes in a reference file.
The Output Format
Define exactly what the final output looks like. Use a template with section headers so Cowork knows what to produce. Without this, output structure varies every time.
This is where your personal preferences live. Your template, your section names, your formatting choices. The skill ensures consistency across every run.
Pro tip: Include "Writing Notes for Each Section" after the template. These are micro-instructions that tell Cowork how to fill each section, not just what sections exist.
The Anti-Pattern Checklist
A table of common failure modes with their fixes. Cowork reviews its own output against this list before delivering. Think of it as a quality gate.
These come from experience. Every time you see Cowork produce something you don't like, add it to the anti-pattern list. The skill gets better every time you refine it.
Pro tip: Frame anti-patterns as "if this, then that" pairs. Cowork applies them more reliably than vague instructions like "make it better" or "be more engaging."
The Full Folder Structure
A skill is a folder. Here is what goes inside.
S
youtube-script/
skill folder
M
SKILL.md
required, under 500 lines
R
references/
detailed knowledge
H
hook_archetypes.md
6 hook styles, line-by-line
P
pacing_models.md
5 pacing patterns
When to Build a Skill
Four questions to decide whether your task deserves to become a skill.
1 Have you given Cowork this type of instruction more than twice?
If yes, you are repeating yourself. A skill captures those instructions once and runs them every time. If no, do it manually a few more times first. You need the reps to know what works.
2 Does the output need to follow a consistent format or quality bar?
If your proposals, scripts, reports, or emails need to look and sound the same every time, a skill enforces that. Without one, quality drifts between conversations.
3 Is there domain knowledge Cowork needs but does not have by default?
Hook archetypes, brand voice rules, pricing models, compliance checklists. If you find yourself re-explaining the same context, put it in a skill and a reference file.
4 Could this skill connect to another skill you already have?
The compound effect. A script skill feeds a slides skill. A research skill feeds a proposal skill. Skills that read each other's output become production pipelines.