voice-skill-template
Skill0 starsChanged 3 months agoMITView on GitHub
TEMPLATE — fill this in for your own account, then rename it. Governs the actual words in short-form content: scripts, hooks, captions. Triggers on any content-writing request: writing or evaluating scripts, generating hooks, writing captions, repurposing clips, briefing editors. The scheduling skills (get-it-out-there, trial-reels) defer to this one for wording, so this must exist before captions get written.
In this repository3 other files it gives its agents.
- get-it-out-thereSkill
- transcribe-anythingSkill
- trial-reelsSkill
The file
Files
Voice skill — TEMPLATE
This is a blank. It does nothing useful until you fill it in.
get-it-out-there and trial-reels handle mechanics: platform limits, media
verification, scheduling, approval gates. They deliberately do not decide how
your content sounds. They look for a skill like this one and defer to it.
That split is the point. Mechanics are universal; voice is not. Keeping them apart means you can share the mechanics without publishing your brand voice.
Fill in the sections below, delete the guidance in > blockquotes, rename the
folder and the name: field to something account-specific, and the scheduling
skills will pick it up.
Reference fetch rules
If your voice guidelines live outside the skill — Notion, Drive, a wiki — list them here as a table of task → what to fetch first. Be explicit about which fetches are mandatory, because the failure mode is writing captions from a transcript alone and producing something generically competent and off-voice.
Keep the list short. A skill that fetches six documents before every task gets disabled by its own user.
Mandatory before the task, every time
| Task | Fetch first | Where |
| Writing captions | Voice reference | <url> |
| Writing a script | Voice reference | <url> |
| Generating hooks | Hook library | <url> |
Conditional — only when relevant
| Fetch when… | Resource | Where |
Who this account talks to
The audience, in concrete terms. Not demographics — the specific problem they have and the words they use for it. If you have real phrases from real customers (support tickets, sales calls, comments), collect them here verbatim. Borrowed language outperforms invented language every time.
What this account is about
The 3–5 recurring themes. Every piece of content should be identifiably one of them. If a script doesn't map to one, that's a signal to question the script.
Voice
How this person actually sounds. Be specific and testable, not aspirational — "warm and authentic" is unusable. Things worth pinning down:
- Sentence length and rhythm
- Capitalisation conventions
- Punctuation bans (many creators ban em dashes, semicolons, or exclamation marks)
- Words and phrases that are theirs
- Words and phrases that are forbidden
- First person or second person
- Whether jokes are on-brand
Video structure
The shape of a script: hook, body, payoff, CTA. Give target durations per section, and say what happens in the first few seconds.
Hook guidelines
Hooks are where most short-form content fails, so this section earns its length.
Document: the hook types that work for this account, with real examples; rules for on-screen text hooks (character limits, how they relate to the spoken hook); and an explicit list of what to avoid — the clichés, the fake urgency, the patterns this account has decided are off-limits.
What to avoid
A "never do this" list is often more useful than a "do this" list, because it stops regression toward generic AI phrasing.
Caption formats by platform
Per-platform structure. Note that
get-it-out-therealready specifies character limits and CTA placement mechanically — this section is about wording within those constraints. Don't duplicate the mechanics; if the two ever disagree, say here which wins.
| Platform | Notes on wording |
| X / Twitter | |
| Threads | |
| Instagram Reels | |
| Facebook Reels | |
| TikTok | |
| YouTube Shorts |
Capitalisation and emphasis
Applies across platforms. Specify how emphasis is done — caps, italics, line breaks — and where the exceptions are.
Hashtag rules
Whether to research fresh each time or use a fixed bank, and how many per platform. If research is required every time, say so explicitly, or it will get skipped.
Current CTAs
The live calls to action and what each one actually delivers. Keep this current — a CTA promising something that no longer exists is worse than no CTA. If you use comment/DM trigger words, this is where the live list belongs, and the scheduling skills will check against it for collisions.
Trigger commands
Optional. Short commands for common requests, so you can type
hooksinstead of a paragraph. Define what each one outputs and how many variants.
| Command | Does |
hooks | |
captions | |
check this |
Script evaluation
The rubric for judging a draft. Useful for pasting in a script and getting a verdict rather than a rewrite. Say whether the output is a score, a list of problems, or a fix.
Hard rules
The non-negotiables. Keep this list genuinely short — five rules get followed, twenty get averaged. Anything here should be something you'd reject a whole draft over.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

