> ## Documentation Index
> Fetch the complete documentation index at: https://www.tella.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent skills guide

> Install Tella's open-source agent skills and use ready-made prompts to cut, clip, zoom, source B-roll, and publish your videos with an AI agent.

Skills are ready-made workflows that teach your AI agent how to do a specific video editing job — cutting silences, clipping for social, adding zooms, sourcing B-roll — so you get a consistent result without writing a long prompt every time.

Tella's skills are open source and free. Install one, then ask for what you want in plain language.

<Card title="Browse the Skills directory" icon="wand-magic-sparkles" href="https://www.tella.com/skills" horizontal>
  See every skill, what it does, and its install command.
</Card>

## How skills and the MCP fit together

These are two different things, and most people want both:

* **Skills** work on **video files on your computer**. They run media tools like `ffmpeg` and Whisper locally. Depending on the skill and agent, selected images or video frames may be sent to your AI model provider for analysis.
* The [**Tella MCP server**](/docs/help/integrations/mcp-server) connects your agent to **your Tella workspace** — the videos you've already recorded — so it can cut, lay out, and zoom them in Tella itself.

You don't need the MCP to use a skill, and you don't need a skill to use the MCP.

## What you need

* An AI agent that supports skills. Most skills below are built and tested for **Claude Code**; several also run on Claude Cowork, Codex, or Cursor.
* The tools listed by each skill, commonly `ffmpeg` and `python3`. Platform support varies: Cut Video and Clipify lean on Apple's hardware video decoding and need a Mac. B-Roll Finder can fall back to CPU transcription, while Add Zooms, Tella to YouTube, and Slack Assets Bot aren't tied to one platform.
* Each skill's repository has its exact requirements and setup steps.

## Install a skill

Most skills install by cloning them into your agent's skills folder. For Claude Code:

```bash theme={null}
git clone https://github.com/louisedesadeleer/cut-video.git ~/.claude/skills/cut-video
```

Restart your agent, and `/cut-video` is available as a slash command. Swap the repository URL and folder name for any other skill.

<Tip>
  You don't have to memorize the slash command. Once a skill is installed, asking in plain language — "tighten up this recording" — is usually enough for the agent to pick the right one.
</Tip>

<Note>
  A few skills install differently — Tella to YouTube is a Claude Code plugin, and Slack Assets Bot needs a one-time Slack app setup. The exact command for each skill is on its page in the [Skills directory](https://www.tella.com/skills).
</Note>

## Cut Video

<Card title="Install Cut Video" icon="github" href="https://www.tella.com/skills/cut-video" horizontal>
  Install command, features, and source code.
</Card>

Tightens a long recording. It removes silences, filler words, false starts, and repeated takes, while deliberately keeping laughs and comedic pauses. A typical talking-head recording comes out 40–60% shorter.

It's more accurate than cutting from a transcript alone: it uses forced alignment to find word boundaries to within \~10–20ms, so cuts land between words instead of clipping them.

#### ⚡ Prompt it now

* *Cut the silences and ums out of this recording*
* *Tighten this video — aggressive, but keep the laughs*

<Tip>
  Pick a tone when you ask. `aggressive` is the default and matches fast social pacing; ask for `documentary` when long pauses are doing real work.
</Tip>

## Clipify

<Card title="Install Clipify" icon="github" href="https://www.tella.com/skills/clipify" horizontal>
  Install command, features, and source code.
</Card>

Turns a long video into social-ready clips. It scans the transcript for punchlines, reversals, and audio peaks, proposes candidate clips with timestamps, reframes 16:9 → 9:16, and burns word-by-word captions.

For two-person footage it can follow whoever is speaking with hard-cut pans, or stack both faces in a split screen.

#### ⚡ Prompt it now

* *Find the best moments in this podcast and clip them for TikTok*
* *Cut this into 9:16 clips with captions*

<Tip>
  If you want a particular caption look, paste a screenshot of one you like and ask it to match — it will copy the font, size, weight, color, and position.
</Tip>

## Add Zooms

<Card title="Install Add Zooms" icon="github" href="https://www.tella.com/skills/add-zooms" horizontal>
  Install command, features, and source code.
</Card>

Adds emphasis motion to a flat talking-head clip: hard-cut punch zooms on emphasised words, a ratcheting zoom that steps up through a list, and slow Ken Burns pushes on your key line. It's non-destructive and writes a new file.

#### ⚡ Prompt it now

* *Add punch zooms to this clip*
* *Add some energy to this hot-take — zoom on the emphasised words*

<Tip>
  Zooms land best when they're rationed. Long stretches at normal scale are what make each punch register, so resist asking for one on every line.
</Tip>

## B-Roll Finder

<Card title="Install B-Roll Finder" icon="github" href="https://www.tella.com/skills/b-roll-finder" horizontal>
  Install command, features, and source code.
</Card>

Sources B-roll for a talking-head video and places each cutaway on the exact word. It reads your transcript, works out what each line is about, searches trusted sources, and hands back a short list of vetted candidates.

It proposes; you pick. The final choice is a taste call, and the skill is built to narrow the options rather than decide for you.

#### ⚡ Prompt it now

* *Find B-roll for this video*
* *Source cutaways for the intro and show me the options*

<Tip>
  It works best on scripts full of proper nouns and checkable claims — products, companies, events. Abstract lines have no obviously correct footage, so it will defer to you.
</Tip>

## Slack Assets Bot

<Card title="Install Slack Assets Bot" icon="github" href="https://www.tella.com/skills/slack-assets-bot" horizontal>
  Install command, features, and source code.
</Card>

Keeps your asset library within reach. Drop files, YouTube links, or image URLs into a Slack DM and they land sorted into folders on your machine, renamed so you and your agent can find them later.

#### ⚡ Prompt it now

* *Use the reaction meme I saved yesterday in this edit*
* *Add the sound effect from my sounds folder*

<Tip>
  Screenshots with names like `CleanShot 2026-01-04` get a descriptive name automatically, so you can search your library by what's actually in the image.
</Tip>

## Tella to YouTube

<Card title="Install Tella to YouTube" icon="github" href="https://www.tella.com/skills/tella-to-youtube" horizontal>
  Install command, features, and source code.
</Card>

Takes your latest Tella recording, drafts a title and description from the transcript and chapters, and uploads it to YouTube in 4K.

#### ⚡ Prompt it now

* *Publish my latest Tella recording to YouTube*
* *Upload that demo to YouTube — draft a title from the transcript*

<Warning>
  The generated metadata defaults to a **public** upload. Ask for **unlisted** or **private** first if you want to review the video before it goes live.
</Warning>

See the [full setup guide](/docs/help/integrations/tella-to-youtube-agent-skill) for connecting your Tella API key and Google account.

## Edit your Tella videos by prompt

If you've connected the [Tella MCP server](/docs/help/integrations/mcp-server), you can edit videos already in your workspace by asking — no skill needed. Your agent can cut clips, remove silences and fillers, switch layouts, add zooms, highlights, and blurs, then export.

#### ⚡ Prompt it now

* **Clean up a recording:** *Take my latest Tella video, remove the filler words, and cut any silence longer than a second.*
* **Cut by transcript:** *Find where I fumble the intro in this video and cut it — keep the second take.*
* **Layouts:** *Switch to camera-only fullscreen when I'm talking to the viewer, and side-by-side when I'm demoing. Use a spring transition.*
* **Zooms:** *Add zooms on the parts where I'm clicking through the settings panel — nothing on the talking-head sections. Preserve any existing hand-tuned zooms.*
* **Redact:** *Blur the email address that's visible in the top-right while the dashboard is on screen.*
* **A social cut:** *Duplicate this video, keep only my best reactions, and export it as a 20-second clip with captions on.*

<Tip>
  Ask the agent to show you a thumbnail at a given moment to check its work before you export — it renders the frame with layouts and zooms applied.
</Tip>

<Note>
  MCP edits remain adjustable in the Tella editor, but some operations replace matching edit state. Automatic zoom generation replaces existing tracking zooms by default. Ask the agent to preserve existing zooms (using `replaceExisting: false`) or duplicate the video before experimenting.
</Note>

## Tips for best results

* **Be specific about the moment.** "Cut the bit where I lose my train of thought in the intro" beats "clean this up".
* **Say what to protect.** Skills cut hard by default. If a pause is a joke landing, say so.
* **Review before you publish.** Ask for the plan or a preview frame first, especially on a long recording.
* **Iterate in the same conversation.** "Too aggressive, keep more of the pauses" is faster than starting over, since the transcript and proxy are already there.
* **Combine skills.** A common chain is Cut Video → Add Zooms → Clipify: tighten the recording, give it motion, then cut it for social.

## FAQ

<AccordionGroup>
  <Accordion title="Do I need a paid Tella plan to use skills?">
    No. The skills are open source and run on your own machine — they work on any video file. You only need a Tella account for the skills and prompts that touch your Tella workspace, like Tella to YouTube or editing through the MCP.
  </Accordion>

  <Accordion title="Does my footage get uploaded anywhere?">
    Core transcription and rendering happen on your machine. Depending on the skill and agent, selected images or verification frames may be sent to your AI model provider for analysis. Skills that publish or source footage — Tella to YouTube, B-Roll Finder — also reach the internet by design.
  </Accordion>

  <Accordion title="Can I use these outside Claude Code?">
    Sometimes. A skill is a `SKILL.md` file plus a few scripts, so any agent that can read instructions and run shell commands can use one — Claude Cowork, Codex, and Cursor generally work. The install steps and slash commands in each repository assume Claude Code.
  </Accordion>

  <Accordion title="Can I change how a skill behaves?">
    Yes. Each skill's `SKILL.md` holds its defaults in plain English — pacing, zoom strength, which sources to trust. Fork the repository and edit it to match your own style.
  </Accordion>

  <Accordion title="Something went wrong or a skill isn't installing.">
    Check the skill's GitHub repository first, since setup instructions there are always current. For anything Tella-specific, contact us at [support@tella.com](mailto:support@tella.com).
  </Accordion>
</AccordionGroup>