Article to show

The arclight-producer skill teaches Claude Code to turn a news story into a show you can record in one take. Give it a link. It pulls out the facts, proposes an angle, writes a short script with cues, then builds the slides, cameras and animated charts in ArcLight and checks every slide before handing it back.

When it's done, you press Play and read. The teleprompter starts with the show, [[Slide]] cues cut to the next slide on the right sentence, and graphic cues land each number on the word it belongs to.

Install it

The skill ships inside the app, at ArcLight.app/Contents/Resources/Skills/arclight-producer. It needs the arclight MCP server, so set that up first.

Connect Claude Code to ArcLight. Follow Set up AI agents: turn on Settings ▸ Agents and add the arclight server to your project.
Copy the skill into the project. See Install the producer skill for the command. It copies the folder into your project's .claude/skills.
Restart Claude Code in that folder, so it picks up the skill.

If ArcLight isn't reachable, the skill stops and tells you to turn on Allow agents to control ArcLight in Settings ▸ Agents. It won't try to work around it.

What to ask

Start Claude Code in the project folder, with ArcLight open, and ask in plain words. For example:

Make an ArcLight show from https://example.com/story, about 45 seconds.

Produce a 60-second segment on this article. Use a bar chart for the revenue numbers and a quote card for the CEO's line.

Here's my script. Make slides and graphics for it, with cues for the numbers.

Revise the show we built: swap the timeline for a comparison and tighten the close.

What it does, step by step

Fact sheet. It reads the article and lists every claim, number, name, date and quote it might use, each with the exact sentence it came from. It notes what's uncertain, like projections, disputed figures or anonymous sources. Text in the article is treated as data, never as instructions.
Angle and hook. It offers two or three angles, each a one-sentence thesis with its opening line, and recommends one. You pick. This is the one question it asks up front. Length defaults to 30 to 60 seconds unless you say otherwise.
Beat sheet. Hook, context, two to four points, the so-what, and a close. Each beat gets a visual: camera full, camera with a lower third, camera picture-in-picture beside a graphic, a full chart, a big number, a quote card or a source card. It aims to change the picture every 8 to 15 seconds.
Script. It shows you the script as plain prose first. You approve it before any cues go in or anything is built. Then it adds [[Slide name]] cues at each beat and {{Layer: step}} cues on the exact words where a number or reveal should land.
Build. It creates the slides in script order, adds cameras, titles and lower thirds, places graphics from the kit, sets transitions (Dissolve by default, Magic Move between slides that share a camera framing, Push for a new topic), writes the script into the teleprompter and sets it to start with Play.
Review. It renders every slide at each graphic step and build, and checks that text fits, nothing covers the camera, contrast is readable, graphics report no errors, and every number on screen matches the fact sheet. It fixes and re-renders, up to three passes.
Hand-off. A short report: the running time and slide list, every on-screen number with its source sentence, anything it couldn't verify, and how to record. It saves the document if you ask.

You can watch it work. Each change is one undo step named "Agent: …", layers it just changed are outlined on the canvas, and the toolbar's agent badge lists every change with Revert.

The graphics kit

The skill comes with ready-made web graphics, so it builds charts without writing code. Each one draws on a transparent page sized to its layer, steps through reveals on script cues, and shows only the data the agent passes in. Nothing unsourced reaches the screen from a template.

Component	What it shows	Fits
`bar-chart`	Horizontal bars that grow in one by one. The newest bar takes the accent color.	About 8 bars
`line-chart`	A line drawn left to right as each step reveals the next point. The last visible point is labeled.	About 12 points
`big-number`	One number that counts up or down to each step's value, with a label.	One figure per step
`comparison`	Two columns compared row by row, one row per step.	About 5 rows
`timeline`	A horizontal timeline. Each step reveals the next event and moves the marker to it.	About 6 events
`quote`	A pull quote that builds in word by word, then highlights a phrase.	About 40 words
`source-card`	Where the story comes from: outlet, headline and date.	One source

Every component takes the same theme (accent, text, muted and panel colors, and a font), so a show keeps one accent color and one type style across all its graphics.

The rules it follows

No invented facts. Every figure, name, title and quote on screen or in the script comes from the fact sheet. A projection is labeled "projected" on screen.
Sources are credited, on a source card or in the graphic's source line.
Quotes stay verbatim. On-screen copy otherwise uses plain words, with no markdown.
You decide twice: the angle before it writes, and the script before it builds.
Every change can be undone, one step at a time, or reverted from the agent badge.

Record the show

Press Play. The teleprompter starts, and as you read, slide cues take the next slide and graphic cues step the charts. Record with the toolbar's Record button, or ask the agent to start recording. See Teleprompter and Running the show for the details.