Open Edits Update 2026: AI Video Editing Harness for Claude Code

If you edit video with Claude Code, the open-source Claude video editor is worth your afternoon. Open Video Edit is an open-source set of Claude Code skills that turns a raw talking-head clip into an animated short: captions, logos, faces, GIFs, b-roll cards, counters. It renders locally with Python and ffmpeg. Nothing uploads to somebody else’s render farm. The skills include one that reverse-engineers an editing style from reference footage, so you can point the agent at a look you like and get a scaffold back.

raw clip (talking head)
        |
        v
 Claude Code + Open Video Edit skills
        |
        +-- split-animated-talking-head
        +-- longform-animated-talking-head
        +-- fireship-style-edit
        +-- fireship-style-short
        +-- split-animated-short
        +-- reverse-engineer
        |
        v
 Python + ffmpeg render (local)
        |
        v
 finished vertical or long-form file
        |
        v
 Creator OS: schedule to Reels, Shorts, TikTok

What Open Video Edit Is and Why It Matters for Creators

Open Video Edit lives at github.com/kevinbadi/open-edits. It is not a subscription editor. It is a harness: a set of skills your Claude Code agent loads and runs on your machine. You bring the footage, the agent brings the timeline.

Why it matters: most editing tools ask you to learn their interface. This one asks you to name a style. You say “Fireship style,” or “the way my competitor cuts their shorts,” and the agent assembles the edit from skills. The output is a normal video file. No proprietary project format, no lock-in.

All six skills ship in the repo today:

• split-animated-talking-head
• longform-animated-talking-head
• fireship-style-edit
• fireship-style-short
• split-animated-short
• reverse-engineer

Kevin walks through the harness on the KevBuildsApps YouTube channel, and the launch walkthrough is at the Open Video Edit launch video.

Reverse-Engineer Any Editing Style Into a Reusable Skill

This is the one people will copy first. The reverse-engineer skill takes reference footage and produces a scaffold: a first draft of a new skill that mimics that style on your own clips.

The workflow, roughly:

1. Drop the reference clip or clips somewhere the agent can read them.
2. Ask Claude Code to run the reverse-engineer skill on that footage.
3. Read the scaffold it writes. It is plain files, so you can edit it.
4. Run the new skill on your own raw footage and compare.
5. Adjust the scaffold until the cuts, captions and pacing land where you want them.

What this replaces: watching forty minutes of someone else’s video with a notebook, then trying to hand-code the timing. The scaffold gives you a starting structure you refine instead of a blank page. It is the difference between “make it punchy” and a file that says where the punch lands.

One caution. A scaffold is a first iteration, not a finished house style. Expect two or three rounds. The skill name says it: reverse-engineer, then build forward.

fireship-style-short: A Vertical Short Editor From the Same Harness

Vertical is not a crop. A horizontal edit shrunk into 9:16 leaves you with dead space and captions that fight the platform UI.

fireship-style-short is built for the vertical frame from the start. That matters if your distribution is Reels, Shorts and TikTok, which is exactly where most creators get their reach. If cross-posting those three formats is part of your week, the mechanics of doing it without re-uploading three times are covered in cross-post Reels, TikTok and Shorts.

Pair it with the long-form version and you get one recording session feeding two outputs: the full conversation, and the cut-down vertical. Record once, edit twice, publish in three places. Because the render happens locally with Python and ffmpeg, long renders cost you time and electricity, not credits.

fireship-style-edit: The General Purpose Style Skill

fireship-style-edit is the horizontal, long-form counterpart. It is the skill you reach for when the clip is a full video rather than a vertical cut, and it shares the same rendering setup as the rest of the harness.

The practical split looks like this:

You do not have to pick one and stay there. Run the long-form skill on the full recording, then run a short skill on the same source file. Two outputs, one camera session.

How This Complements the Open-Source Social Agents

Open Video Edit is one half. The other half is the agent that runs your accounts after the file is done. That is Social Agents, at github.com/kevinbadi/social-agents.

Start it with one command:

npm start creatoros social-agents

It interviews you about your brand first, then posts, replies to comments and DMs, runs automations, and reports back on a local dashboard. Bring your own model: a logged-in Claude Code session, an ANTHROPIC_API_KEY, or any Anthropic-compatible API.

The split is clean. Open Video Edit makes the file. Social Agents moves it. There is a longer write-up on the whole approach in social agents, open source, and the background on the editing half is in the open-source Claude video editor post. If you run a channel network rather than a single channel, managing multiple YouTube channels covers the account side.

From Edit to Scheduled Post: Wiring Open Edits Into Creator OS

Here is a worked example. You recorded a 22 minute interview. You want one long upload, one Short, and a TikTok cut, all scheduled for the same week, without opening a browser nine times.

Step one: edit twice. Run the long-form skill on the full recording, then run fireship-style-short on the same source for the vertical cut.

Step two: install the Creator OS side. The CLI is @creatoros/cli, and the skills come with it:

npx @creatoros/cli@latest init

That installs 14 Claude Code skills, including post-longform, post-shortform and schedule-posts. Later, creatoros sync updates them without overwriting files you have edited. Your edits come back as .new files instead of disappearing.

Step three: check the caption before you send it. Platform limits differ and finding out at publish time is expensive:

creatoros validate:post-length --text "Your caption here"

Step four: create the posts. One command per post, targeting the accounts you want:

creatoros posts:create \
  --text "Full interview is live" \
  --platforms youtube,linkedin \
  --media med_abc123 \
  --title "The 22 minute cut" \
  --scheduledAt 2026-10-06T09:00:00Z \
  --timezone America/Toronto

Step five: send the vertical to the short-form networks. Same shape, different platforms, and a cover pulled from a specific frame:

creatoros posts:create \
  --text "60 seconds on why this matters" \
  --platforms instagram,tiktok,youtube \
  --media med_abc123 \
  --coverAt 4200

That single med_ id is worth understanding. Upload the file once with creatoros media:upload and the returned id is reusable on every platform in the post. No re-upload per network. If you want the full calendar pattern for a month at a time, scheduling a month of content walks through it.

Step six: check the timing. Before you commit to 9am, ask:

creatoros analytics:best-time --platform instagram

Then, once things are live, creatoros analytics:posts --sortBy engagement tells you which cut actually earned attention, not which one you liked making.

The agent version of the same workflow

If you would rather not type commands, connect Creator OS to Claude as a custom connector. Go to Settings, then Connectors, then Add custom connector, paste https://mcp.creatoros.ca/mcp, sign in, pick the workspace, and choose read only or read and write. Read-only connections only see read tools, and the API refuses writes from them.

Then you can say something like: “Take the file at this path, upload it, write a Shorts caption under 100 characters, and schedule it for Thursday morning.” The agent calls upload_media_from_url, check_caption_length, then create_post. Destructive tools like delete post and disconnect account are marked so your AI app asks first. Setup for other editors is in the MCP docs, and the Claude-specific page is at creatoros.ca/claude.

Two more CLI pieces worth knowing about after an edit lands. creatoros posts:retry <postId> re-pushes a post that failed on one network, leaving the successful ones alone. And if you publish and immediately spot a typo on a network that supports editing, creatoros posts:edit <postId> --platform twitter --text "fixed" fixes it in place.

Get Started With the Open Video Edit Harness

Two installs, one afternoon.

1. Clone the harness: github.com/kevinbadi/open-edits. Run the reverse-engineer skill on a reference clip, or jump straight to fireship-style-short if you just need a vertical cut today.
2. Install the publishing side: npx @creatoros/cli@latest init.

The Creator plan is $19.99/month or $59.99/year, and includes up to 8 connected accounts: 7 socials plus Skool. API keys, the MCP server, the CLI and the agent skills are included on every plan, so you are not paying extra to let an agent do the work.

Sign up at creatoros.ca/sign-up, then read the full command reference at the CLI docs. If you would rather watch the harness build a real edit end to end, the Open Video Edit launch video and the rest of the walkthroughs on KevBuildsApps are the fastest way in.