/brag: a Claude Code skill that renders a launch video from the project you just finished

What /brag actually automates

Most launch videos fail at the story stage, not the render stage. Someone has a working repository and no idea which three moments are worth showing. /brag takes that decision away from you: the skill owns the product angle, the tone, and which moments to show, then hands a focused brief to Hyperframes, which builds, times, and renders the video. The README frames the split plainly, with the skill on the story side and Hyperframes on the build side.

The audience is narrow and specific. This is for someone who just pushed a side project and wants a short, shareable clip for a launch post. It is not for a marketing team with a brand kit, and it is not for anyone who needs to cut footage they already shot. The whole premise is that the input is a repository, not a media library. The README's own launch site says the looping video there was made by /brag on this repository, which is the only worked example the project offers.

The story brief and the render handoff

The architecture is a two-stage pipeline with a file-based boundary. Stage one is the skill running inside your agent: it reads the project, decides on an angle and tone, and writes a plan plus a composition brief. Stage two is Hyperframes, invoked through its CLI, which turns that brief into timed motion and a rendered file. The README describes the handoff as a focused brief, which is the important detail: the skill does not drive the renderer frame by frame, it constrains what the renderer is asked to build.

Everything lands in a brag-output/ folder containing the plan, the composition brief, share copy, and the rendered brag.mp4. That folder layout matters more than it first appears. Because the plan and brief are written to disk before rendering, you can read them and see what the skill decided before you commit to the output. It also means the intermediate artifacts are the natural place to intervene if the angle is wrong. The README does not document a way to resume a render from an edited brief, so treat the folder as an inspection surface rather than a documented re-entry point.

Tone is a first-class input. The README gives the example of steering it toward a fake Series A launch from 2016, which tells you the skill is designed to accept a stylistic frame rather than a shot list. Voiceover is the other explicit switch, off by default and enabled with a flag. When enabled, narration uses Kokoro through Hyperframes.

Installing the skill and running a first brag

There are three install paths, and which one you take depends on your agent. In Claude Code, add the marketplace and install the plugin. The README shows exactly these two lines:

/plugin marketplace add latent-spaces/brag
/plugin install brag@brag

After that, running /brag inside any project is the entry point. For other agents (Cursor, Codex, Copilot, Gemini CLI, opencode, and others), the README points at the skills CLI instead:

npx skills add https://github.com/latent-spaces/brag --skill brag

Adding -g installs globally so the skill is available in every project; leaving it off scopes the install to the current one. If you have no installer at all, the README's fallback is a direct copy into the Claude skills directory:

rsync -a --exclude '.DS_Store' skills/brag/ ~/.claude/skills/brag/

Restart Claude Code after that copy. Before the first run, confirm the runtime requirements hold, because a missing FFmpeg or an old Node will surface as a render failure rather than a clear install error:

npx hyperframes doctor

The README lists Node.js 22+, FFmpeg on PATH, and the Hyperframes CLI as requirements. Once doctor passes, ask the agent to run the skill, optionally with a tone, and optionally with voiceover:

/brag --tone "fake Series A launch from 2016"
/brag --voice

What you should see afterwards is a brag-output/ directory holding the plan, the composition brief, the share copy, and brag.mp4. Read the plan and the brief before you watch the video; that is where a wrong angle is cheapest to catch.

Where /brag breaks down

The dependency chain is the first real constraint. You need an agent that supports Agent Skills, Node.js 22 or newer, FFmpeg on PATH, and the Hyperframes CLI reachable through npx. Four prerequisites for a tool whose selling point is one command. If any link is missing, the failure appears late, at render time, after the skill has already done its planning work.

Windows is the sharpest edge. The repository exposes the skill to each agent through symlinks, and the README warns that Git on Windows needs git config core.symlinks true, or a clone with core.symlinks=true, plus Developer Mode or Administrator privileges to create them. The stated fallback is to copy skills/brag/ into the agent's skill directory by hand. That is a manual workaround for what is otherwise an automatic discovery mechanism, and it is the kind of detail that quietly disqualifies a tool for a team standardized on Windows machines.

There is also a control ceiling. Because the skill owns the angle and Hyperframes owns timing and rendering, the output is a short launch clip shaped by whatever the skill inferred from your repository. If you want a specific shot at a specific second, this is the wrong tool. The README does not describe a timeline editor, keyframe control, or a way to pin a particular visual to a particular beat. The tone flag steers mood; it does not give you a shot list. And with no releases retrieved, there is no versioned artifact to pin, so you are tracking the main branch.

How this differs from a template-based video tool

The obvious alternative is a template editor such as Canva or CapCut: pick a launch template, drop in screenshots, type a headline, export. The difference in approach is where the intelligence sits. A template tool gives you a fixed composition and asks you to fill the slots; you supply the judgement about what matters. /brag inverts that. It reads the project, decides which moments to show, and generates the brief itself, so the composition is derived from the repository rather than chosen from a gallery.

That inversion has a cost. A template tool gives you deterministic, repeatable output: the same slots, the same timing, the same export every time. /brag produces a plan and a brief that you should read before trusting the render, because the creative decisions were made by an agent. You trade predictability for not having to make the creative calls yourself. If your project has a strong visual identity you want preserved exactly, the template route is more faithful. If your problem is that you have no idea what to show, /brag addresses the actual bottleneck.

Maintenance, licensing and what you are tracking

The repository is not archived, and the last push was on 2026-08-17, roughly a month before this writing. That is recent enough that the project is moving, but with no releases retrieved there is no tagged version to depend on. You install from the marketplace or from the repository URL, which means you are following main. For a skill that shells out to an external renderer, that is worth knowing: a change in the skill and a change in the Hyperframes CLI can land independently.

The project is MIT licensed, which is permissive and places few conditions on reuse. The licence covers the repository; it does not automatically cover everything the tool pulls in. The credits section names the bundled music from ende.app, sound effects from Kenney, video generation through Hyperframes, and the fake demo sites built with Impeccable. If you plan to ship the rendered output commercially, check the terms attached to those assets and to the rendering service separately. That is a practical check, not a legal opinion.

The bundled music and sound effects are the part most likely to date. A fixed library of tracks means launch videos made with /brag will tend to sound alike, and swapping in your own audio is not described in the README.

What the examples directory tells you about intent

The repository ships five fake product sites under examples/: bicycles-for-snakes, fish-flight-school, horse-tinder, psychologists-for-chatbots, and taxi-for-taxis. The README calls them a benchmark suite. That is a meaningful design choice. Absurd fictional products are a way to test whether the skill can find a story angle in a repository that has no real market, no real users, and no obvious hook. If it can make horse-tinder look launchable, the reasoning is doing work rather than pattern-matching on a known brand.

For a prospective adopter, the examples double as a way to judge output quality before pointing the tool at your own work. You can read what the skill produced for a fictional project and decide whether that level of invention is what you want applied to something you actually care about. The README does not publish evaluation criteria or pass/fail thresholds for the suite, so it functions as a demonstration set rather than a scored benchmark. That is a gap worth naming: a benchmark suite without stated criteria is really a gallery.

Editorial conclusion

Adopt /brag if you ship small projects alone and want a launch clip without opening a video editor: install the plugin, run /brag --voice only when you want Kokoro narration, and inspect the plan and composition brief that land in brag-output/ before you share the render. Skip it if you need frame-level control, a long-form edit, or a pipeline that runs without an agent in the loop. Verify first that Node.js 22+, FFmpeg on PATH, and npx hyperframes doctor all succeed on your machine, and on Windows confirm git config core.symlinks true before trusting the discovery symlinks.

Frequently asked questions

Official sources

1. Issues
2. latent-spaces/brag on GitHub
3. License: MIT
4. README