The cli-explainer Skill | Claude Code on Screen
Liam, in for Bear, tears down the cli-explainer skill file to show how a single markdown document, not a recording, directs Claude to write and structure a CLI explainer video.
Liam, in for Bear, opens the cli-explainer skill file to show what actually happens when someone types "CLI" at the Claude composer. The point of the video is a single distinction: cli-explainer does not record a terminal session, it commissions a reel. Claude reads the skill file and writes the video from it.
What a skill folder actually contains
A skill in this toolkit is just a folder. For cli-explainer, that means two files: skills/make/cli-explainer/SKILL.md, roughly 365 lines of plain markdown, and a reference directory holding one example beat sheet. There's no compiled binary and no background service. Claude reads the file, and the file is what writes the reel. The file isn't a loose suggestion either, it fixes the shape of every CLI reel to a nine-step spine: cold open, problem, ask, code, output, revision, code again, better output, summary, handoff, outro. Every reel follows the same sequence.
The trigger and the revision law
The entry point is a verbatim trigger from the skill file's own trigger table: "CLI" followed by the concept, no model version, no temperature setting. That trigger tells Claude which skill to load; the skin defaults to Claude and the persona defaults to Liam. The skill also names its own mandatory beats for the 16x9 cut: the problem beat, at least one revision cycle, and a summary with next steps before the outro. A CLI video without a check-in-change moment is treated as incomplete, a rule written directly into the skill file rather than left to a reviewer's judgment. The one exception is the shorts cut, which ships a single revision cycle to fit its three-minute cap and points viewers to the 16x9 video for the full example.
Swapping the interface skin
The skill separates the interface skin, Claude or Codex, GitHub, shell, from the narrative spine underneath it. Choosing "CLI Codex" instead of "CLI Claude" renders the same spine through the Codex composer instead of Claude's; the body logic and bookends don't change. The swap is described in the file as a one-to-one rename: "Claude composer ask" becomes "Codex composer ask," "Claude code beat" becomes "Codex code beat," and so on. Props like the greeting, the command, and the title carry over unchanged. It's a substitution, not a reauthoring, which is why the same lesson can teach through a different tool without the underlying beat sheet being rewritten.
The verdict
Liam's summary is direct: cli-explainer is instructions, not software. It reads like an editorial style guide because that's what it is, and it doesn't record anyone's terminal, it commissions a real explanation of what was actually asked of Claude and what came back. The closing argument is that reading the skill file beats watching a tutorial about it.
Key takeaways
- cli-explainer is a markdown skill file that directs Claude to write and structure a video, not a terminal recording tool.
- Every CLI reel follows the same nine-step spine: cold open, problem, ask, code, output, revision, code again, better output, summary, handoff, outro.
- The 16x9 cut must include at least one revision cycle; the shorts cut is the sole exception, with a single cycle.
- Swapping tools, Claude to Codex, is a one-to-one rename of the interface skin, not a rewrite of the narrative spine.
- The trigger phrase is verbatim from the skill's own trigger table: "CLI" plus the concept, nothing else.
Who this is for
This is for anyone using Claude Code who wants to understand how the cli-explainer skill turns a real coding session into a structured explainer video, and for skill authors curious how a single markdown file can lock a video's structure without any custom software.
Full transcript(auto-generated, with timestamps)
[0:00]Namast, this is Liam in for bear. I asked Claude what happens when I type CLI at the composer. Not because I don't know, because half the room has a Claude subscription and no idea a skill by that name exists. Here is the whole video in one breath. CLI-Explainer does not record a terminal. It commisss a reel. Claude reads the file and writes the video. 12 minutes from now. That distinction is the only sentence I want you to keep. A skill in this toolkit is a folder. Two files sit under skills/make/clyexplainer skill.md 365 lines of plain markdown and a reference directory with one example beat sheet that is the whole skill. No
[0:38]Compiled binary, no demon, no service. Claude reads the file and the file writes the reel. The file is not a suggestion. It fixes the shape of every CLI reel to a ninestep spine. Cold open problem ask code output revision code again. Better output summary handoff outro. Same in every reel. The creative work happens inside the frame. The frame itself is locked. The ask verbatim from the files own trigger table. CLI followed by the concept. No model version, no temperature. The trigger tells Claude which skill to load. The skin defaults to Claude. The persona defaults to Liam. This is the only entry point. This is the section that owns the
[1:16]Video you are watching. Mandatory in the 16x9 cut. One, the problem beat. Two, at least one revision cycle. Three, the summary and next steps before the outro. A CLI video without a check-in change is incomplete. That is a rule in the skill, not a note in a review. Read it in the files own words. The revision law. The 16 by9 video always contains at least one revision cycle. The one exception is the shorts cut. To fit the 3minut cap, it ships a single cycle and points the viewer to the 16x9 for the complete example. The doctrine names its own edge case in the same paragraph. The revision
[1:50]CLI the concept- tool codeex one flag. The same spine renders through the codeex composer instead of claws. The body logic and the bookends don't move. The interface skin does. Nothing in the beat sheet has to be rewritten. This is how the skin swap is written in the file. A onetoone rename. Claude composer ask becomes codeex composer ask. Claude code beat becomes codeex code beat. Claude title outro becomes codeex title outro. Props carry over. Same greeting, same command, same title. It is a substitution, not a reauthoring. Here is the design decision I keep coming back to. The skill separates the interface skin clawed codeex on GitHub shell from
[2:28]The narrative spine. Choose a new tool. The same lesson still teaches. That is the thing the file gets right. The spine survives the skin. The verdict cli-explainer is instructions not software. It reads like an editorial style guide because that is what it is. It doesn't record your terminal. It commisss a real about what you asked Claude for. Reading the file beats watching a tutorial. Your turn. Here is the prompt. Read it with me. Use CLI-E explainer to build a brutalist reel about the smallest command in my own workflow. Read skill/make/clexplainer/skill.mmd first, then propose a beat sheet before you draft narration. The bite is the second sentence. You are asking Claude
[3:07]To read the file, not summarize its name. Liam in for bear. The Cly explainer skill. Liam in for bear.
More from Brutalist (Film as Code)
3:49The anthropics Skill | Claude’s Official Skill Library
3:30Automating the YouTube Review Pipeline | Vikhyat
3:34claude liam brutalist skill deep explainer
3:07The ai-explainer Skill | How Brutalist Builds Explainers
3:34claude liam brutalist skill duration planner
4:13