Communitygithub.com

showcase

Video and animation skill. Turn real digital work (a website, web app, mobile app, automation or brand identity) into a narrated, cinematic Instagram Reel and Instagram cover with the Claude Showcase Kit. Ask whether the user wants an on-camera presenter or a voice-over-only Reel. In presenter mode, write a recording or avatar-generation prompt and wait for their footage before editing. Use for showcase videos, portfolio Reels, Instagram covers, case studies and promos of real projects.

showcase 是什么?

showcase is a Claude Code agent skill that video and animation skill. Turn real digital work (a website, web app, mobile app, automation or brand identity) into a narrated, cinematic Instagram Reel and Instagram cover with the Claude Showcase Kit. Ask whether the user wants an on-camera presenter or a voice-over-only Reel. In presenter mode, write a recording or avatar-generation prompt and wait for their footage before editing. Use for showcase videos, portfolio Reels, Instagram covers, case studies and promos of real projects.

兼容平台✓Claude Code~Codex CLI~Cursor
npx skills add https://github.com/mchine-ltda/claude-showcase-kit/tree/HEAD/skills/showcase

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

showcase 是做什么的?

Make a premium portfolio Reel of the work in the current repository with the Claude Showcase Kit engine. The engine is a separate application (its folder is in engine.json next to this file). It owns the template, capture, voice, music, cover, QA and the portfolio registry. This repository only provides the work. Never copy the engine into this repo. Never modify this repo (no edits, installs, builds, commits or generated files) unless the user explicitly authorises a specific change.

Arguments: $ARGUMENTS (a URL, the client's name, the kind of project or notes; may be empty).

Choose the presenter format

Before capture or production, ask “How would you like the Reel presented?” in the user’s language. Use AskUserQuestion if available, otherwise ask in chat. If the user already made an explicit choice, use it and continue.

  • Voice-over only (Recommended) — Use the existing generated narration and device walkthrough.
  • On-camera presenter — Use the person or avatar's recording for the narration. Show them in the opening and closing cards, with the real walkthrough and their voice in between.

Wait for the choice. For voice-over only, continue through the usual interview and workflow. For on-camera presenter, read PRESENTER.md and prepare the recording brief first. Use read-only inspection to ground the walkthrough and script. Deliver the copy-ready recording or generation prompt and wait for the footage before initialising the project, recording the walkthrough or rendering previews. If the user already supplied usable presenter footage, inspect it and confirm its spoken script against the walkthrough before proceeding.

The engine

Run every engine command through the bundled launcher, from any directory:

node "${CLAUDE_SKILL_DIR}/scripts/showcase.mjs" <command> [options]

Below, SC means exactly that prefix. Start with SC where, SC doctor and SC voices. If the engine is missing or has no dependencies, follow INSTALLATION.md and stop until it works. Relative paths you pass resolve against the directory you run it from. Projects live in the engine's projects/<slug>/, and they are private: never commit or upload them.

One idea for everything

Every kind of work is shown the same way: a real recorded session of the real thing, in the same cinematic frame, in the same order.

  1. Cover lockup, about 1.2 s. Its settled frame is the Instagram cover.
  2. Walkthrough: real scrolling, clicks, typing and page changes.
  3. Outro: the agency signature.

Narration, subtitles, ducked music and subtle sound effects run throughout. The kind only changes what is recorded:

KindWhat the walkthrough showsDevices (default)
websiteHome, a strong section, menu, a service/product page, projects, mobile menu, contactDesktop ~11 s + mobile ~11 s
web-appDashboard, opening a record, a real action and its result, the responsive viewDesktop + mobile
appThe app on a phone: home or onboarding, the main flow, a real actionMobile ~20 s
automationThe trigger (a form, an email, a new row), the workflow running, the result in each app, joined by cutsDesktop ~20 s
brandingThe brand guidelines or brand site: logo, palette, type, applications, with cutsDesktop ~20 s

The template and the agency's identity define the look. See STYLE-GUIDE.md. Adapt the walkthrough, narration and the presenter's framing to the material.

The interview

Ask before you decide. Use the AskUserQuestion tool for every round:

  • up to 4 questions a round, with 2–4 short options each;
  • the recommended option first, labelled "(Recommended)";
  • one line of description per option;
  • the user can always type their own answer.

Base the options on what you found in the repository. Keep the creative interview to the three rounds below, after the presenter-format question.

Round 1: the project. Ask this after a quick read of the repo (README, package.json, routes, deploy config):

  1. What are we showcasing? Offer the kind you detected as Recommended, then the next most likely ones:
    • Next.js/Astro marketing pages → website;
    • a dashboard or auth routes → web app;
    • Expo or React Native → app;
    • n8n/Make/Zapier exports or workflow scripts → automation;
    • brand.json, brand guidelines HTML, logo files → branding.
  2. What is the post for? Portfolio piece (Recommended) · Client delivery · Promotion/ad.
  3. Where do we record it? The deployed URL you found (check it returns 200) · An existing local build (--serve dist) · A dev server (only with the user's permission, since it may write files) · Their own screen recordings (import).
  4. Which devices? The kind's default (Recommended) · Desktop only · Mobile only.

Round 2: the story. Ask this after inspecting the real pages, screens or workflow. In voice-over mode, a first capture can help. In presenter mode, move this round before capture and skip question 3 because the recording supplies the voice.

  1. What should the video highlight? (multiSelect) Up to 4 real features, pages or steps you found, named the way the product names them.
  2. Tone of the voice-over? Confident and warm (Recommended) · Calm and premium · Upbeat.
  3. Voice? British female, Kokoro bf_emma (Recommended: the approved voice) · British male, Kokoro bm_george · A cloud voice. Only offer a cloud voice if SC voices shows it is configured.
  4. Closing line? "Want a <website|web app|app|automation|brand> like this for your business? Send us a message." (Recommended) · Their own line. Keep it to about 3.5 s: the 3.6 s outro holds the signature for a second after it.

Round 3: approval, before rendering. Show the user:

  • a storyboard table: part, seconds, what is on screen, the narration line;
  • the cover preview (SC cover <slug>) and the profile grid (SC grid <slug>), sent as images.

Then ask Render it? with the options Render (Recommended) · Change the script · Change the walkthrough. Loop on changes until they say render.

Save the answers in the project config under brief, so the video can be remade the same way.

Workflow

Follow WORKFLOW.md. In short:

  1. Inspect (read-only). Framework, scripts, build output, deployed URL, languages (prefer /en), pages, navigation (plain links vs dropdowns, the mobile menu), strongest visual sections, privacy risks.

  2. Interview, round 1.

  3. Create the project:

    SC init <slug> --kind <kind> --url <url> --name "<Client>" --category "<Category>"
    

    This writes the config with the approved template, the kind's timing and the cover turned on. Then edit capture.journeys, narration.lines, project.displayUrl and video.output. Examples are in CONFIG-EXAMPLE.md.

  4. Web apps and automations need a demo account. The user runs SC login <slug> --url <app> and signs in in the Chrome window themselves. You never type passwords. The session is saved to the project (capture.auth).

  5. Capture: SC capture <slug>. Read capture/report.md and a few steps/*/after.jpg. Hide banners and overlays (capture.hide) and blur anything personal (capture.blur).

  6. Interview, round 2. Then write 7–8 short lines tied to their steps and run SC narrate <slug>. In voice-over mode, aim for about 25 s. In presenter mode, use the script and source takes reviewed before capture, and follow the timing of the recording.

  7. Review: SC still <slug>, SC cover <slug> and SC grid <slug>. Then interview round 3.

  8. Render:

    • SC render <slug> writes the MP4, instagram-cover.png, the grid-crop preview and the QA report, and records the post as a draft in the portfolio;
    • it takes a few minutes, so run it in the background.
  9. Validate: read out/qa/report.md and fix every FAIL. Look at the contact sheet and the cover preview.

  10. Deliver (below). When the user says it is posted, run SC portfolio publish <slug> [--url <post url>].

The profile stays one portfolio

  • Keep the studio's structure, identity, cover layout, music and outro consistent. Use the selected narration source for each Reel.

  • Eyebrow taxonomy on the cover: <Kind> · <Category> (Website · Restaurant, Automation · Lead follow-up, Branding · Coffee roaster).

  • Before rendering, run SC grid <slug>. It previews the profile with this post first and warns about:

    • three posts of the same kind in a row;
    • a cover colour too close to a neighbour;
    • a title too long for a thumbnail.

    It also names the kinds missing from the last nine posts. Share the warnings with the user and let them decide.

  • The cover is a frame of the video, the settled cover lockup, exported losslessly. The title is the client's name or a 2–3 word project title. No extra text, stock images or effects.

Narration

  • Natural, confident, conversational English for an Irish business audience. Short sentences.
  • Describe only what is on screen, in the part where it appears. No invented features, claims, prices, numbers, awards or reviews.
  • For automations, say what triggers it, what it does and what the result is, in plain words.
  • Preserve the delivery pace. Shorten scripted lines before recording or adjust the edit to fit an existing take. For a generated voice provider that needs a key, name the environment variable.

Rules

  • Only real work. Show only real pages, screens, data flows and interactions. Use demo accounts and demo data for anything behind a login.
  • No private data on screen. Never capture or show credentials, customer or personal data, API keys, admin panels, staging banners or dev overlays. Blur or hide what cannot be avoided.
  • Never handle passwords. The user signs in themselves (SC login). Never write keys into files.
  • Keep the identity. No zooms, camera rotations, flashy transitions, sped-up scrolling or site animations, long static screenshots or generated walkthrough footage. Presenter mode may use the avatar clip supplied by the user.
  • Never edit the template for a project. Don't change the engine's src/ or a published template's numbers. Everything project-specific goes in the project config.
  • Music: the engine's own bed, or a track the user supplies with its licence recorded. Never download music.
  • Never publish anything. Don't post, upload or publish the video, cover or projects anywhere. portfolio publish is local bookkeeping only.
  • Check this repo. At the end, run git status here: it must be unchanged.

Deliver

Report:

  • the MP4 path, duration and kind;
  • instagram-cover.png and the grid preview;
  • the walkthrough (what was shown);
  • the narration and voice;
  • the music source;
  • the QA result;
  • the grid check;
  • what the user should still check: client approval to publish, the agency details in agency.config.json, licences.

Leave the files in <engine>/projects/<slug>/out/ unless asked otherwise.

Supporting files

相关技能