What does this skill do, and when should you use it?
video-shotcraft is an AI agent skill for producing cinematic promos of web and desktop products. It bundles 157 shot recipe cards (214 styles with 214 motion previews), a validated 36.2-second Ink Press template, reusable Remotion components, and 149 SFX assets. It offers three modes: template-based reproduction, fully autonomous agent-driven creation, or stage-by-stage co-creation with the user. Its methodology emphasizes real page screenshots, a visual language grown from the product's design system, beat-synced cutting, deterministic rendering, and an independent final review.
The skill reads SKILL.md plus recipe cards, pipeline docs, and aesthetic rules under references/. After a minimal read-only product inspection, it captures real pages via headless browser (full-page 2x textures, element-level crops, layout. coordinate tables). It then generates a product brief, storyboard, and Remotion project according to the chosen mode, reusing tuned TSX implementations from demos/. Before rendering it runs librosa beat analysis, and after rendering it verifies cut-point accuracy; it can also export a JianYing draft or open a browser-based motion workbench for post-delivery edits.
- An indie developer needing a launch-day promo for a web/desktop product — use the Ink Press template route and swap in assets
- A marketing team wanting a brand-new-style launch video, granting the agent full autonomy from product understanding to final render
- A brand wanting creative control — pick co-creation mode and approve the brief, visual direction, and storyboard at each checkpoint
- A video creator who only needs one high-quality motion shot for an existing video — pick a card from the gallery and use it standalone
- An editor who wants to keep tweaking after delivery — export to JianYing or edit captions, retiming, and audio in the motion workbench
How do you install this skill?
- The delivery flow directs users' videos and social handles to the author's external showcase (a GitHub issue form); review before opting in.
- remotion.media CDN and GitHub Pages may be unreachable from mainland China; prepare a local chrome-headless-shell and offline previews.
- The Windows JianYing export path is self-admittedly unverified; Windows users should run smoke_test.py first.
- Remotion carries its own commercial license; companies must verify terms before use.
- Card/preview counts are inconsistent between SKILL.md and the description (157/214 vs 152/209); defer to the actual repository.
- Shell / CLI
- Network access
- Local filesystem
Node.jsRemotionPython 3headless browser (chrome-headless-shell)
The most direct way: hand the repo link to your agent — in Claude Code / Codex say "Install this skill for me: https://github.com/Vincentwei1021/video-shotcraft". Or install manually: git clone https://github.com/Vincentwei1021/video-shotcraft.git, then ln -s "$(pwd)" ~/.claude/skills/video-shotcraft (Claude Code) or ln -s "$(pwd)" ~/.codex/skills/video-shotcraft (Codex). Alternatively run npx skills add Vincentwei1021/video-shotcraft.
git clone --depth 1 https://github.com/Vincentwei1021/video-shotcraft.git ~/.claude/skills/video-shotcraftGenerated from the source repository and skill path; it copies only this skill's folder. If the author's install steps above differ, follow those first. To scope it to one project, replace ~/.claude/skills with that project's .claude/skills.
How do you use this skill?
Once installed, send your agent any of these to trigger it:
- Use video-shotcraft to create a promo for my desktop product.
- with the Ink Press template
- Use the deck-deal-flyin and row-embed shot cards to present this feature.
After install, issue prompts in Claude Code / Codex, e.g. "Use video-shotcraft to create a promo for my desktop product." (add "with the Ink Press template" for template mode) or "Use the deck-deal-flyin and row-embed shot cards to present this feature." If no mode is specified, the agent performs a minimal read-only product inspection and presents the three modes. You can pre-pick shots at https://vincentwei1021.github.io/video-shotcraft/. Rendering requires Node.js; on headless Linux boxes pass --concurrency=1 and use a chrome-headless-shell binary.
What are this skill's strengths and limitations?
- Large, pre-tuned asset library: 157 recipe cards with exact demo source code, 214 motion previews browsable online
- Complete methodology: real-screenshot capture, design-token re-skinning, librosa beat analysis, independent subagent final review
- Ships with a runnable complete template and 149 free-commercial-license SFX — production-ready out of the box
- Rich post-delivery ecosystem: browser motion workbench, JianYing project export, dual BGM/no-BGM renders
- Heavy local engineering dependencies: Node.js, Remotion, Python 3, headless browser — API-only callers must supply a shell
- JianYing export only verified on JianYing Pro 11.2 for macOS; the Windows module is untested on real hardware
- Remotion has its own license — free for individuals and small teams, but companies may need a paid license
- Template screenshots are demo assets only and must be replaced and data-sanitized before publishing
How does this skill compare with similar options?
Side by side with related skills; every score comes from the same FSRS standard.
| Skill | FS score | Stars | Last updated | License |
|---|---|---|---|---|
| Video ShotCraft: Cinematic Product Video Skill this page | 61 · Recommended | ★ 17k | 8d ago | Apache-2.0 |
| Vox Director: One-Topic Vox-Style Collage Videos | 56 · Use with care | ★ 2.6k | 1mo ago | MIT |
| Narrator AI Skill | 56 · Use with care | ★ 3.2k | 4mo ago | MIT |
| capcut-cli: Edit CapCut/JianYing Drafts from the Terminal | 69 · Recommended | ★ 1.1k | 2d ago | MIT |
| Huashu Design | 42 · Not recommended | ★ 29k | 2d ago | MIT |
The source material names no competing tools, so no comparison is warranted. Its sibling project video-talkcraft targets narration videos (78 cards, word-level timestamp alignment) and complements rather than competes with this product-promo skill.
How did FollowSkills review this skill?
SKILL.md codifies a minimal read-only product check, sensitive-data freezing/redaction, confirmation gates, and deterministic-seed rules; strong data-flow transparency. Deductions: delivery flow promotes the author's social accounts and a showcase submission (outbound user content), the Windows JianYing path is unverified, and rollback/permission scoping are not formalized.
Workflow is self-consistent, deterministic-render rules explicit, headless pitfalls documented with fixes, CI runs tsc+vitest+smoke render. Deductions: static review cannot execute reproduction, the Windows JianYing path is self-admittedly untested, and failure-feedback quality on abnormal input is not demonstrated.
Trigger conditions are precise (explicit skill/template/shot-card invocation), three-mode boundaries and non-fit ranges are clear, Chinese docs are complete, and the remotion.media CDN blockage risk in mainland China is disclosed with a local-binary workaround. Deduction: gallery previews depend on GitHub Pages downloads, a residual mainland reachability risk.
Well-layered docs (SKILL → references → demos), LICENSE (Apache-2.0), audio ATTRIBUTION, acknowledgements/trademark notes, changelog-style updates. Deductions: publisher identity and maintenance commitment not verifiable, and card/preview counts in SKILL.md (157/214) contradict the description (152/209).
Ships a validated template, copyable components, audio assets, and test/CI evidence; the template is reproducible via npm install + render, with clear marginal value over manual production. Representative output quality cannot be independently verified statically, so the score is capped at 7.
Repo includes a CI workflow (pr-checks.yml), vitest unit tests, and a demo smoke-render script; fact/inference separation is decent. Deductions: nothing was executed in this review, tests cover only helper functions rather than the full production path in SKILL.md, and multi-source corroboration is thin.
Open a dimension to read why it scored that way
Evidence confidence：Low — Mostly static review, author material or a limited demo; useful for discovery, not high-risk decisions.
See the full review method →