HyperFrames Domain Skills Roundup: Which of the 5 Do You Need
HyperFrames ships five domain skills and the naming lies: hyperframes-creative contains no animation code, hyperframes-keyframes owns almost no scene strategy, and hyperframes-audio — the one with the steepest install curve — did not exist three weeks ago. I read all five SKILL.md files in heygen-com/hyperframes (43,150 stars, Apache-2.0) and pulled fresh install numbers off skills.sh on 2026-08-31. The boundaries between the skills turned out to be the real content, and the curves backed up the file sizes.
The Five Skills and the Line Between Them#
The pack splits by concern, not by name. hyperframes-core is the technical contract: the data-* timing attributes, class="clip", sub-composition templates, the single paused gsap.timeline registered at window.__timelines[compositionId], determinism bans, and a 14-row reference index. It is the only skill you must read before writing composition HTML. hyperframes-creative is the brand layer — design-spec handling (frame.md → design.md → DESIGN.md precedence), palettes, typography, narration pacing, beat planning, audio-reactive visuals — and its frontmatter says it outright: “intentionally non-animation.”
The split matters because the two motion skills collide. hyperframes-keyframes covers visual camera work: punch-in/out, Ken Burns pans, reframes, whip handoffs, masks, SVG morphs, and 2D/3D keyframes, framed as a “pose contract” with a mechanism-choice table (FLIP for box changes, path travel for routes, stroke draw for traces). hyperframes-animation holds everything else about motion: atomic rules, multi-phase scene blueprints, transitions, 24 named text effects, and seven runtime adapters — GSAP default, plus Lottie, Three.js, Anime.js, CSS keyframes, WAAPI, and TypeGPU/WebGPU. Their shared boundary rule: keyframes only animates visible handoff on wrappers inside clips, and never touches clip timing or source ranges — that is core’s data-start/data-duration/data-media-start domain.
The File Sizes Predict the Newcomer#
Size order inside the pack: hyperframes-audio 24,734 bytes, hyperframes-keyframes 15,901, hyperframes-core 13,066, hyperframes-animation 7,785, hyperframes-creative 6,354. The largest is the youngest. My skills.sh scrape shows audio at 89,089 lifetime installs with a weekly series of five straight zeros — [0, 0, 0, 0, 0, 17,702, 36,501, 32,178] — meaning it hit the registry roughly three weeks ago and jumped from nothing to mid-30K weekly installs. Every other domain skill traces the pack’s lockstep shape instead: core, creative, keyframes, and animation all peaked near 43-44K weekly in the same week and drifted to the 25-33K band since, totals running 250K-348K. Audio is the exception that confirms the pattern — a genuinely new skill grows alone, an established pack member just counts along.
Why Audio Is Worth a Deeper Look#
The size is not padding. hyperframes-audio reads like a mixing console spec: a mix is “a set of relationships, not a stack of processors,” and every tool exists to express one relationship. Effects live on the element as data-fx-chain; envelopes ride data-automation; a voiceover carve keeps its own data-fx-carve settings so you can re-derive the ducking instead of guessing filters. Twelve effect families ship (gain, EQ, compressor, limiter, gate, saturation, delay, reverb, chorus, phaser, bitcrush), and — the detail that sold me — preview and render run the same Web Audio graph builder, one offline, one live, so “what you hear while scrubbing is what gets written. You never tune twice.” A <hf-audio-group> bus carries a chain, a fader, and an automation clock across many tracks at once. The whole file reads like a manual for a tool that does not exist anywhere else.
Rounding Out the Family#
Two companion skills hold the tooling around the domain five. hyperframes-cli (461,670 installs, the top of the pack) documents lint, check, snapshot, preview, render, and the keyframe-proof flags — --shot, --ghost, --layout strip — while hyperframes-registry (403,886) manages reusable blocks versus components with a 24-hour TTL. The main hyperframes entry point (446,618) is the mandatory router: it resumes project state, runs a BRIEF/STORYBOARD intent layer, and owns the 10-workflow routing table we dissected in the hyperframes-vs-remotion comparison last week. If you are installing by hand, the honest shortlist is: core for anyone writing compositions, animation next for motion, keyframes when camera work appears, creative for brand-heavy deliverables, and audio — right now, before the pack curve absorbs it — for anything with narration over music.
Where to Start#
Do not install all five at once; the boundary rules do the routing for you. Start with npx skills add https://github.com/heygen-com/hyperframes --skill hyperframes-core, let the composition contract sink in, then add animation and keyframes when a scene actually needs motion. Audio you can grab today and watch its install curve climb for another week or two — the file is the strongest writing in the pack, and the 12-family effect chain has no equivalent in Remotion’s toolkit. The summary, in one line: read core, route by boundary, and trust the newcomer with the biggest file.
Related Articles
Azure Skills Roundup: Spot the 4 Removed Skills
4 microsoft/azure-skills - azure-rbac, azure-hosted-copilot-sdk, cost-optimization, observability - deleted while install totals kept climbing.
HyperFrames vs Remotion: HTML Video vs React Video in 2026
HyperFrames renders HTML video via a 10-workflow skill router; Remotion renders React. Audits cover seek-driven motion, graded migration, 457K installs.
Anti UI Slop vs Web Design Guidelines: Stopping Generic UI
Two skills fight AI-generated UI slop: Vercel audits code against 16 rule groups, UIZZE grounds builds in 800K real screens. Weekly installs: 21K vs 121K.
Prototype vs TDD: When to Explore Throwaway and When to Go Test-First
The Matt Pocock skills offer two ways to build something new: Prototype throws code away to answer a question, TDD locks it down test-first. Pick right.