← LeaderboardThe Skill Leaderboard

openmotion-growth-teardown

OpenMotion — Reverse-engineer how a company grew and turn it into a sourced growth-map video. Use when someone names a company (a competitor, a breakout startup, their own product) and asks how it grew, what moves drove its growth, for a growth teardown, growth timeline or "copy how X grows". Produces dated, cited growth moves on the metric curve, three plays worth copying and a 1920×1080 motion film rendered from editable HyperFrames source.

Published by nexu-io ↗nexu-io/openmotion

Try your first task

A source-grounded starter task has not been verified for this skill yet.

SKILL.md

name: openmotion-growth-teardowndescription: OpenMotion — Reverse-engineer how a company grew and turn it into a sourced growth-map video. Use when someone names a company (a competitor, a breakout startup, their own product) and asks how it grew, what moves drove its growth, for a growth teardown, growth timeline or "copy how X grows". Produces dated, cited growth moves on the metric curve, three plays worth copying and a 1920×1080 motion film rendered from editable HyperFrames source.

OpenMotion · Growth Teardown

An OpenMotion Skill. Every film it makes is branded as made with OpenMotion: the OpenMotion wordmark opens the film, the OpenMotion mark leads every scene headline and the wordmark signs the outro. The company being analysed keeps its own logo where it appears (the command line, its gallery card); never replace or restyle either brand.

Name a company, get every move that grew it — on one timeline, with the receipts. The output is a short film and the data behind it:

  1. growth.json — the metric curve (revenue run rate, ARR, users, stars…), every dated growth move in 3–4 lanes, and three plays. Every number and move carries its source URL.
  2. An editable HyperFrames project generated from that file (9 scenes, ~45 s) and its MP4.
  3. A short written teardown (three plays, each with evidence and the one instruction a team should act on).

The film's power comes from the research, not the animation: a curve that bends right after the moves that caused it, and real headlines, thumbnails and pages as proof. Spend most of the effort in step 1.

1. Research the moves

Follow research.md. In short:

  • Search wide before reading deep: the company name across web, news, Reddit, X, YouTube, GitHub, Hacker News and Wikipedia, limited to the last 12–24 months; then read the press pieces that report metrics, the pricing page, the blog/changelog index and the sitemap.
  • Extract dated moves only: free windows and trials, pricing changes, creator/affiliate/partner programs, contests, launches, distribution deals, SEO/programmatic pages, integrations, and every funding round. Each move: date (day if known), a ≤ 5-word label, the source URL and the source's own headline.
  • Extract metric points: each reported value with its date and who reported it. Never interpolate a value you did not read; mark company claims vs. third-party estimates.
  • Group the moves into 3–4 lanes named by mechanism (e.g. Free windows · Creators · Search · Fundraising). Fundraising is usually the last lane.
  • Pick three plays: the lanes whose moves line up with a bend in the curve. A play is a claim ("free windows sped up growth."), a counted fact ("4 free windows in 16 months"), 2–3 evidence moves and one instruction a team can execute.

If a company has no public metric series, use the best proxy with a clear label (GitHub stars, downloads, reported users, traffic estimate) and say so. Never invent a curve.

2. Write growth.json

Use the schema. Keep labels short enough to read in a half-second glance; keep the full source headline in source.title. Validate it:

node scripts/build.mjs --check growth.json

The check rejects undated moves, metric points without a source, plays without evidence and labels that will not fit.

3. Collect the evidence images

node scripts/evidence.mjs growth.json evidence/

Material is what makes the film credible, so collect it deliberately:

  • Real pages. For each evidence move the script screenshots the source page in a headless browser (cookie banners removed), uses the thumbnail for YouTube links, and falls back to the page's og:image.
  • Thumbnail walls. A play with wallQuery gets up to 48 real video thumbnails from that search (e.g. every creator review of the product), shown behind the evidence cards while the counter runs.
  • Icons and logo. Favicons for each research source and the company's own icon.
  • Made graphics. Some proof is better drawn than screenshotted: give a move a card instead of an image — chips (e.g. the four blog languages), list (e.g. sitemap paths such as /mcp/...), or stat (one big counted number). Draw only what the data says.
  • Moves left without any image become press-clipping cards carrying the source's real headline and date.

Use only what the source published; do not generate stand-in images of real people or products. Look at every saved image before rendering and replace any that show a login wall, an error page or an unrelated page.

4. Generate and render the film

node scripts/build.mjs evidence/growth.json out/
cd out && npx hyperframes check && npx hyperframes render -o growth.mp4

Build from evidence/growth.json: it is the copy with every collected image, wall and logo filled in. The generator writes a native HyperFrames project: index.html plus one composition per scene. The storyline is fixed because it works for any company:

SceneWhat it shows
IntroThe command being typed with the company's logo: the question
SourcesThe research sources lighting up, with the real call count
MapThe metric counter ticking while the curve draws and every move lands in its lane
Plays ×3The play's lane lifts out of the map, the curve turns orange after its moves, a counted stat, the thumbnail wall, then three evidence cards with leader lines to the exact dates
Full mapEverything at once: the screenshot people share
Plays to copyEach play's stat and the instruction to hand the team
Outro"Reverse-engineer any company", with other finished teardowns when --gallery a.json,b.json is given

Copy, colours and timings come from growth.json; scene files stay hand-editable. --lang zh writes Chinese headlines (or set "lang": "zh"). A license-free score is synthesized from the film's own cues (typing, pops, whooshes, hits) into assets/score.wav; --music <file> uses a licensed track instead and --no-score leaves it silent.

The generator uses the GSAP installed with OpenMotion; when this Skill is installed on its own, set OPENMOTION_HOME to the OpenMotion checkout or run npm install gsap in the working directory.

In OpenMotion, import or open out/ as a project and continue through the usual editor preview and export. Updating an existing managed project uses the shared project-edit contract with its current revision; do not write around version protection.

5. Review before delivery

Watch the render at full speed and check against review.md: every on-screen number traceable to growth.json, the curve bend lines up with the moves, labels never collide, each evidence card is the real source. Then deliver the MP4, the editable project, growth.json and the written teardown. Rendered output, source accuracy and the user's creative acceptance are separate claims; state which ones you checked.

Rules that keep it honest

  • Every visible number and move has a source in growth.json. Unsourced = cut.
  • Correlation is what the film shows; say "sped up after", not "caused", unless a source attributes it.
  • Company-reported, press-reported and estimated figures stay labelled as such in metric.points[].kind.
  • Use public material about public companies. Do not scrape behind logins or reproduce paywalled text beyond a headline.