Flows diagram
Produce a branded Amigoscode animated BACKEND-DIAGRAM video: a React/SVG flow diagram (controllers, queues, threads, caches, DB cylinders, arrows, packets) animated over 10s and rendered to a 1440x1440 square MP4, a 960x960 looping GIF, and a 1080x1920 branded reel, then optionally given an ElevenLabs voiceover, word-by-word captions with animated emojis, an infographic-style caption.txt, and an upload title.txt. Use whenever the user wants an animated backend, Java, Spring, database, or concurrency DIAGRAM, flow, or 'how X works' reel, or to add a voiceover, captions, or a title to one of these diagrams. Examples: 'make a flows diagram for the outbox pattern', 'add a diagram about backpressure', 'render the square + reel for saga', 'give the retry diagram a voiceover and captions', 'generate title.txt for these'. This is the SVG-scene pipeline (make.mjs + puppeteer + ffmpeg). It is NOT the HyperFrames explainer-video skill and NOT the CSV reel skill.From its SKILL.md
npx -y skills add amigoscode/skills --skill flows-diagramAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 2 stars2 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
7.0 KB, as published. Nobody here has run it
Flows diagram
Turn one backend concept into a set of branded animated diagrams. Every diagram is a self contained React/SVG scene animated across 10 seconds and rendered three ways from one command, then voiced, captioned, and titled. The voiced, captioned reel is part of the standard deliverable, not an afterthought.
Zero en dashes or em dashes anywhere (titles, captions, voiceover). Periods and commas only.
Where things live
$SKILL— this skill folder. It carriesscripts/(the pipeline),assets/(brand logos and the scene template), andreferences/. Never write output here.$FLOWS— the workspace where diagrams are authored and rendered. One folder per diagram (<name>-flow/). Resolution order:FLOWS_DIRenv var, thenworkspaceDirin~/amigoscode-skills/flows-diagram-config.json, then~/flowsif it exists, then~/amigoscode-skills/flows(created on demand).
Set both once per session, then every command below is copy-paste:
SKILL=~/.claude/skills/flows-diagram # or the installed plugin path
FLOWS=$(node -e "import('$SKILL/scripts/paths.mjs').then(m=>console.log(m.workspaceDir({create:true})))")
First render fetches React and Babel into $SKILL/.cache automatically. puppeteer-core must be
installed once (cd $SKILL && npm install); rendering needs system Chrome (CHROME_PATH to
override) plus ffmpeg and ffprobe.
The pipeline at a glance
- Scene — author
$FLOWS/<name>-flow/<name>-scene.jsx(+ copyanimations.jsx) from$SKILL/assets/template/. This is the only creative step. Seereferences/scene-contract.md. - Render —
node $SKILL/scripts/make.mjs <folder> <sceneFile> <outBase> "<Title>" "<EYEBROW>" "<caption>"emits the square MP4, the GIF, and the reel. TuneCROP_TOP/CROP_BOTTOMper diagram so the diagram is centered between the title and the footer (see below). Seereferences/pipeline.md. - Words — write
caption.txt(the /infographic post copy) andtitle.txt(the upload name). Seereferences/voiceover-captions.md. - Voiceover + captions — write
vo-script.txt(~44 words), then run the voiceover + Submagic caption pipeline to produce<base>-reel-voice-over.mp4. Do this for every diagram by default (the reel ships voiced and captioned); only skip it if the user explicitly says they just want the silent base videos. Seereferences/voiceover-captions.md.
Fast path for a brand new diagram
cd "$FLOWS"
mkdir foo-flow
cp "$SKILL/assets/template/animations.jsx" foo-flow/animations.jsx
cp "$SKILL/assets/template/example-scene.jsx" foo-flow/foo-scene.jsx # then edit the diagram
# iterate on layout with the fast one-frame check (writes $FLOWS/.work/foo/foo-check.png):
MODE=check node "$SKILL/scripts/make.mjs" foo-flow foo-scene.jsx foo "How Foo Works" "SUBSYSTEM" "x"
# when the frame looks right, full render (square mp4 + gif + reel):
CROP_TOP=150 CROP_BOTTOM=140 node "$SKILL/scripts/make.mjs" foo-flow foo-scene.jsx foo \
"How Foo Works" "SUBSYSTEM" "Left → Middle → Right"
MODE=check renders a single mid-timeline frame in ~5s so you can dial in the diagram and the
CROP_TOP/CROP_BOTTOM padding before committing to a full 30fps render. Always iterate on the
check frame first.
Conventions (non negotiable)
- Palette: purple
#7f56d9primary, teal#2FA39Bsecondary/success, red#E5484Dfailure. Ink text#20273e. The scene template'sCobject already holds these. - Layout is locked in
make.mjs: purple mark at top, eyebrow + title pinned near the top, the diagram in the middle, ink wordmark pinned at the bottom. Do not restyle the frame; only the diagram inside#stagechanges. - Center the diagram between the title and the footer. This is the look we want: roughly
equal breathing room above the diagram (below the title) and below it (above the wordmark), not
jammed against the title with a big empty gap below.
CROP_TOP/CROP_BOTTOMare how you land it: a tall diagram fills the body naturally, but a short or single-row diagram needs a smallerCROP_TOPso it drops into the vertical middle. Always confirm the balance on theMODE=checkframe before the full render. Example: a thin single row centers atCROP_TOP=95 CROP_BOTTOM=130, whereas a tall diagram like a connection pool uses180/165. - 10 seconds, 30fps, k=1 (no breathing zoom). Scenes animate on
useTime()(0..10). - Voiceover ~44 words so the brand voice lands near 18 to 21 seconds. Longer scripts drift past 22s.
- Caption placement for reels is fixed:
caption-render.mjs ... 1080 1920 30 <dur> 1545 60 1420. - Clean up after every render:
make.mjswipes$FLOWS/.work/<outBase>automatically (setKEEP_WORK=1to keep frames for debugging). The voiceover driver deletes its caption frames. Never leave.workorcaps-*scratch behind; it grows to many GB fast.
Scaling to many diagrams (use agents)
Writing scenes, caption.txt, vo-script.txt, and title.txt is per diagram authoring work
that parallelizes well: dispatch one general-purpose agent per small batch (3 to 4 diagrams) to
write the text files, giving each agent the exact style spec. Do the RENDERING sequentially in a
single resumable driver (a done_file the loop skips) because puppeteer/Chrome and ffmpeg
contend when run in parallel. See references/pipeline.md for the batch-render pattern and
references/voiceover-captions.md for the batch-voiceover driver.
References
references/pipeline.md— themake.mjsCLI contract, every env var, the tuned 26-diagram render matrix (folder / outBase / CROP values / title / eyebrow), and the resumable batch-render driver.references/scene-contract.md— how to author a<name>-scene.jsx: the helpers,KCard/KIcon/ICONS, the DB cylinder,Defs, theSceneRouter/window.*exports, and how motion is driven by time.references/voiceover-captions.md—gen-vo.mjs,vo-align.mjs,caption-render.mjs, the emoji map, the exact ffmpeg composite command, and thecaption.txt/title.txtstyles.