New Vox-style reel
Follow AGENTS.md (rules, repo map) and docs/WORKFLOW.md (full process). This skill is the
checklist — create a task per step and stop at every ✋.
- Read
docs/STYLE.md,docs/REELS.md, and the code of the closest existing reel. - Align ✋ — AskUserQuestion: angle, length, screenshot sourcing. Present a beat table; get a yes.
- Data —
npm run fred -- …(or save another cited source todata/); compute every figure with a script; report numbers + sources; flag contradictions. - Screenshots ✋ —
npm run capture, crop with ffmpeg, show the user. Real articles only. - Script ✋ — 80–110 words / 40s, numbers spelled as spoken, saved to
scripts/reels/<reel>.txt. - Costs — preflight Higgsfield costs (
get_cost: true) and state the total before generating. - Assets — VO with word timestamps →
public/audio/<reel>/; music; stills (check each one). - Build — copy
src/reels/_template/→src/reels/<reel>/; beats viavoFrame("word", nth); register insrc/Root.tsx; next series number. - Verify —
npm run typecheck; stills at every beat/annotation, tiled and inspected; render; check frames + loudness from the MP4. Run the review checklist in WORKFLOW.md. - Hand off — send the MP4 with a beat summary, sources, credits spent, and caveats.