git:20260810.1f29900 to git:20260810.f69fb10

2 added, 2 removed. Audit A to A.

# narova

Prompt-to-video CLI. Turns a scene script (`reel.config.mjs`) into an MP4 with

local TTS, word-level captions, and speech-timed visuals rendered through

- HyperFrames. The standalone CLI package lives at `tool/`; the agent skill at

- `skills/narova/` contains instructions and references only.

+ HyperFrames. CLI code lives in `tool/`; agent instructions and references live

+ in `skills/narova/`.

## Key gotchas

- `data-cue="k"` counts turns from 0. `data-cue="0"` = the first turn.

- No CSS animation/infinite/hover/transition in theme.css — the renderer

jumps between frames. Motion comes from `reveal`/`data-cue` + `data-*` animators.

- SVG ids are namespaced per scene at compose. Style with classes, never `#id` in theme.css.

- Never edit `out/` or `out/hf/` — every compose regenerates them.

- Agent shells don't persist env vars — use `narova` or spell out `node tool/bin/narova.js` every call.

- Post-processing renders with ffmpeg concat need `setsar=1` in every video chain.