@koda-sl/baker-cli 0.80.0 → 0.81.1

package/README.md

@@ -3631,6 +3631,8 @@ It then scaffolds the full pipeline: per scene, two **static-ad-grade frames** (
 
 **Overlays are agent-painted HTML, not props.** The clips are concatenated, then the `video-overlay` composition (copied next to the canvas) composites the overlay layer. The scaffold **bakes the reference's overlays into that composition's `index.html` as real, editable HTML** (each overlay is a plain element with its text, a `.pos-*` position class, and `data-start`/`data-dur` timing); a tiny generic runtime only shows/hides each element at its timestamp (with an optional `data-anim` entrance). It makes **no styling decisions** — bars, tickers, colors, fonts, and a real logo `<img>` you drop into the dir all live in the HTML/CSS you edit. Floating elements (logo bugs) are seeded as commented `<img>` stubs so an un-edited render stays clean. Drop `brand-bold.otf`/`brand-regular.otf` for on-brand type.
 
+**Re-craft the script — the hook is the #1 decision.** A reproduction is *inspiration* from a proven ad, not a clone: its structure (hook → body → CTA) carries the persuasion, and the hook is *targeting*, so a competitor's hook often does **not** transfer. `metadata.todo.script_recraft` tags each scene with its `narrative_role` (from the deconstruct, else inferred) and carries the original line **flagged** so it is never shipped as-is — and the per-scene `recraft` instruction is **role-aware**: the **hook** scene's entry carries the diagnose → decide (keep/adapt/rebuild) → criteria (statement not question, benefit by ~2s, first frame legible **sound-off** in ~1s, no bait-and-switch) inline and routes to the skill's `references/hook-craft.md`. A dedicated top-level **`metadata.todo.hook`** key foregrounds it as the highest-leverage beat, mapped onto scene-0's artifacts (`s0_start` first frame, scene-0 overlay text, `s0_clip` line, micro-hook, hook-ramp).
+
 The emitted canvas is validated (`validateCanvasDeep`) before it's written, so it always runs. It also carries a **`metadata.video`** timing plan that `baker canvas validate` proves **statically, before any billed render**: no two voiceover turns overlap, the audio length ≈ the video length, and every single-on-camera-speaker scene is a native talking head (its clip carries `generate_audio` and is wired to an `audio_voice_convert` node). The full editable checklist is embedded as **`metadata.todo`** (with a step-by-step guide in `metadata.description`). stdout returns `{ ok, canvas_path, prompt_path, models, stats, checklist }`.
 
 ```bash

package/dist/cli.js

@@ -9443,7 +9443,7 @@ function buildMotionBoard(blueprint, sceneTurns) {
     ].sort((a, b) => a.at_s - b.at_s);
     return {
       scene: i,
-      role: scene.narrative_role?.trim() || inferNarrativeRole(i, blueprint.scenes.length),
+      role: resolveSceneRole(scene, i, blueprint.scenes.length),
       window_s: [round(start_s), round(end_s)],
       storyboard_frames: [`s${i}_start`, `s${i}_end`],
       spoken,
@@ -9480,22 +9480,44 @@ function inferNarrativeRole(index, total) {
   if (index === total - 1) return "cta";
   return "body";
 }
+function resolveSceneRole(scene, index, total) {
+  return scene.narrative_role?.trim() || inferNarrativeRole(index, total);
+}
+function findHookSceneIndex(blueprint) {
+  const total = blueprint.scenes.length;
+  return blueprint.scenes.findIndex((s, i) => resolveSceneRole(s, i, total) === "hook");
+}
+var HOOK_OPENER_CRITERIA = "statement not question; benefit by ~2s; first frame legible SOUND-OFF in ~1s (the feed plays muted); no bait-and-switch (the hook sets up the body)";
+function recraftTextForRole(role, sceneIndex) {
+  if (role === "hook") {
+    return `[RECRAFT HOOK \u2014 the #1 decision, but usually the LIGHTEST move wins: this hook already worked, so default to KEEPING it and building on top (swap only the specifics for ours). REBUILD only when it doesn't transfer \u2014 a claim we lack or a different funnel/awareness stage \u2014 by reaching for its deeper INNER MECHANIC and delivering that truthfully. DIAGNOSE (device + mechanic + stage) \u2192 DECIDE keep/adapt/rebuild \u2192 hold the opener to criteria: ${HOOK_OPENER_CRITERIA}. It lives on s${sceneIndex}_start (first frame) + the scene-${sceneIndex} overlay. See references/hook-craft.md + meta-ads-playbook \xA710/\xA739.]`;
+  }
+  if (role === "cta") {
+    return "[RECRAFT CTA \u2014 the close: state what to do (where to go) + what to expect (offer/bundle/discount) for OUR brand; soft for top-of-funnel, hard for bottom. Do NOT render the reference's words. See references/script-craft.md.]";
+  }
+  return `[RECRAFT: rewrite this ${role} for OUR brand \u2014 true claims only; do NOT render the reference's words. See references/script-craft.md + meta-ads-playbook.]`;
+}
 function buildScriptRecraft(blueprint) {
   const total = blueprint.scenes.length;
   return blueprint.scenes.map((scene, i) => {
-    const role = scene.narrative_role?.trim() || inferNarrativeRole(i, total);
+    const role = resolveSceneRole(scene, i, total);
     const original = (scene.dialogue ?? []).map((d) => d.line?.trim()).filter((l) => Boolean(l)).join(" ");
     return {
       scene: i,
       role,
       original_line: original || null,
-      recraft: `[RECRAFT: rewrite this ${role} for OUR brand \u2014 true claims only; do NOT render the reference's words. See references/script-craft.md + meta-ads-playbook.]`
+      recraft: recraftTextForRole(role, i)
     };
   });
 }
 function buildVideoTodo(report, overlayCount, floatingCount, opts, blueprint) {
+  const hookSceneIndex = findHookSceneIndex(blueprint);
+  const h = hookSceneIndex;
   return {
-    recraft_the_script_first: "VIDEO IS INSPIRATION, NOT A CLONE. Before rendering, re-craft the script (hook \u2192 body \u2192 CTA) for OUR brand \u2014 the reference already won in-market, but its hook is targeting and may not transfer. Work the per-scene `script_recraft` checklist below; see references/script-craft.md + the meta-ads-playbook skill.",
+    recraft_the_script_first: `VIDEO IS INSPIRATION, NOT A CLONE. Before rendering, re-craft the script (hook \u2192 body \u2192 CTA) for OUR brand \u2014 the reference already won in-market, but its hook is targeting and may not transfer.${h >= 0 ? " The HOOK is the #1 decision (see the `hook` todo);" : ""} ${h >= 0 ? "then work" : "Work"} the per-scene \`script_recraft\` checklist. References: references/hook-craft.md (the hook), references/script-craft.md (body/CTA) + the meta-ads-playbook skill.`,
+    ...h >= 0 ? {
+      hook: `THE HOOK IS THE HIGHEST-LEVERAGE BEAT \u2014 the first frame + first 3\u20134s decide whether the ad is watched at all, and the hook is TARGETING. But highest-leverage does NOT mean always rewrite: this hook already won, so MOST OF THE TIME you KEEP it and build on top (swap only the specifics). REBUILD is the exception \u2014 only when it doesn't transfer (a claim we lack or a different funnel/awareness stage), and then by reaching for its deeper INNER MECHANIC and delivering that truthfully, not inventing a new opener from nothing. For scene ${h}: DIAGNOSE it (device + mechanic + what stage it targets), DECIDE keep/adapt/rebuild, then hold the opener to the criteria \u2014 ${HOOK_OPENER_CRITERIA}. The hook lives across s${h}_start (the scroll-stopping first frame), the scene-${h} overlay text, the s${h}_clip line, an optional ~0.5s micro-hook, and the ramp into the body. Full diagnose\u2192decide\u2192(keep/adapt/rebuild) discipline + the proven hook-type menu: references/hook-craft.md (+ meta-ads-playbook \xA710/\xA717/\xA739).`
+    } : {},
     script_recraft: buildScriptRecraft(blueprint),
     edit_frames_in_place: "Each s<i>_start / s<i>_end node has its own editable params.prompt (FRAME DESCRIPTION). Edit per frame; the blueprint is only a shared style reference. Frames are CLEAN PLATES \u2014 they render no on-screen text; all text is the overlay HTML layer.",
     frames_mode: opts.frames ?? "generate",
@@ -9527,7 +9549,7 @@ function buildVideoTodo(report, overlayCount, floatingCount, opts, blueprint) {
         "Nail the PACK SHOT \u2014 the final product hero sells (motivated camera move, light matched to the rest of the ad).",
         "Geometry/objects drifting frame to frame? A MAP beats a paragraph \u2014 drop a high-angle schematic still (marked positions) as an extra el_*/reference rather than adding more words.",
         "Before/after (dry\u2192wet, skeptic\u2192believer) = two SEPARATE locked reference images \u2014 don't ask words to transform a subject mid-scene (cheaper and more reliable than re-rolling clips).",
-        "The hook is VISUAL-FIRST: the feed plays muted, so scene 1's opening frame + its overlay text must read sound-off in ~1 second \u2014 don't bury the hook in the spoken line alone (meta-ads-playbook \xA739 visual-hooks-beat-audio, \xA748 the 1-second/feed-native rule)."
+        "The hook is VISUAL-FIRST: the feed plays muted, so the hook scene's opening frame + its overlay text must read sound-off in ~1 second \u2014 don't bury the hook in the spoken line alone (meta-ads-playbook \xA739 visual-hooks-beat-audio, \xA748 the 1-second/feed-native rule). The hook is also the #1 re-craft decision \u2014 diagnose/decide/rebuild it against criteria in references/hook-craft.md (see the `hook` todo)."
       ]
     },
     transitions: "Scene-to-scene cuts the deconstruct flagged as fade/whip/zoom/dissolve/swipe are reproduced as an ffmpeg xfade at the boundary (everything else stays a hard cut). The overlap is consumed from extra generated footage, so the picture stays exactly on the audio timeline. To change a transition, edit the scene's `transition_out.type` in prompt.json and re-scaffold, or hand-edit the `spine` node's ffmpeg args. For a richer HERO cut (whip-pan, glitch, light-leak, gravitational-lens\u2026), the overlay layer can run a Hyperframes shader/CSS transition instead \u2014 see references/hyperframes/blueprints-and-transitions.md (pick 2\u20133 transition types total; the motion IS the handoff).",