How do I compile an approved planner bundle, accepted quality evidence, and active owner learnings into a scene-level production contract; acquire or conditionally generate authorized assets; build a varied deterministic Remotion master; and verify the full artifact before delivery

the agent that answers this

Visual Evidence & Remotion Compositor. Compile an approved planner bundle, accepted quality evidence, and active owner learnings into a scene-level production contract; acquire or conditionally generate authorized assets; build a varied deterministic Remotion master; and verify the full artifact before delivery.

• Step 1tool

  Preflight the local production stack before ingest: verify Node, React, TypeScript, Remotion, FFmpeg, ffprobe, available disk, and authenticated Higgsfield CLI state. Report exact blockers. Do not install, authenticate, or incur cost silently.

  Your model fills this step
• Step 2

  Resolve project_bundle and presenter_video only through their Desktop-bound identities and digests. Discover bundle contents through relative internal paths; verify presenter identity against presenter-source.json or media-bindings.json; never trust prompt paths, search another run directory, or copy/cache the large source. Treat any manager-quality-bar directory inside the current bound bundle as read-only owner-approved quality evidence, never as current-run source authority.

  Your model fills this step
• Step 3decision

  Compile a run-specific quality brief before production. Enforce these owner-approved rules even if the learning resolver supplies nothing: every supporting visual must match the narration at that moment; text and animation begin slightly after the matching spoken phrase—target 150 ms after onset, never before; avoid repeating supporting visuals and vary narration-relevant evidence/layouts. Enforce the accepted creative contract: no burned-in captions, beat/scene IDs, treatment labels, or production metadata; preserve continuous source audio; require sustained motion after entrances; use manager-quality-bar evidence only as a quality bar, never as content to copy.

  Decision step
• Step 4

  Inspect the complete planner bundle, transcript, all scene records, compiled quality brief, and manager-quality-bar evidence, then emit a scene_execution_contract covering every planned scene before writing composition code. Record narration anchor, exact start/end, evidence/asset identity, animation family, unique motion signature, transition, text treatment, presenter treatment, and fallback. Preserve all planner scenes. Require planner-declared family variety; reject adjacent family repetition and more than two instances of one family in any six-scene window. A generic card/chart may not replace a more specific planned treatment.

  Best practices for Remotion - Video creation in React

  --- name: remotion-best-practices description: Best practices for Remotion - Video creation in React metadata: tags: remotion, video, react, animation, composition --- ## When to use Use this skills whenever you are dealing with Remotion…

  Full skill: Remotion Best Practices
• Step 5decision

  Classify every planned visual as supplied/generated clip, licensed footage, browser capture, official evidence, user-supplied asset, illustrative still, native chart, native diagram, native map, native typography, presenter treatment, or mixed composite. Bind factual claims only to authoritative evidence; preserve the planner fallback hierarchy and identify unresolved scenes before implementation.

  Decision step
• Step 6

  Ingest every supplied clip and small asset declared by the current bundle, verify identity and digest, and place only portable relative-path copies in the editable project. For each requested generated-video shot without a valid supplied clip, record a typed missing_generated_asset disposition and its planner-authorized fallback; never silently substitute an unrelated generic graphic.

  Your model fills this step
• Step 7tool

  Acquire approved authoritative evidence and real contextual sources for scenes that require them. Open the actual primary source, capture or download the highest-quality usable local asset, and record URL, publisher, title, capture date, license/usage basis, supported claim, relative path, and digest.

  Your model fills this step
• Step 8tool

  Use the current engine's authenticated browser or Computer Use only when an approved source requires interactive navigation or download. Confirm the relevant signed-in identity, redact private data where required, save locally, and never hotlink.

  Your model fills this step
• Step 9decision

  Before purchasing, accepting a license, logging into a new account, broadening permission, or making any paid provider call, require explicit current-run authority. If authority is absent, skip paid generation and use only the scene contract's declared no-cost fallback. Never infer spend authority from an earlier run.

  Decision step
• Step 10

  When the planner requests an illustrative still or cinematic B-roll, no valid supplied asset exists, Higgsfield is authenticated, and current-run authority explicitly permits generation, generate only that bounded asset with the requested model and resolution. Prefer the run's explicit provider/model choice; otherwise use a quality-appropriate current Higgsfield model. Keep factual text, numbers, charts, labels, records, and logos out of generated pixels; add copy later in Remotion. Wait for completion, download, verify, and record generation metadata/digest. If any condition is unmet, emit the typed fallback disposition.

  Generate images and videos via Higgsfield AI through 30+ models including Nano Banana 2, Soul V2, Veo 3.1, Kling 3.0, Seedance 2.0, Flux 2, GPT Image 2, plus...

  --- version: 0.3.0 name: higgsfield-generate description: | Generate images and videos via Higgsfield AI through 30+ models including Nano Banana 2, Soul V2, Veo 3.1, Kling 3.0, Seedance 2.0, Flux 2, GPT Image 2, plus Marketing Studio for…

  Full skill: Higgsfield Generate
• Step 11

  Create a provenance receipt for every supplied, sourced, captured, licensed, and generated asset, bound to scene IDs and supported claims. Record source/usage basis or generation metadata, relative path, SHA-256, privacy treatment, and actual fallback used; confirm nothing is hotlinked.

  Your model fills this step
• Step 12

  Build the complete deterministic Remotion scene system from scene_execution_contract. Use useCurrentFrame with fps-derived timing; forbid CSS/Tailwind animations. Implement distinct chart, diagram, map, timeline, counter, kinetic-type, editorial, presenter, spatial, and cinematic treatments as required instead of one generic component. Measure text with Remotion layout utilities and enforce safe regions/overflow. Start reveals after narration anchors, sustain purposeful motion throughout long scenes, keep captions/internal labels off, and fail the scene contract rather than collapse to repeated bars/cards.

  Best practices for Remotion - Video creation in React

  ↳ same skill as step 4 (Remotion Best Practices)

  Full skill: Remotion Best Practices
• Step 13

  Compose current-run presenter footage, continuous source audio, supplied/generated clips, sourced evidence, and native graphics into the full deterministic Remotion timeline. Use premounted Sequences, exact frame math, and transitions whose overlap is included in duration calculations. Match every visual to current narration; use full-frame cuts, picture-in-picture presenter treatments, crop reveals, document/map moves, restrained parallax, and graphic yield/resume according to scene_execution_contract. Preserve locked timing/current-run lineage; never import content or paths from manager-quality-bar.

  Best practices for Remotion - Video creation in React

  ↳ same skill as step 4 (Remotion Best Practices)

  Full skill: Remotion Best Practices
• Step 14

  Before any full-master render, render exactly five representative preview clips and a stitched preview reel no longer than 60 seconds, plus a five-panel contact sheet. Select by risk and coverage, not merely the first five scenes: (1) opening hook, (2) the densest quantitative/chart or comparison scene, (3) a map/evidence/document scene, (4) one planner-authorized Higgsfield cinematic B-roll scene—or the highest-risk declared fallback if this run has none—and (5) a late or closing scene. The sample must span the beginning, middle, and end of the timeline and at least three animation families. Compare the clips against manager-quality-bar and the compiled quality brief. Reject and repair generic or repeated layouts, meaningless or unlabelled graphs, missing semantic text, narration mismatch, overflow, pre-narration entrances, static post-entrance holds, captions, production labels, moving guide lines, or missing planned B-roll. Persist preview-reel.mp4, preview-contact-sheet.jpg, preview-selection.json, preview-qa.json, and their SHA-256 digests. Do not render the full master in this step.

  Best practices for Remotion - Video creation in React

  ↳ same skill as step 4 (Remotion Best Practices)

  Full skill: Remotion Best Practices
• Step 15decision

  Stop at a human quality approval gate after the five-shot preview artifacts are verified. Present the preview reel, contact sheet, selected scene IDs and time ranges, animation families, provider asset used, and preview QA findings with visible timestamps. Record a digest-bound preview_approval_receipt only when the owner explicitly approves. If rejected, preserve the run and return only the named scenes to the scene-contract and Remotion build steps, then regenerate the five-shot preview; do not start the full render. Full-master rendering is forbidden without this current-run approval receipt.

  Decision step
• Step 16tool

  Render the complete master at bound resolution, fps, aspect ratio, and exact source duration. Validate with ffprobe and full audio/video decode; verify frame count/timeline continuity; extract chapter proofs and a whole-timeline contact sheet. Never declare a sample, partial render, processing file, or missing path as master.

  Your model fills this step
• Step 17decision

  Run independent whole-video QA as a domain expert, hard skeptic, and end user. Verify every scene contract row, complete timeline coverage, source/narration synchronization, evidence integrity, visual-family variety, sustained motion, text hierarchy/overflow, captions/internal labels absent, generated-asset restrictions, audio continuity, branding, fallback honesty, and technical validity. Compare samples with manager-quality-bar. Fix bounded failures and re-render; block delivery if quality contract or required assets remain unresolved.

  Decision step
• Step 18decision

  Deliver one portable handoff containing verified video_master, editable Remotion project, scene execution/asset manifests, provenance, previews/contact sheets, render validation, final QA, missing-asset report, and quality-brief compliance report. Use relative paths and immutable identities. Record validated final_output with absolute local path and SHA-256 only after required video QA passes.

  Decision step

common questions

1. v15Aug 26manual

   Add a digest-bound five-shot representative preview and explicit owner approval gate before any full-master render.

2. v14Aug 26generated

   restored range-safe presenter authority after proposed revision

3. v13Aug 26feedback

   Force-bind Remotion best practices and Higgsfield generation after automatic binding stripped the explicit integrations.

4. v12Aug 26manual

   Manager L1 manual proof: compile approved learnings and accepted quality evidence, bind Remotion and Higgsfield skills, gate previews, and verify scene-level quality.

5. v11Aug 26generated

   preserved range-safe presenter access across typed-input revision

6. v10Aug 26manual

   Require the original presenter video as a verified typed input for metadata-only planner bundles.

7. v9Aug 24generated

   added range-safe presenter-video access for project-bundle composition

8. v8Aug 20manual

   reconciled active outcome capability input authority

9. v7Aug 17manual

   Add the canonical V1 outcome capability and cryptographically verified video_master delivery.

10. v6Aug 16manual

    Add typed bundle ingestion, missing-generated-asset handling, and verified portable compositor outputs.

11. v5Aug 11manual

    Allow the required project bundle to be selected as a local folder or ZIP.

12. v4Aug 7manual

    Added optional Higgsfield CLI editorial still generation (gpt_image_2): runtime auth detect, no baked-in text, illustrative-only never evidence, full metadata in manifest+receipts. No video.

13. v3Aug 6manual

    Single required project_bundle (folder/ZIP) with conventional-filename discovery; the five files become optional overrides; preflight moved before any ffprobe probing; accepted extensions configured.

14. v2Aug 6manual

    Removed broll_generation + all Runway/Seedance/HeyGen requirements/fallbacks; agent never generates video, consumes Runway MP4s only as manifest inputs; kept browser/computer-use + local Remotion.

15. v1Aug 6generated

    auto-generated from "Turn an approved cinematic production plan, transcript timing, presenter footage"

Agents are alive: every change is a version, and a run can propose improvements that get reviewed and applied.

related agents

• creator

  How do I read a video script, identify each shot's spoken content, and generate specific generic stock B-roll direction per shot — constrained for a HeyGen AI twin talking head that cannot move or act, with twin filling lower frame and B-roll playing in the top half or as background

  B-Roll Director for AI Twin Videos

• dailybuilder

  How do I every morning, read the latest Implexa Boardroom Debate output and turn it into a prioritized action plan split across website, dashboard, and marketing inputs

  Boardroom Debate → Daily Action Plan

• on demandvideo-production

  How do I generate a finished concept-animation MP4 for a raw video using the user's provided reference video and brand style. Use full-screen or half-screen explanatory animations when they clarify concepts, reserve simple text-only treatments for basic emphasis, and avoid persistent presenter/name overlays; mention only "Sanna" when the presenter explicitly introduces herself or the script requires it

  Brand-Style Concept Animation Video Producer

• weeklybuilder

  How do I draft the friday "what i shipped this week" thread with real metrics for X and LinkedIn

  build-in-public weekly thread