atlantis-mk/three-dimensional-remotion-template

Create the episode video using the template

Entry and Responsibilities

Locate the project root directory containing the root rules, README, and this skill. Initially, the work/ directory may not exist. First read the root rules, README, this skill's references/template.md, and any existing work/task.json; if no workspace exists, treat it as a new production. Current projects and references are already in place, do not request the template again. Essential questions, phases, and default deliverables follow the root AGENTS.md.

The project template and reference materials are placed together with this skill under assets/, technical specifications under references/, and runtime tools under scripts/. The root directory keeps this package's rules; a single-episode workspace is created when starting production. Use the entire template package as-is. For existing tasks, verify the actual files and preserve any modifications. When performing maintenance only, do not initialize the episode project, request assets, or advance episode state.

Production Workflow

1. Examine assets, requirements, reference visuals, and source code; collectively gather essential information that affects the result. When topics and directions are clear, proceed continuously without asking for confirmation at every stage.
2. Read references/task-state.md and record actual parameters, asset usage, storyboards, and timelines. Narration, images, and videos are not required to be complete; do not force new voiceovers or music.
3. Read references/remotion.md. Use pipeline.py env to check the environment; on first run, execute init --install to create the workspace and a blank task state, copy the baseline, install locked dependencies, and export a full content.json. For existing projects, adapt without overwriting or rebuilding.
4. Modify the episode parameters and scenes. When changing the subject matter, check model props in Scenes.tsx and Visuals.tsx; replacing a sign does not equal a subject change. If images or videos need to be displayed, integrate them into the episode code.
5. When narration is present, measure its duration and reorder scenes, subtitles, metrics, and sound effects based on semantic content. The script does not provide TTS, transcription, or automatic alignment; if narration is supplied, do not check narration service keys. An updated schema or total duration does not imply audio-visual sync.
6. Set content.cover text and scene selection. The three cover layouts are already separately defined; there is no need to recreate the entry point for each episode. Check the main subject, title, and boundaries individually.
7. Use validate to check parameters, timing relationships, and audio files; use check to verify types and actual composition; use preview to generate keyframes and a preview of a specified interval with audio. Review visually and audibly, then correct.
8. Use render to output the actual deliverable list for this episode; use verify to decode and check media information and the current rendering log. Upon technical passing, still complete the root rule's audiovisual acceptance check, then update the status and deliverables.

The script outputs JSON on its last line; exit codes are 0 for success, 2 for usage error, 3 for missing environment, 5 for invalid parameters, and 7 for execution or check failure. First read the report log, then make corrections; do not set a cumulative check limit. The script does not automatically mark the task as completed.

Reference on Demand

• Template style, parameters, and validation scope: references/template.md. The project baseline is at assets/project-template/, and reference keyframes are at assets/reference/.

• Initialization, parameters, assets, preview, rendering, and technical checks: references/remotion.md.

• Task resume, phase progression, and acceptance records: references/task-state.md.

• Modify README machine-readable metadata: references/readme-format.md.

• Module and maintenance verification: root directory docs/architecture.md. Evidence from tests/ is not part of this episode's deliverables.