This Skill first completes content analysis, then hands the actual artifacts to the local Motion Craft for production. Analysis and production are clearly sequential stages; "starting to write scenes after watching part of the SRT" is not treated as completed analysis.
Each new video stores inputs, analysis, project files, and outputs in an independent working directory; sequels and partial revisions continue using the current project and qualified assets specified by the user. When the user explicitly requests independent retesting, recreate from the original video, SRT, explicit references, and current rules provided this time, without reading project files, finished videos, analysis, matting caches, or historical conversations from other rounds, and without requesting old answers from other tasks; identical inputs are not automatically treated as a sequel. Public fonts, general mathematical tools, and references with usage rights may be used.
Fully read the analysis rules and inputs
Execute from the project root:
node scripts/read-stage.mjs --stage srt --kind talking-head
For pure SRT animation use --kind animation. Execute the returned NEXT_READrepeatedly until END_READING; each tool call returns only one page, and do not batch-print all rules. This stage fully includes semantics, key points, and timing, the input scope, and the content fields of the original production plan. First fill in any pages that were not fully returned; do not read only titles or the first few lines of each section.
Read the complete SRT, the user's original requirements, and valid corrections; distinguish original footage/original audio from visual references; verify aspect ratio, duration, and retention scope. When voice-over packaging has no authorization for deletion or modification, preserve order, content, and timing; when only sentence-level SRT is available, do not fabricate word-by-word timing. For facts, totals, and proper names, first verify the context; when a missing item would change the content and cannot be determined, ask only about the specific gap and continue work that is not affected.
Artifacts handed to production
Save the analysis in the current project's existing content review,PLAN.md、timeline.json or corresponding analysis file; if none exists, create a content analysis in this video's working directory. It is a production input, not an additional stage checklist, and does not require a separate storyboard or anchor image.
The content should directly support production: complete content structure corresponding to the original timing; each segment's original meaning, visual expression task, screen emphasis, and necessary conditions; character/evidence/material requirements; final numeric values, enumerated members, and summary callbacks; the opening and main turning points; the user's latest requirements and the superseded scope; source files and reference entry points, audio conditions, and timecode precision. Preserve an accessible path to the complete original SRT; do not replace the original text with a summary.
If a qualified analysis already exists, first read it completely and verify the currently valid requirements, reuse its content, and do not generate another one. For an analysis-only task, delivery of this artifact ends the task; when production has been authorized, then have Motion Craft first read this artifact completely, then enter production from assets , and do not proceed from just a verbal summary.
Production handoff and scope
Production follows Motion Craft's stages: "material character and background determination → visual direction and key composition → continuous sample segments, segment-by-segment expansion → audio → full-video review." Before starting work, each stage fully reads all of its corresponding rules; during the animation stage, do not write the entire video while catching up on reading. In the layout stage, do not first finalize static pages for the entire video; later, complete composition, motion, and transitions together segment by segment. There is no default stage-by-stage user approval.
For real-person talking-head production, first establish an independent character layer synchronized with the original footage, then decide the outline based on the actual background: when retaining the character's original video background, character outlines are prohibited; when replacing the original background behind the character, a wide and relaxed outline must be applied. Placing the original scene into a small window does not change this condition; the card must still have a 3D floating hierarchy, and the character within the frame must be re-centered in the shot. The complete conditions are unified in Character Rules, and implementation is maintained by the corresponding space and container sections of Motion Craft. Existing qualified keying should be checked and reused. Pure animation with only SRT does not apply character keying. Complete detailed rules enter by stage; do not load all layout, keying, animation, and rendering materials in advance during the analysis stage. When the user only modifies audio, a partial shot, or text, retain their scope and continue using unaffected analysis and confirmed content. Rule applicability and priority are uniformly executed according to the Motion Craft entry point.
The error correction and font poster research newly added in phase three are still in the formal rules; the colors, frame shapes, aspect ratios, and specific seconds of each phase are not generalized into fixed templates. The original production requirements enter the corresponding stages by chapter via Production Principles and other references; migration and historical evidence are only reviewed when tracing sources. The current structure does not enable the revoked multi-Agent segmented production scheme.