- name
SKILL.md
The skill file stays in the publisher's repository.
Paste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add nexu-io/openmotion --skill openmotion-recreate -a cursor -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add nexu-io/openmotion --skill openmotion-recreate -a claude-code -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add nexu-io/openmotion --skill openmotion-recreate -a codex -yPaste the copied command in a terminal. It installs this SKILL.md into that agent.
npx skills add nexu-io/openmotion --skill openmotion-recreate -a grok -ySend the copied message in a Grok Bot chat. The Bot fetches this SKILL.md from GitHub and adds it to the skills all your Bots share.
Install this skill: https://github.com/nexu-io/openmotion/tree/HEAD/skills/openmotion-recreateFrom the publisher
The user supplies a video and wants a version of it for themselves. Deliver the editable OpenMotion project, open in the editor, where the user reviews it and keeps editing. The default workflow reads the reference, builds the complete requested film, makes one focused preview comparison at up to four representative moments and fixes clear defects in at most one batch. It then exports with automatic technical verification and hands over the result. Detailed frame comparisons, scoring and extended visual revision follow an explicit request. Rebuild covers reference understanding and authoring. Read difficult visual reconstruction when the reference requires that method. Export and requested review owns this route's delivery and review scope, including where shared guidance describes a broader review. Use only the Tools actually returned by tools/list.
Rebuild everything native code can carry: text, UI, charts, logos, captions, shapes, camera and layer motion. Parts of the reference that native code cannot rebuild, such as portraits, product photography, live-action or AI-generated shots, narration and music, become media in the same editable project: generated through openmotion-media under the shared generation boundary, supplied by the user, or kept from the reference soundtrack. A generated shot matches the original shot's role, framing, subject type, motion and timing, not its exact pixels or a real person's identity; keep text, logos and UI as native layers above it rather than baking them into generated pixels. Without a usable media provider, say plainly which parts cannot be produced, finish the rest natively, and offer the choices: add a provider in Model providers, upload their own footage or images, or accept a native approximation. State limits as facts about the workflow; never mention, quote or link this Skill or its files in the reply.
For an accepted onboarding recommendation, follow accepted template selection. Retrieve the accepted ID through openmotion_get_template and use that reference's instructions and motionDocument. An accepted sample experience retains the reference's content and motion unless the user requests adaptation; incidental workspace branding does not restart intake or change the template. Preserve the ID and source in project working notes and recovery. For that first successful sample delivery, also read the first-run template experience for its reply, next-step invitation and failure rules.
Read the shared context before working; it lists the references a recreation needs and how to reuse documents already read. When the work changes an existing draft or project, also apply the shared edit workflow. If the reference turns out to be only loose inspiration for an original film, switch to openmotion-director.
Before asking anything, look in the current workspace for the user's own material: README and docs, product or marketing copy, brand files, logos, screenshots, colour or font definitions, package or site metadata. Read what is there. This is the user's brand, product and latest features; use it as the content of the film.
Then read the reference as rebuild describes: a local decodable copy, its URL and checksum, the whole video and consecutive frames around important changes. Links and local files take the same route. A cover image or an unreachable link is not a viewed video; never claim to have watched it.
Work out the film the user wants from everything available: their words, whose film the reference is, what the workspace holds and the conversation so far. Short requests often mean more than they say. A bare link in a workspace with product material usually asks for a film like this about that product; "do this for our new feature" or "I want this feel" usually asks for the reference's craft carrying the user's story; an explicit list of changes asks for exactly those changes.
Whether the reference's brand stays or becomes the user's is your judgement from the request and context, not a fixed rule. When a wrong guess would change the whole film, such as "make one identical" while the workspace holds the user's own brand, put it in the single question below; otherwise decide and state the assumption at delivery.
Parts the request does not cover keep the original. Requested changes are in scope without restriction: wording, brand, images, colours, typography, pacing, duration, adding or removing scenes.
Ask only when the workspace and the request together still lack something only the user can supply (brand, product, claims, numbers, prices, logos, images), or when an ambiguity would materially change the result. With usable material in the workspace, do not ask who the film is for. Offer "you decide" where it fits. When selecting, obtaining or replacing an image, clip or sound, apply the shared material policy within this workflow's supported media scope. Ask for a material gap only when it cannot be resolved within the current scope, available sources and authorization. When parts need generated media, plan them during this step and put the media cost-consent question from openmotion-media (planned assets, count, provider/model, what the key cannot make) into this same single batch. If nothing else is missing, that consent is the only question.
Ask all of it in one batch through the host's question interface, in the user's language. If nothing is missing, do not ask. Do not ask about technical choices, intermediate approval or storyboards. After the answers, work through to the finished film without stopping.
Follow the runtime adapter and Native authoring. A continued recreation reuses its projectId. Otherwise create one project with openmotion_create_project and a stable requestId: projectType: "reference_video" with referenceUrl for a link, or projectType: "natural_language" for a local file, with the file path and the agreed brief in requirements. Creation makes an empty carrier, not a film, and never overwrites another project.
Read openmotion_get_project({projectId}) for the directory, revision, current version, hyperframesCli and authoringGuide; read the pinned engine guide before writing source. Commands you run yourself (node <hyperframesCli>, the measurer, ffmpeg) use node and ffmpeg from PATH. The self-contained package does not put them on PATH: its directory is the one holding node_modules/hyperframes in the hyperframesCli path, with Node at runtime/node/bin/node (Windows runtime\node\node.exe) and FFmpeg/ffprobe in runtime/media/bin. For an npm installation there is no bundled runtime/ directory: the application root is the directory containing node_modules/hyperframes, and scripts/npm-runtime.mjs exports ensureNativeRuntime(appRoot) and nativeConfiguration(appRoot). Prepare through the former when needed; the latter returns paths.HYPERFRAMES_FFMPEG_PATH and paths.HYPERFRAMES_FFPROBE_PATH for the verified cache. Pass these as environment variables to the measurer or direct media commands. When neither is available, record the affected check as blocked with the reason; never as passed.
For detailed moving material, scrolling text or textured breakup, use the reusable rendering code. It supplies complete-texture transport, source-coloured sprites, release masks and motion exposure. Use its optional source observer when existing source views leave a concrete mark-size, brightness or transition question unresolved; batch those observations and sprite preparation. Read that reference for the applicable mechanism and save the rendering helper with the editable project.
Follow rebuild to choose the editable representation from source observations and build the user's version directly. Resolve appearance-bearing materials, visible bounds, motion controls and event timing together before writing the complete implementation. Carry observed texture and mark-size distributions through motion, exposure and breakup. Reuse suitable supplied implementations and the pinned engine's supported capabilities with their coordinate and timing contracts. For dense multi-shot work, use the segment reconstruction process, delegate independent units when available and authorized, and integrate their source. Keep source observations, representation decisions, timed controls and known gaps in existing working notes. Default creation follows the focused preview pass and delivery below.
Import real media with openmotion_import_assets and use the returned paths; save text files (HTML, CSS, JS, SVG, JSON) with openmotion_edit_project using the latest expectedRevision and expectedVersionId. Re-read the project after a conflict instead of replaying an old batch.
Before grouping reference regions into materials, inspect each proposed material region through time and inventory its independently changing contents, including small changes within stationary boundaries. Connect those changes to source controls in the reconstruction.
When the reference is the user's own finished film and the request changes only a few details (a year, a line of copy, a logo) while everything else stays, the project may keep the original video as its base layer: import it, cover only the changed areas with clean patches and set the new text or logo as editable native elements that follow the original's timing and motion. Fit each patch to its surroundings, including edges and underlying texture. This is still the editor project and delivery below; never deliver an MP4 edited outside the project. Otherwise, placing the whole original video on the timeline is not a recreation, and an unrelated built-in template is not a substitute.
New content must keep the reference's craft, not just its palette. Match its visual density (a screen full of tiles stays full), camera and 3D motion (a tilted orbit stays tilted), type scale and type character (the closest available face of the same kind, such as a pixel, mono or wide display face, instead of a generic system font), and its background layers (line drawings, grids, textures). A line, label or headline that is clearly visible in the reference must be equally visible in the remix, and the remix must not be darker or lower-contrast than the reference. Preserve the density and spatial relationships of dense and 3D passages in the implementation.
Apply the changes with care: fit longer text with related layout and reading-time changes, update every relevant appearance of a replaced brand or object, keep unrelated content and motion, never invent facts, and check audio timing when changes shift pictures.
Sound: when the reference has sound, the remix has sound. Keep the reference soundtrack by default (extract it as AAC .m4a and import it) and re-time it to the new cut. A voiceover that names the reference's brand or product is replaced with the user's lines in the reference's language, delivery and timing when speech generation is available and consented. An unanswered consent question stays pending; do not remove or replace the narration to get around it. If generation is unavailable or declined, follow the audio alternative the user chose; if the narration cannot be separated from the music, say so. Deliver a silent film only when the user asks or the reference is silent.
Once this turn's final video HTML and assets are complete and ready for preview, complete the editable preview handoff automatically, before final export or delivery. Keep intermediate authoring silent; present the final chat video after technical export verification through this route's delivery workflow.
Follow the single focused preview pass in delivery, then export the current version and use that job's automatic technical verification. Recover concrete render, asset-loading or file-integrity failures from their reported cause, then hand over the result.
Numeric reference comparisons, full playback-review sessions, reviewer delegation and extended visual repair follow an explicit review, evaluation or visual-change request. A review-only or evaluation-only request returns findings for the inspected version. Apply visual fixes when the user requests revision, covering the affected interval and dependent appearances. Report known visual gaps and distinguish technical export success from visual acceptance.
The project open in the editor is the deliverable. Do not copy, link or attach the exported MP4 or other files unless the user asks for them.
Reply briefly in the user's language. Give the previewUrl returned by openmotion_open_preview and say the editor is open only after confirming visible latest content through the shared handoff; otherwise state the display failure or confirmed inactive-chat deferral, even if the MP4 exported successfully. Say in one or two sentences what changed, which assumptions you made from the workspace material, and any limitation the user would notice. When the reference, analysis or tools fall short, name the specific gap and its effect; never claim a faithful result that was not achieved.
Apply follow-ups to the same project with the shared edit workflow. Use §6 for export and review scope, and validate requested changes over their affected intervals and dependencies. Never start over. openmotion_list_versions and openmotion_restore_version bring back an earlier version as a new one. When the user is unhappy without saying why ("not satisfied", "第一版不太满意"), ask at most three multiple-choice questions in one batch, for example: follow the reference's story or tell our own; which feature is the hero; size and duration. Offer "you decide" as an option. Then revise the existing project.
© 2026 Skills Rack.
---name: openmotion-recreatedescription: >----# skills/openmotion-recreate/SKILL.mdFAQ
Download skills/openmotion-recreate/SKILL.md from the publisher, or copy the path and open the repository. PlugMCP lists the skill and does not host the file.
The publisher keeps it at https://github.com/nexu-io/openmotion.
Claude, Cursor, Codex, and Grok Build can read a SKILL.md you add from the publisher's repository. Grok Bot installs it when you send the skill's GitHub folder link in a chat.