agentsope/SkillAlchemy / skills/agentsope-ai-video-director
agentsope-ai-video-director
UnverifiedDirect and implement programmatic videos from a natural-language sentence: brief, shots, style, timeline, assets, engine-specific coding prompts, draft, review and repair. Use when asked to make a video, product launch, explainer, Reel/Short, kinetic typography or docs-to-video with HyperFrames, Remotion or Motion Canvas, or to improve an existing code-video project. 一句话视频制作与导演工作流。
examples/data-explainer.md
View on GitHub# Synthetic data explainer · 30 seconds
Authored C01–C07 application, not the P08 temperature-film marks/reveal or P34 eCPM counter repair. Geometric bar growth and fixed printed endpoints are authored requirements for this demonstration.
## Input and fact ledger
Input: “用30秒横屏解释收入从40增长到55,简洁数据动画。” Preserve target 30 s, starting/ending revenue values 40/55. Their units, periods, measurement source, precision, sampling and causes are unknown. Treat these as user-supplied demo values, not independently verified observations. Derived arithmetic 55−40=15 and (55−40)/40=37.5% are labeled calculations from supplied data; avoid implying a scientific mechanism, trend beyond two points or causation. Package defaults: horizontal web target, 16:9, 1920×1080, 30 fps, Chinese labels and explicitly silent evaluation draft. Audience is a general viewer. Main message: “给定的第二个数值比第一个大 15(原单位)”.
## Director plan
| Shot | Seconds / 30 fps frames | Focal subject/action | Framing/layout/copy | Transition |
|---|---|---|---|---|
| S1 | [0,4) / [0,120) | Two revenue labels establish comparison | Center title; “收入的两个数值”; demo/source unknown note | Inner title fade |
| S2 | [4,10) / [120,300) | 起点 bar geometry rises to 40 | Aligned baseline, direct label 起点 / 40; illustrative display axis 0–60 | Short mark reveal, printed 40 fixed |
| S3 | [10,17) / [300,510) | 终点 bar appears to 55 | Same scale/baseline, direct label 终点 / 55; 起点 remains | Short mark reveal, printed 55 fixed |
| S4 | [17,24) / [510,720) | Equation appears beside bars | “55 − 40 = 15”; “差值 15(原单位) · 增幅 37.5%(计算)” | Small equation opacity ramp |
| S5 | [24,30) / [720,900) | Full comparison holds | Both bars; “用户提供的演示数据;单位、时期与来源待确认” | Stable hold |
Closure: 4+6+7+7+6=30 s; 900 frames, last 899. Style tokens: near-white #F5F6F8, navy #182C47, teal #1A8F9C, title 68 px, direct labels 44 px, note 32 px, 100 px working inset. These are authored design choices. Single neutral axis scale is illustrative, with no unit or period invented. Narration/captions: none in this declared silent draft, and no source creator's sonification ratio is applied.
Assets/rights: two original bars, original equation and user demo numbers (authored/supplied evaluation use); no external icon, media, music or bundled font. A future dataset export/image/font/audio requires separate permission and source records. Engine production eligibility remains independently unresolved.
## Conditional route and full coding-agent prompt
Conditional Remotion **4.0.534 server** route when exact Node/browser/encoding tools and subject/use permission are established; do not substitute a browser compositor or v5 terms. Unknown eligibility blocks declaring a production route selected. HyperFrames 0.8.142 is a separate possible adapter using the seconds table; it requires its own exact dependency/readiness/rights probe. With neither exporter, the concrete result is plan/prompt/code plus gaps.
Send the entire original [Remotion coding-agent prompt](remotion-prompt.md) with this ledger. It already contains this application's exact 900-frame Composition, five shot intervals, visual tokens, labels, static/frame determinism, stable resource waits, pinned commands and file probe, acceptance/repair and truthful delivery. If changing engine, recompile the entire plan through [that adapter](../references/engine-adapters.md); do not merely replace an engine name in the prompt.
## Draft, QA and repair expectations
Expected draft: source and lock/version record, fixed values and starting/ending labels, clear unknown-unit/period/source note, actual encoded file if allowed/runnable and retained probe/frame evidence. Static QA verifies dataset literal values, derived difference, consistent scale/labels, half-open frame closure and isolated engine namespace. Samples include 0,119,120,299,300,509,510,719,720,899. Compare repeated/reordered representative frames within the actual tested environment. Technical QA compares actual dimensions/fps/duration/frame coverage; do not assume pixel format from an old smoke. Full viewing decides pacing/animation, and actual audio absence is checked in the probe for the silent draft.
If actual labels overlap, adjust this authored layout and repeat the affected frame checks. If units/periods/source later arrive, update the ledger/copy, recompute any changed numbers and revalidate those frames. Do not borrow the eCPM blur value or note-opacity threshold as a universal repair.
Current state: plan_available=pass; code_available/static_checked/preview_observed/encoded_artifact_available/technical_probe_passed/sample_frames_reviewed/full_motion_reviewed/reproducibility_scope=not_executed; audio_listened=not_applicable for explicit silence; rights_confirmed=unknown for final use. This file is a complete plan/prompt application, with no claimed rendered draft. [Quality checklist](../tests/quality-checklist.md), [record format](../tests/manual-review.md).
## Complete authored coding-agent prompt
Build a project-local synthetic data explainer: 30 s, 1920×1080, 30 fps, 900 frames, last frame 899. Data supplied for this demo: starting revenue 40, ending revenue 55. Units, periods, source and cause are unknown. Labels must say “用户提供的演示数据;单位、时期与来源待确认”. No accuracy/causal claim may be invented. Target audience is a general viewer, and horizontal web presentation is a labeled package assumption. Silent draft is an explicit evaluation assumption. Use original vector marks, system-rendered text and no third-party media.
Before production, record the exact subject/use eligibility basis under the pinned v4 license. Individual/nonprofit/for-profit with at most three employees/noncommercial evaluation conditions differ from other for-profit production uses. Unknown eligibility stays unknown and prevents a production render route from being declared selected. No Company License purchase is authorized. The local noncommercial smoke establishes no user production permission. Keep hosted services, codecs, media rights and server/client telemetry scopes separate; do not substitute announced v5 requirements for v4.
Use version-matched packages and `registerRoot(Root)`. Register a `Composition` with id `DataDemo`, component, fps 30, width 1920, height 1080 and `durationInFrames=900`. Derive visible state from zero-indexed `useCurrentFrame()`. `Sequence` intervals are [from,from+duration), and each child sees a local frame. Explicitly pass absolute frame only if needed by the authored design. Use `spring({frame,fps,...})` with fps 30 and `interpolate` with explicit left/right clamp where the visual requires bounded values. Do not use wall-clock timers or mutable prior-tab state. Concurrency 1 is not a remedy for wall-clock animation.
Implement five shots: S1 [0,120) title “收入的两个数值” on near-white #F5F6F8 with navy #182C47; establish the two supplied revenue values without animation of their digits. S2 [120,300) show a bar for 起点 fixed at 40 on a clearly labeled 0–60 illustrative display scale; animate geometry, keep its printed value fixed. S3 [300,510) reveal the 终点 bar fixed at 55 beside the first, with aligned baseline and direct starting/ending labels. S4 [510,720) show the authored arithmetic “55 − 40 = 15” and “差值 15(原单位) · 增幅 37.5%(计算)”; these are computations from supplied values, not a cause or newly sourced measurement. S5 [720,900) hold both bars and the provenance/unknown note. Title 68 px, direct labels 44 px, source note 32 px, 100 px working inset are authored design tokens. Use straight cuts/short inner-mark opacity ramps; keep the note readable in each relevant shot and test actual overflow. No universal reading threshold is asserted.
For asynchronous resources on this server path allocate a stable lazy `delayRender` handle with `useState`; call `continueRender(handle)` on completion or `cancelRender(error)` on failure. Do not allocate on every rerender. Load any authorized font before measuring text. Default readiness timeout at this pin is 30 s. The draft has no requested audio; if later requested, import `Audio` from `@remotion/media`, use `staticFile` for authorized local audio, and keep `from`, `trimBefore`, `durationInFrames` in frames. `durationInFrames` selects source frames and occupies durationInFrames/playbackRate parent frames. `Caption` from `@remotion/captions` is plain data with `text`, `startMs`, `endMs`, `timestampMs:number|null`, `confidence:number|null`; preserve whitespace. Captions use milliseconds, not frame counts, and creating data is not transcription or alignment.
Use the admitted local server commands as separate invocations after verifying executable/browser paths:
```sh
./node_modules/.bin/remotion versions
```
```sh
./node_modules/.bin/remotion studio <entry.tsx> --port=<port>
```
```sh
./node_modules/.bin/remotion still <entry.tsx> DataDemo <output.png> --frame=300 --image-format=png --browser-executable=<chrome-path>
```
```sh
./node_modules/.bin/remotion render <entry.tsx> DataDemo <output.mp4> --codec=h264 --concurrency=2 --browser-executable=<chrome-path>
```
```sh
ffprobe -v error -show_format -show_streams -of json <output.mp4>
```
Concurrency 2 is a reproduced smoke profile, not a performance optimum. The equivalent documented server API route is bundle/serve URL → selectComposition → renderMedia; do not replace it with `renderMediaOnWeb`. Compare repeated and reordered still capture at frames 120,300,510,720 within the actual environment. Verify frames around those transitions and final frame 899, numbers/labels, chart baseline, source note, dimensions/duration/fps and actual audio state. A deliberate empty starting spring frame requires judging its authored purpose; it is not automatically an export defect. Fix observed geometry/text defects against this plan, rerun affected checks and retain changes/logs.
Return code, plan, detailed prompt, command log and actual artifact/probe only when they exist. Report eleven C07 states separately and nine QA dimensions. A build or still does not prove the whole video; full-motion/listening stays not_executed until observed. Server licenseKey/success telemetry differs from browser completed/failed telemetry; do not hide that distinction. If tools or eligibility prevent export, deliver plan/prompt/code and the concrete gap, with encoding/probe states not_executed.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 472k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 198k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |