Skip to content

agentsope/SkillAlchemy / skills/agentsope-ai-video-director

agentsope-ai-video-director

Unverified

Direct 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。

references/research_notes.md

View on GitHub
# Research coverage and limits

This package was independently distilled through SkillAlchemy → Lens → LEAP Branch A at pinned workflow revision `6ea799f6deb10ee48d66a644e595b1ffb84ef9a6`. Structural exemplars affected organization only; no skill fusion/persona imitation was performed. Runtime instructions are 20 General and 20 Scoped admissions, within exact sigma, plus nine explicitly authored C contracts. No excluded candidate was promoted. Forty cards preserve 46 unsupported empty components and 16 admitted engine-profile bindings.

## Counts and evidence categories

- Fourteen seed repositories plus two counted upstream projects were studied; official documentation and asset/font/service pages expand the citation ledger. A framework maintainer is not counted as a creative author just because its API was read.
- Forty-two creative case units, seventeen original public prompts and two author-published briefs were retained. Prompt and case counts overlap and must not be added. Revisions/language variants with verified provenance are deduplicated; ambiguity remains recorded in aliases/case categories.
- Seven independent creator teams and at least seven production scenario families appear across the creative cases. This does not imply equally strong evidence for every requested video type.
- Ten research outputs had sampled-frame observations. No research case had independent full-motion viewing or actual audio listening. Encoded metadata, code and author verdicts remain separate evidence classes.

See [case ledger](case-studies.md), [source index](sources.md) and [machine trace](provenance.json). They preserve original URLs/revisions, statement origins, confidence and rights. No complete third-party prompts, assets or code are shipped.

## What is supported

Version-bound engine contracts support HyperFrames composition seconds/registration/readiness/ownership; Remotion frame clocks, separate export paths, resource handles and typed audio/caption timing; Motion Canvas released generator/project/editor exporter/audio/events. These contracts are not comparisons of artistic superiority. P15 applies only to TikTok auction advertising placement, not every social clip.

P02, P05, P06 and P09 are optional source-authored suggestions, 待验证, even though admitted reusable within their scopes. Their aesthetic/communication effectiveness was not independently validated. Other creative decisions and repairs remain source-local Scoped notes. Literals such as 300→80px, 9px blur, 2.5s/50% note visibility and a 1.5 decay ratio retain film/style-specific scope, never universal defaults.

## Research-time execution observations

- HyperFrames 0.8.142: original 2s / 60 frames / 30fps / 640×360 silent H.264 smoke encoded and probed; selected frames/repeated seeks observed. Async media/fonts/audio and complete playback were not established. npm gitHead differs from the docs commit; only the recorded compared pages were byte-identical.
- Remotion 4.0.534: original 2s / 60 frames / 30fps / 320×180 silent H.264 smoke encoded and selected/repeated stills observed. Actual pixel format was yuvj420p; no yuv420p result is claimed. Browser export, audio and complete playback remain unexecuted. Concurrency 2 is the executed configuration, not an optimum.
- Motion Canvas released 3.17.2: TypeScript, Vite build and local server checks passed. Editor render/video export/UI/audio were not executed. The released wrapper is MIT; same-number main exporter metadata differs (GPLv3). The exact inspected cached FFmpeg binary’s nonfree/nonredistributable build is excluded from the archive.
- Source-local Canvas/WebGL renderer: code/encoding route inspected only; not executed. Its browser path is preview, not demonstrated browser-only encoding.

These are research-time original smoke fixtures, not the package’s authored example films and not final user productions. Package/fixture/runtime/review results are recorded separately when an actor actually executes them. A structural validator cannot establish visual or audio quality.

## Remaining limits

Cinematic depth, language substitution, silent-versus-voiced comprehension and controlled creative matched-pair evidence remain limited or absent. No diffusion prompt/output method was admitted. Remotion production eligibility depends on the actual subject/use under pinned v4 terms and remains unresolved here; upcoming v5 terms cannot substitute for that contract. Third-party media/font/music/service rights remain separate from repository licenses.

The framework-neutral workflow, seven genre application guides, chosen short-duration defaults, nine-dimension rubric, state vocabulary and packaging/test conventions are C01–C09 authored task contracts. They carry no fabricated empirical proof. New synthetic examples have pending implementation/render/full-motion/audio states until actually produced and reviewed.

## Executed package evaluation — additive release record

A fresh delegated worker loaded this compiled skill and produced 14 retained actual responses: the three sentence requests (P01/P02/P03), five routing cases and six missing/error cases. The strengthened actual-output validator passed the final retained response set (exit 0). The first failure, exact-fixture clarification and follow-up reruns remain in the separate local audit. These are delegated instruction-workflow evaluations, not external Claude Code or Codex CLI integration tests. The authored fixture definitions retain their original `not_executed` labels; the executed responses and reports are recorded independently.

The root also implemented and checked an independent original B01 HyperFrames benchmark: four shots, 15 seconds, 1080×1920, 30 fps and 450 decoded frames; H.264/yuv420p, silent MP4, 1,410,286 bytes. It differs from the worker’s five-shot 720×1280 P01 response and is excluded from research-case counts. Its actual technical probe passed. Browser checks reported zero runtime/layout errors and 42/42 contrast checks, with five structural warnings and motion audit disabled. Observed implementation, font and shot-boundary defects were fixed and rechecked using before/after evidence. The root inspected eight representative frames and twelve boundary frames; four ordered/shuffled seek samples had identical SHA-256 hashes.

Full-motion viewing remains **not_executed**. Audio listening is **not_applicable only for this explicitly silent benchmark**; it does not clear pending narration in P03 or the narrated-docs application. Final production rights and product claims remain unknown, and `publish_ready` is false. The encoded draft, frames, logs and reviews remain in the separate local audit, outside the runtime archive. Static/technical/frame evidence establishes only its actual coverage and does not certify audiovisual or production completion.

Two independent reviews completed with explicit evidence/rights/scope and routing/runtime limits. Twenty-one checker mutation probes were executed and rejected their invalid contracts; these are validator checks, not actor responses. The historical research observations above remain unchanged.

The final actor run included R08, giving 14 actual retained responses in total. Its two isolated prompts share the same 15s director plan but keep HyperFrames seconds and Remotion450-frame clocks separate. Both independent reviews completed with explicit limits; 21 checker mutation probes were independently rerun and passed their expected rejection checks. Actual ZIP contents and project discovery links are verified separately during packaging.
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: 472kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexUnverifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall