LEAP
LEAP builds skills through two pipelines: Branch A distills a skill from raw data, while…
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
$ npx -y skills add agentsope/SkillAlchemy --skill agentsope-ai-video-director --agent claude-codeHow it fires
How this skill gets triggered: by you, by Claude, or both.
/agentsope-ai-video-directorContext preview
The summary Claude sees to decide when to auto-load this skill.
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
name: agentsope-ai-video-director description: >- 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。
Use this skill for an authorized video-production task, including:
Do not activate solely for:
Execute the following authored application chain (C01–C09). Every step produces a retained result before its dependent step starts. Read [operation cards](references/sop_models.md) and match each operation’s exact scope before use. Do not turn a source-local case into a rule for a new film.
Read [prompt fields](references/prompt-patterns.md). Extract goal, audience, main message, platform, supplied facts, duration/aspect, language, audio intent, assets, engine preference and delivery requirements. Preserve explicit values. Mark inferred audience, style and format as assumptions. If unstated, use these reversible **package defaults**, subject to user context: 15s for a short product/social draft; 30s for a compact data explanation; 60s for a compact docs/knowledge draft; 30fps; 16:9 for general explainers, 9:16 when vertical social is explicit; original geometric visuals and system fonts. These values are authored conventions, not evidence-proven creative optima. Retain supplied facts with their source. Record unknown facts and synthetic placeholders separately. A missing product metric must not become a fabricated benefit claim. Clarify only critical facts, permissions or assets that prevent the requested result; continue a provisional brief and authorized work while those inputs are pending. Output `director_brief` with `declared`, `assumptions`, `facts`, `unknowns` and `audio_intent`.
Read the matching [video guide](references/video-types.md) and [director fields](references/director-playbook.md). Choose an authored communicative pattern from the goal and available material. Write a shot table: ID, interval in seconds, subject/focal action, framing/layout, exact copy, animation intent, transition, asset, audio/caption cue and acceptance. Close the timeline at the declared duration; record intentional overlaps or gaps. Define palette, typography, hierarchy, backgrounds, safe-area assumptions and motion intent. Build an asset ledger with provenance, permission, intended use and unresolved rights. Keep original synthetic visuals available when third-party assets are pending. For voice-led work, use real authorized audio timing when ready (P28); if audio is pending, label the entire scene/caption clock provisional. Do not apply P05’s 2–10-minute or P06’s 90–180-second guidance to short examples as proven rules. P02/P05/P06/P09 are optional author suggestions, 待验证; they are never quality guarantees. Output `director_plan`: brief, shots, style tokens, timeline, assets, audio, target and acceptance.
Read [isolated adapters](references/engine-adapters.md). Probe actual Node, browser, FFmpeg/exporter, installed versions and local scripts; record command output/version/license evidence and missing capabilities. Use an existing compatible project when available; do not silently upgrade it. Select HyperFrames only within the verified 0.8.142 profile and its local dependencies. Select Remotion server only with matched 4.0.534 packages, browser and intended-use eligibility (P37). Its browser compositor is a separate route with its own supported DOM/media/codec subset (P23). Motion Canvas 3.17.2 uses its Vite editor and registered FFmpeg exporter (P25), not a fabricated headless render CLI. Preserve release/main and binary-license distinctions (P39). The Canvas/WebGL renderer is only the recorded source-local profile (P27), not a universal fallback to recreate or vendor. Do not buy services or assets. If production rights are unknown, retain the plan/prompt, use only eligible authorized work and report the unresolved eligibility. If no runnable eligible exporter exists, output its concrete gap and the achieved state. Output `engine_route`: engine/profile/version, environment evidence, eligibility and gaps.
Read the selected adapter’s exact contract and relevant [HyperFrames example](examples/hyperframes-prompt.md) or [Remotion example](examples/remotion-prompt.md). Include the complete director brief, shot table, style, timeline, asset ledger, audio/caption timing, dependency pins, file/output requirements and acceptance. Include only the selected adapter’s admitted API/CLI bindings. For HyperFrames, seconds, static finite root duration and completed paused registration apply (P16–P19). For Remotion, frame-derived state, zero-based local Sequence clocks and explicit clamps apply (P21). Audio fields use frames; Caption fields use milliseconds (P40). For Motion Canvas, generator `yield*` timing uses seconds and persisted event metadata (P25/P31). Never transpose APIs or clock units between engines. Specify an actual draft, relevant static/ren
Turn people, methods, and experience into installable, reusable agent skills. SkillAlchemy is an open-world agent skill creation system that turns underspecified skill briefs and open-world sources into installable, reusable agent skills.
LEAP builds skills through two pipelines: Branch A distills a skill from raw data, while…
Lens — Add a cognitive lens to any problem. It accepts a task description and produces an…
Cross-framework enhancement overlay for choosing a multi-agent topology BEFORE writing any…
SOP for terminal-based, git-native AI pair programming with Aider (git work-tree +…
Screens biomedical / life-science papers for signs of data fabrication, image manipulation,…
Universal discipline for any LM-driven loop — agent retries, plan-act-observe, multi-agent…