Content & Media Creation · heygen-com/hyperframes
general-video
Author or edit a custom HyperFrames composition when no specialized workflow fits, or when BRIEF.md sets flow: companion. Use for longer or multi-scene pieces, brand and sizzle reels, montages, static loops, static title cards, footage remixes, and freeform builds. Use motion-graphics instead for a short unnarrated motion-first unit, including an animated title. Route fresh creation through hyperframes before using this skill.
L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandNetwork AccessWrite FileDual-pass Audit
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.
1How It Works & Mechanism
Positioned as a "fallback workflow": used only when no specialized workflow matches, or when BRIEF.md specifies flow: companion; all fresh creation must first route through the /hyperframes intent layer.
skills/general-video/SKILL.md
Route fresh creation through hyperframes before using this skill.
What is happening here: This skill is one card beneath the HyperFrames router (/hyperframes) and does not take on "which workflow to select" decisions itself. The trigger surface is intentionally narrowed to prevent the model from heading straight to fallback execution whenever it sees "make a video".
Perform state-machine evaluation before starting work: take the first match across four priority rows ("Specific edit / Has BRIEF / Has hyperframes.json or STORYBOARD / Fresh creation") and cease further evaluation.
skills/general-video/SKILL.md
Apply the first matching row; do not evaluate lower state rows:
What is happening here: Eliminates ambiguity between "continuation" and "restarting"—specific edits bypass discovery interviews; existing BRIEFs skip further questioning; only genuine ground-zero creation loops back to the intent layer. Each row in the table provides actions and explicit constraints ('Do not reopen discovery.', 'Ask no brief questions.').
Cross-cutting capabilities (media, Figma) are not implemented internally, but delegated to sibling skills via adapters: any audio/image/icon/logo/voiceover/color grading/LUT/VFX/subtitles must first read resolve + setup-providers in /media-use.
skills/general-video/SKILL.md
**Media:** For any audio, image, icon, logo, voice, grade, LUT, treatment/effect, caption, or media-operation need, load `/media-use` and follow `../media-use/references/resolve.md` (resolve, adopt, reuse) and `../media-use/references/setup-providers.md` (providers, auth).
What is happening here: Asset procurement, licensing, and brand asset handling have a single owner (/media-use); this skill only manages "when to invoke them". Figma inputs similarly enforce delegation to /figma, for reasons stated in the body: bypassing SVG sanitization, media provenance, and brand token binding.
Runtime form is derived from the flow/storyboard fields in BRIEF.md (automation/companion × yes/no), with explicit prohibition against inventing synonyms.
skills/general-video/SKILL.md
Do not invent synonyms for these states. An ongoing “just build it” signal is handled by the intent layer and arrives as `flow: automation`, `storyboard: no`.
What is happening here: Gate behaviors (whether to open a storyboard, whether to halt and ask) are determined by mode, which is derived from brief rather than improvised on the spot by the model—the core mechanism of "controlled automation" across the HyperFrames family.
The composition contract is a hard constraint (class="clip" + single paused seek-safe timeline + render-time determinism), sharing the same deterministic discipline as the keyframes family.
skills/general-video/SKILL.md
Timed elements use `class="clip"`; the root and relevant ancestors are sized; each composition registers one paused, seek-safe timeline on `window.__timelines`; rendering is deterministic. Do not use render-time network fetches, clocks, or unseeded randomness.
What is happening here: The renderer seeks on a per-frame basis, strictly prohibiting render-time network requests (which introduce non-deterministic outputs due to network timing), clocks, and unseeded randomness. This contract is a shared constraint across all 8 workflow skills.
The only included script is a thin wrapper around the packet builder: pinning down only this workflow's path and one behavioral distinction (design truth resolution order: frame.md → design.md → DESIGN.md), while reusing hyperframes-core shared implementations for all other logic.
skills/general-video/scripts/frame-packets.mjs
// Thin wrapper over the shared packet builder in hyperframes-core — this file pins // this workflow's paths plus its one behavioral difference: design truth resolves // frame.md → design.md → DESIGN.md (general-video § 6 order). Everything else has // one owner: ../../hyperframes-core/scripts/lib/frame-packets-core.mjs
What is happening here: Explicit "single ownership" convention—shared build logic cannot be forked inside individual workflows. Actual disk writes (mkdirSync + writeFileSync for per-frame packets and _role.md) occur within the core implementation, generating `.hyperframes/frame-packets/` under outDir.
Sub-agent dispatch carries economic thresholds and quantitative rationale: inline execution is faster for small clips; fan out only at scale, and dispatch all workers in a single wave.
skills/general-video/SKILL.md
**Dispatch pays for itself only at scale.** Authoring packets and warming fresh worker contexts costs real minutes and tokens: a film of up to ~6 short scenes builds FASTER inline, in this context, one scene after another (measured: 5 short scenes ≈ 9 min inline vs ≈ 21 min packetized).
What is happening here: Overturns the intuition that "parallel is always faster" using empirical data (packet assembly and warming context carry fixed overhead), requiring 2–3 scenes per worker and single-wave dispatch (a second wave nearly doubles the time window). This is the key differentiator from generic "one-agent-per-frame" designs.
2Core Capabilities
01Freeform authoring and editing when no specialized workflow fits (multi-scene pieces, brand reels, montages, static loops, title cards, footage remixes)
02companion mode: adopt the posture of a "director rather than a contractor", providing a ceiling-level proposal for the user to subtract from
03Timeline-level editing guidance (hard cuts, trims, splices, reordering)—multiple instances of the same asset + data-media-start/data-start/data-duration
04Frame packet builder: generate bounded packets (scene blocks + blueprint content + reference rules) and complete role files (_role.md) for dispatched workers
05Secondary artifact contract for frame-worker output: must provide motion.json (duration + in/out vectors) alongside HTML to allow upper layers to stitch scene transitions
06Self-refresh and borrowing discipline: execute `skills update` before relying on dependencies; borrow only story shapes and taste when drawing from other workflows, not their private scripts or directory contracts
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| cli | hyperframes CLI（init / check / lint / snapshot / preview / render / skills update） |
| cli | hyperframes skills update（自刷新，需网络拉取最新 skill 集） |
| cli | media-use 的 prefs.mjs / recipe.mjs（偏好记录与配方应用，本地写盘） |
| package | hyperframes-core 共享库（frame-packets-core.mjs，本 skill 脚本的唯一实现来源） |
| package | 兄弟 skill 依赖（/hyperframes 路由、/hyperframes-core、/hyperframes-creative、/hyperframes-animation、/hyperframes-keyframes、/media-use、/figma） |
4Recommended Governance & Evidence
Suggested Controls for Safe Operation
For Agent Runtime
Graded/per-use authorization, sandbox recommended
For Administrators
Approval workflow, rollback support, full audit
Evidence & Triggers (File Paths & Excerpts)
- scripts/: Execute local scripts or commands: node scripts/frame-packets.mjs --project … --storyboard … (local Node script; via hype
- 网络端点: Declare external network access: npx hyperframes skills update general-video (command semantics imply fetching skill sets from remote; this skill's own scripts
- 本地工作目录: File I/O and artifact output: .hyperframes/frame-packets/* (per-scene packet) and _role.md (via frame from hyperframes-core
Audit Notes: Behavioral scope: Local Node scripts + external hyperframes CLI, performing only local file read/write (generating packets, recording preferences, project files), with no credential retrieval (no .env / API key / keychain / token access), and zero outbound data transfer (zero fetch/network APIs inside scripts). The only network actions are the workflow-mandated `npx hyperframes skills update` and subsequent render/capture CLI calls (triggered in other skills), communicating solely with the official CLI itself. Note: When dispatching sub-agents, full role files and packets must be passed to workers, effectively introducing host sub-agent permissions into the build process—read and write boundaries must be enforced by the host.
5Second-Pass Verification
- [ok] 唯一脚本是否真的『薄』（有无隐藏功能/网络） — frame-packets.mjs 全文 37 行，仅 import existsSync/path/url 与 core；导出两个函数并透传 CONFIG；无 fs 写操作、无 fetch、无 env 读取。写盘与模板拼装在 core 中，符合注释自述。
- [ok] 网络 / 凭证 / 注入面 — skill 内 0 处 URL、0 处 fetch、0 处 process.env、0 处凭证关键词；grep 'heygen' 命中的两行是 Figma 段把 figma.com 当输入 URL 的说明与派发成本段，不是 API 调用。第二遍未发现漏掉的网络或凭证路径。
- [ok] 『外部资源』条目逐条回查调用点 — npx hyperframes init/lint/check/snapshot/preview/render/skills update 均在 SKILL.md 正文出现；prefs.mjs/recipe.mjs 路径在 §2 明确定义 <MEDIA_DIR> 为已安装的 /media-use skill 目录；frame-packets-core.mjs import 在脚本原文可验。
- [ok] frame-worker 输出契约与编排者所有权是否自相矛盾 — frame-worker.md 要求 worker 写两个文件（html + motion.json）并禁止写 <audio>、禁止碰邻居场景文件；SKILL.md §5 第 5 步要求编排者收集 motion.json 并在装配时把时长/进出场向量带入缝合。两侧互补、无冲突。
- [ok] 组内一致性（packet 构建路径与自刷新提示） — 与 product-launch-video / faceless-explainer 对比：三者都调 `node <SKILL_DIR>/scripts/frame-packets.mjs`，但 general-video 的 worker 拿到的是『2–3 场景』而另两者是『一帧一 worker』——这是有意的差异（本 skill 面向长片），已在 traits 注明而非记为出入。
6Verdict & Scenarios
Best for: Suitable for users who have a HyperFrames environment and need to produce long-form or multi-scene pieces without a specialized template: brand/sizzle reels, montages, footage remixes, static loops and title cards, or freeform builds. Also fits users who explicitly choose companion mode, where this skill first delivers a comprehensive ceiling proposal (story arc, design specs, per-scene motion, transitions, audio identity, opening/closing scene designs) for the user to subtract from.
Not for: Not suitable for: (1) Tasks that match a specialized workflow—flagship product showcases route to /product-launch-video, pure text explainers to /faceless-explainer, short unnarrated motion to /motion-graphics (explicitly named in description); (2) Starting fresh creation directly from this skill (should route through the /hyperframes intent layer first); (3) Environments missing hyperframes CLI and sibling skills like /media-use; (4) Users expecting zero-interaction one-shot output—even automation mode includes check/preview gates and render approval; (5) Scenarios requiring asset procurement, voiceover, or color grading capabilities directly—those belong to /media-use.
Install Agent-Ready
① Verified Mirror 2026-09-15
Option A · Direct DownloadDownload general-video.tar.gz
sha256:
Option B · Copy Guide for Agent4d140e0c0bd3b0ae…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit b8328f9573.