AGENT SKILL

suede-video

Plan or produce marketing video with scripts, storyboards, shot and asset lists, consent-aware AI or avatar choices, programmatic rendering, edit decisions, pacing, export specs, and video-search metadata. Use when asked to "turn this demo into a short video" or build a repeatable production pipeline. NOT FOR: social calendars (use suede-social), clip-to-guide bridges (use suede-clip-to-guide), paid ad concepts (use suede-ad-creative), still images (use suede-image), or unapproved publishing.

License 许可
Apache-2.0
最近更新
2026-11-14
Stars
★ 1.6k
Forks
491

01 前言 Front Matter

name
suede-video
metadata
{ "version": "2.2.1", "display-name": "Suede AI Video Skills" }
description
Plan or produce marketing video with scripts, storyboards, shot and asset lists, consent-aware AI or avatar choices, programmatic rendering, edit decisions, pacing, export specs, and video-search metadata. Use when asked to "turn this demo into a short video" or build a repeatable production pipeline. NOT FOR: social calendars (use suede-social), clip-to-guide bridges (use suede-clip-to-guide), paid ad concepts (use suede-ad-creative), still images (use suede-image), or unapproved publishing.

02 资源文件 7

03 SKILL.md 正文

Suede AI Video Skills

Compatibility

Planning works from supplied media and product evidence. Rendering, generation, avatars, and frame inspection depend on discovered authorized tools; otherwise return the documented manual editor handoff.

For a concrete input, expected handoff, and failure check, read the worked handoff.

This skill turns an approved message and rights-cleared source material into scripts, storyboards, generation plans, edit systems, and reusable production pipelines. It keeps likeness, licensing, brand assets, render cost, and publishing as explicit gates rather than assumptions.

Before Starting

Check for product marketing context first: If .agents/product-marketing.md exists (or .claude/product-marketing.md, or the legacy product-marketing-context.md filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.

Gather this context (ask if not provided):

1. Video Goal

  • What type of video? (Product demo, explainer, testimonial, social clip, ad, tutorial)
  • What's the target platform? (YouTube, TikTok/Reels/Shorts, website, ads, sales deck)
  • What's the desired length?

2. Production Approach

  • Do you need a human presenter? (AI avatar vs. voiceover vs. screen recording)
  • Do you have existing footage or assets? (Screenshots, logos, product UI)
  • Do you need generated footage? (AI-generated scenes, B-roll)
  • Is this a one-off or a template for repeated use?

3. Technical Context

  • What's your tech stack? (Node.js, Python, etc.)
  • Do you have API keys for any video tools?
  • Budget constraints? (Some tools charge per minute of video)

Choosing Your Approach

Choose the production method from the job, available evidence, rights, and currently callable tools:

Approach Candidate fit Tools to evaluate Use when
Programmatic Templated, data-driven, batch video An installed HTML/CSS or React renderer The format repeats and deterministic output matters
AI generation Original footage from text/image prompts Any currently authorized generation model The scene cannot be filmed practically and output rights are verified
AI avatars Approved presenter without filming Any currently authorized avatar platform Consent, likeness, voice, localization, and output rights are documented
Editing/repurposing Cutting long-form into short clips Any installed transcript or timeline editor Source media and derivative-use rights are available

Programmatic Video

Programmatic video is a candidate for repeatable, templated, or data-driven work; verify fit against the current stack, edit needs, cost, and review burden.

HTML/CSS renderer candidate

Hyperframes, from HeyGen, may fit an HTML/CSS workflow. Docs: hyperframes.heygen.com. First check whether it is installed. If it is, read the installed version's documentation, inspect its license, and confirm the renderer and export path. If it is not installed, use a renderer-neutral handoff or request explicit installation authorization after showing the current package source, version, lockfile impact, and cost. The following commands must not run without that authorization:

npx hyperframes init my-video && cd my-video
npx hyperframes preview               # live preview
npx hyperframes render -o output.mp4  # render index.html

Key concept: A composition is plain HTML, CSS, and JS. Timing is declared in the markup, and the renderer seeks a paused animation timeline frame by frame. Confirm the current composition syntax and any programmatic render API in the installed version's docs before writing code against it.

Candidate fit: Product announcements, changelogs, data-driven reports, and approved personalized outreach videos. Verify determinism with a repeat-render test instead of assuming it.

React renderer candidate

Remotion may fit a React-based workflow. Docs: remotion.dev. It is source-available, not open source, so check its license terms. Verify the installed or callable version, current documentation, license, rendering path, and hosting cost. The scaffold command may install packages or create files; run it only with explicit authorization for that change, otherwise provide a renderer-neutral plan:

npx create-video@latest

Key concept to verify: a composition is a React component that reads the current frame number and returns what is on screen; props drive content. Confirm whether rendering is local or uses an authorized hosted renderer.

Read references/programmatic-renderers.md when the repo already uses React, or when the choice between the HTML/CSS and React paths has to be justified: it carries the React component sample and the renderer selection table. Neither framework is the more powerful one. Both render web pages to video in a headless browser; the difference is what the team writes.


AI Video Generation

Generate original footage from text or image prompts. Use for B-roll, hero visuals, and scenes you can't practically film.

Model comparison

Discover currently callable and authorized models before recommending one. Read current official documentation and run a bounded test on the actual account. Compare:

Criterion Evidence to capture
Output fit Resolution, duration, audio, motion, and consistency from current docs plus a test clip
Control Reference-image, camera, edit, seed, and storyboard controls actually exposed
Operations Queue time, failure rate, export path, and repeatability
Rights Input rights, output rights, model restrictions, consent, and provenance
Economics Current plan, per-generation cost, failed-generation treatment, and budget cap

If no model is callable and authorized, return a shot list, prompt set, reference-frame brief, and manual vendor-evaluation checklist.

Prompting for Video Models

Good video prompts specify: subject + action + camera + style + mood

A close-up shot of hands typing on a laptop keyboard,
shallow depth of field, warm office lighting,
camera slowly pulls back to reveal a modern workspace,
cinematic color grading, 4K

Common mistakes:

  • Too vague ("a person working"): add specifics
  • Ignoring camera movement: specify dolly, pan, static
  • Forgetting style: "cinematic," "documentary," "commercial"
  • Requesting text in video: AI models struggle with readable text

For detailed prompting guides: See references/ai-video-prompting.md

AI Generation vs. Stock Evaluation

Use Case Generation evidence to test Stock/original evidence to test
Specific scene Prompt control, continuity, review burden Search coverage, license, edit fit
Style continuity Cross-clip consistency in a bounded test Match quality across licensed clips
Real location Factual accuracy and disclosure risk Provenance and location release
Product/brand UI and mark fidelity; avoid fabricated product claims Approved real captures or assets
B-roll Queue time, failure rate, cost Search and licensing time

AI Avatars

Create talking-head videos without filming. An AI avatar delivers your script with realistic lip-sync, expressions, and gestures.

Avatar tool evaluation

HeyGen is one possible avatar platform, not a guaranteed or preferred integration. Before naming it or another vendor, verify current official documentation, authenticated access, callable API or connector availability, plan limits, price, output rights, body-motion and script-ingestion controls, localization and export requirements, and the consent requirements for every voice or likeness involved.

If no avatar tool is currently callable and authorized, provide a manual vendor-neutral workflow: approved script, consent record, user-operated upload, export checklist, and a real-camera or voiceover fallback.

Possible fit to test: product explainers, feature announcements, multilingual versions, and approved personalized outreach.

Avatar vs. Other-Approach Test

Scenario Avatar evidence to verify Alternative to compare
Recurring content Consent, repeatability, audience response Recorded presenter or voiceover
Multilingual versions Translation, pronunciation, disclosure Human localization
Personalized outreach Consent, claim review, cost, reply quality Approved text or recorded template
Founder message Trust and disclosure response Direct recording
Product UI walkthrough Presenter value around real captures Screen recording
Creative/artistic video Control, originality, rights Original footage or generated scenes

Editing & Repurposing Tools

Turn existing content into multiple video formats.

Treat the named products as candidates. Discover what is currently installed or callable, verify the user's account and rights, then map the workflow to the available tool. If none is available, return an edit decision list, clip timestamps, captions, and export settings for manual execution.

Candidate capability What to verify Candidate fit
Transcript editor Speaker accuracy, cut control, export, privacy Interviews, podcasts, webinars
Clip assistant Selection controls, provenance, false-positive rate Long-form to short-form review
Timeline editor Caption, effect, audio-rights, and export controls Platform-specific finishing
Accessibility assistant Caption accuracy, eye-contact disclosure, dubbing consent Approved talking-head content

Repurposing Workflow

Long-form content (podcast, webinar, demo)
    ↓
Transcript edit: Clean up, remove filler, polish
    ↓
Clip review: Propose moments with source timestamps
    ↓
Timeline edit: Add reviewed captions, effects, and platform styling
    ↓
Distribute: TikTok, Reels, Shorts, LinkedIn

Reverse-Engineer a Viral Edit

To study the style of a reference edit, first discover whether an authorized media-inspection, browser, or local-file viewer is currently callable. If one is available, inspect the actual frames and timing. Otherwise ask the user to attach the media, screenshots, or a timecoded export and provide a manual frame-sampling checklist. Then extract a reusable beat sheet plus the 3–5 signature moves. Review it before execution in any currently available editor. Copy the editing grammar, never the reference's footage, script, voice, or music. Full method: references/edit-anatomy.md.


Video Production Workflows

Product Demo Video

  1. Script the key features and value props (use suede-copy skill)
  2. Screen record the product flow
  3. Programmatic overlay: use an available authorized renderer for titles, callouts, and transitions, or provide manual editor instructions
  4. AI B-roll: only with a discovered authorized generator and verified rights; otherwise use licensed stock or original footage
  5. Voiceover: record yourself or use AI avatar for narration
  6. Export at platform-appropriate specs

Explainer Video

  1. Script the problem → solution → CTA arc
  2. Choose presenter: an approved avatar workflow when available, or recorded voiceover plus visuals
  3. Build visuals: programmatic slides, screen recordings, AI-generated scenes
  4. Add captions as an accessibility default and verify their accuracy
  5. Export only after checking current destination requirements and previewing the authenticated composer; do not assume a universal orientation

Batch Social Clips

  1. Create master template in an available authorized renderer or a manual editor project
  2. Feed data: product features, testimonials, stats
  3. Render batch: one template, many variations
  4. Add platform-specific captions with an available editor
  5. Return approved export candidates to suede-social for channel strategy; scheduling remains a separate explicit action

Personalized Video per Account

A tailored product film for each target account, built by agents. One-off prompting reaches a decent film after 15 to 20 minutes of back-and-forth. Hundreds of films need the foundation built first.

  1. Product component library: agents walk the app and rebuild key screens as reusable components, one source of truth every film pulls from instead of redrawing.
  2. Motion primitives: reusable animated parts (camera pushes, screen handoffs, click rhythm, product moves). Fix one and every film inherits the fix.
  3. A menu of demo flows: company-agnostic moments buyers ask to see, each a short list of beats.
  4. Per-account script: an agent drafts it from call notes, CRM win/loss data, and account research, with a fictional company on screen. A cloned narrator voice requires that narrator's documented consent.
  5. A render engine: code that turns a beat list into a film, covering camera, transitions, captions, pointer timing, and end card.
  6. Automated QA: render contact sheets and flag clipped text, small type, and stray pointers before a person watches.
  7. One tracker: a row per film (account, flow, status, links) that agents update as they build, export, and deliver.

Storyboard the first film before production; it sits between the brief and the render the way a spec sits between an idea and code. Expect tool limits at volume (one agent per file, daily API caps), so run films in parallel lanes and batch overnight. Any authorized programmatic renderer (Hyperframes, Remotion) or a design tool with agent access can carry it; none is the default. Rendering, voice cloning, and delivery each stay separate approvals under the Halt Contract.

Source: adapted from Harvey's write-up of its agent-built launch and per-account films (Shawn Farsai and Gabe Pereyra, October 2026).


Tool-Aware Video Pipeline

Use automation only after discovering currently callable, authenticated, and authorized tools. Otherwise produce the same artifacts as a manual handoff.

Approved script (from product context)
    ↓
Available renderer/avatar/generator, if authorized
    or manual editor with timecoded instructions
    ↓
Review cut, rights, captions, and brand assets
    ↓
Output: Approved export candidate, not automatically published

When no production tool is available, return the script, shot list, asset manifest, beat sheet, captions, edit decision list, and export settings so the user can execute the cut manually, in these exact headings:

## Script
<scene or beat> | on-camera or VO line | duration target

## Shot List
Shot # | source (screen capture / owned footage / licensed / generated) | framing | notes

## Asset Manifest
Asset | origin | rights basis and holder | expiry or attribution required | file path

## Beat Sheet
Use the row schema in references/edit-anatomy.md (Beat | Time | Shot |
On-screen text | Caption style | Transition/motion | Audio) plus its style
summary. Do not invent a second beat-sheet format.

## Captions
Full caption text with timecodes; state the accuracy check performed.

## Edit Decision List
Clip # | source file | in TC | out TC | placement | transition | audio bed

## Export Settings
Destination | ratio | resolution | frame rate | codec | max duration | verified
against (current spec source + date)

## Open Gates
<missing rights, consent, cost approval, or brand asset; use the halt contract>

For the beat sheet and its style summary, read references/edit-anatomy.md before writing the handoff: it owns that row schema.


Export Specs and Video Search

Verified platform facts (October 2026):

Destination Fact Source
YouTube Shorts Up to 3 minutes for square or vertical uploads made on or after Oct 15, 2024. Over 1 minute with an active Content ID claim, the Short is blocked worldwide. https://support.google.com/youtube/answer/15424877
Instagram Reels Up to 3 minutes; profile grid previews crop to 3:4, so keep the subject inside a 3:4 safe area of the 9:16 frame Instagram announcement, January 2025
Performance Max Supply your own 16:9, 1:1, and 9:16 videos of 10 seconds or more, or Google may auto-generate one https://support.google.com/google-ads/answer/17091269

For other destinations, read the current upload spec and preview the authenticated composer before choosing ratio, resolution, or duration.

Make the video findable. A video that answers a question people search can earn views for years on YouTube and in Google's video results. For any video meant to be found:

  1. Title: the search phrase and the promise, phrase first, under about 70 characters so it is not truncated in results.
  2. Description: the first two lines restate the answer in plain words; the rest adds links and context.
  3. Chapters: timestamps starting at 0:00 with descriptive names, which Google can show as key moments.
  4. Captions and transcript: upload accurate captions instead of relying on auto-captions, and publish the transcript on the page that embeds the video.
  5. Embed on an owned page with VideoObject JSON-LD (name, description, thumbnailUrl, uploadDate, duration, contentUrl or embedUrl) and the video visible on load, so the page itself can rank and be cited in AI answers.
  6. Thumbnail: one readable idea at small size; real product UI, never a fabricated one.

When the video must lead viewers to a written guide, route to suede-clip-to-guide. For site-level video indexing and schema audits, route to suede-seo-audit.


Common Mistakes

  1. Starting with tools, not strategy: decide what video you need before picking tools
  2. AI-generated text in video: models can't reliably render readable text; use programmatic overlays instead
  3. Unverified avatar quality: test a short consented sample before paying for or producing a batch
  4. Inaccessible cut: provide accurate captions without asserting a universal sound-off viewing rate
  5. Unverified export spec: read current destination requirements and preview the actual composer before choosing ratio, resolution, or duration
  6. Unsupported production-style claim: compare polished and informal treatments on the current account instead of claiming one universally wins

Halt Contract

Use this exact format when tool authorization, cost approval, rights, consent, or an approved brand asset blocks the requested result, including the install and scaffold gates above and the brand-mark gate in Boundaries:

HALT: <one-line blocker>
Why it blocks: <specific missing authority or evidence>
Resolve with:
1. <option>
2. <option>
3. <option, when useful>
Waiting for: <the exact item or approval>

Continue with the manual-handoff artifacts above only when they remain useful and do not imply the blocker was resolved or that a render occurred.


Boundaries

  • Do not generate, render, purchase, upload, or publish media without approval of scope, tool cost, rights, and destination.
  • Do not clone a real person's voice or likeness, use unlicensed source media, or promise an exact copy of a reference style or edit.
  • For Suede-branded visuals, use only the approved transparent Suede S asset at JasonColapietro/suede-creator-skills/docs/assets/suede-ai-logo-transparent.png with SHA-256 83a7ee0317e4debe2e7b076c20ba067feb76a587f9e829dc6310ae4be4b44dfa; if it is missing or changed, omit the mark and halt for the approved file.

Routing

  • Use suede-social to choose channel strategy, cadence, and distribution.
  • Use suede-clip-to-guide when a clip has to bridge to an article, newsletter, or guide.
  • Use suede-ad-creative for paid video ad concepts and variants.
  • Use suede-image for still-image generation and editing.
  • Use suede-copy for source-verified script and messaging polish.

04 同仓库更多 Skills

hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-review-with-annotations

Create, inspect, reply to, update, or resolve situated Worktable annotations. Use when feedback belongs to a Doc section, block, text range, or HTML Doc, or when the user asks to address existing review feedback and coordinate a requested artifact edit.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-find-and-synthesize-context

Find, assess, and synthesize context from Worktable Docs and Records. Use when the user asks what a workspace contains, wants workspace-backed research, needs current facts reconciled across artifacts, or requests a cited synthesis without changing workspace state.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-update-docs

Create, edit, rename, or maintain narrative Worktable Docs. Use for plans, decisions, briefs, notes, research syntheses, and other durable prose when the user requests a new artifact or a safe update to an existing one.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-records

Organize repeated structured information as durable Worktable Records. Use for research datasets, inventories, directories, pipelines, trackers, and other collections whose items need identity, fields, filtering, links, validation, or individual updates.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-create-or-manage-interactive-html

Create or improve durable visual and interactive artifacts in Worktable. Use for dashboards, trackers, calculators, mini apps, slide presentations, and other requests that benefit from a polished HTML Doc.

★ 1.6k
hashgraph-online
hashgraph-online/awesome-codex-plugins

worktable-collaborate-threads

Read, answer, and hand off messages through durable Worktable threads with users or other authenticated participants. Use when the user asks to check addressed messages, reply to a participant, continue a thread, request another agent's help, wait briefly for a response, or watch for comments and messages sent to the agent.

★ 1.6k