AGENT SKILL

product-video

Build or edit deterministic product walkthroughs, SaaS demo videos, feature tours, launch clips, narrated browser tours, and captioned screen recordings. Use whenever a user asks to record a product flow, turn app screens into a video, create a walkthrough/demo/promo, improve an existing product video, or mechanically verify a rendered MP4—even when they do not mention Remotion.

License 许可
MIT
最近更新
2026-11-15
Stars
★ 1
Forks
0

01 前言 Front Matter

name
product-video
description
Build or edit deterministic product walkthroughs, SaaS demo videos, feature tours, launch clips, narrated browser tours, and captioned screen recordings. Use whenever a user asks to record a product flow, turn app screens into a video, create a walkthrough/demo/promo, improve an existing product video, or mechanically verify a rendered MP4—even when they do not mention Remotion.
compatibility
Requires Node 24, the repository pnpm toolchain, FFmpeg/FFprobe, and an installed Chrome-compatible browser. Required narrated OCR verification additionally needs Tesseract. Authenticated capture additionally requires the existing wsl-cdp profile and the repository-pinned Playwright capture runtime; the skill never installs browser tooling at runtime.

02 资源文件 6

03 SKILL.md 正文

Product video

Produce product videos from source-traceable product material. A polished render is not proof of a correct walkthrough: preserve the real interface, keep claims tied to evidence, and verify the final media artifact mechanically and visually.

Route the request

  1. New ProductTour or promo: read workflow.md.
  2. Narrated browser tour: read workflow.md, capture.md, and qa.md. Use NarratedBrowserTour V1 rather than modifying the legacy ProductTour composition.
  3. Authenticated browser capture: read capture.md before opening or attaching to a browser.
  4. Existing-video timing or composition edit: preserve the accepted story and edit only the requested layer; then use qa.md.
  5. Render or verification only: use qa.md.
  6. Music, external publishing, or third-party generation: stop at the boundary in effects.md. Do not improvise credentials or publish. Narration is in scope — synthesize it with narrated:voice using the operator's own provider key, never by cloning a real person's voice without consent.

Non-negotiable boundaries

  • Use real product captures for real product screens. Rebuild only an explicitly stylized scene or one isolated element that must animate.
  • Use either committed fixtures or the intended walkthrough recipient's own read-only account through the product's supported impersonation flow. Never substitute demo data when the request is recipient-specific. Before recording, verify that the viewport contains only recipient-visible product state and no credentials, unrelated customer data, private communications, internal-only values, or secrets. V1 does not redact media or create a human-review/redaction receipt.
  • Never automate login, MFA, CAPTCHA, password entry, token entry, or account recovery. Attach only to the dedicated profile after the operator has authenticated it.
  • Narrated live capture is loopback-CDP only, fail-closed, and dry-run first. Persist only normalized cursor/click/element geometry/scroll/declared-route telemetry—never URLs, selectors, DOM text, storage, requests, or account data.
  • Never install or update third-party skills, CLIs, browser extensions, or providers at runtime. Use only dependencies pinned and reviewed in the repository, and follow UPSTREAM.md for provenance. Do not install the Playwright CLI's bundled skill instructions.
  • Drive animation from frames. Do not use CSS transitions, CSS keyframes, clocks, or unseeded randomness in rendered compositions.
  • Refuse zero-scene plans and zero-frame renders. A successful process exit without content is a failure.
  • Render locally by default. Sending or publishing the result requires explicit authorization for that destination.

Commands

Run commands from the repository root through the workspace package:

pnpm capture --help
pnpm render --help
pnpm verify --help
pnpm narrated:capture --help
pnpm narrated:prepare --help
pnpm narrated:studio --help
pnpm narrated:render --help
pnpm narrated:verify --help

Scripts are black boxes: use --help before reading their source. The legacy fixture smoke path is test:render. The narrated local-only synthetic smoke path is:

pnpm narrated:smoke

It creates ignored neutral media, renders it, and verifies captions with required OCR. A missing FFmpeg, FFprobe, Tesseract, or Chrome-compatible browser is a non-pass, not a skipped pass. It never attaches to CDP or captures product/customer data.

Narrated completion contract

A delivered narrated render includes:

  • the run ID, MP4 path, actual duration, and resolved V1 manifest path;
  • a finalized capture bundle whose source video and telemetry exactly bind the resolved manifest;
  • FFprobe evidence for H264, dimensions, yuv420p, duration, size, decoded frame count, audio stream, and 48kHz sample rate;
  • marker, semantic telemetry, and caption midpoint samples plus the contact sheet;
  • required caption OCR status when captions are expected;
  • any skipped or non-pass check and its concrete reason;
  • the narration source: operator-recorded audio, or the synthesized voice ID and cue count from narrated:voice;
  • confirmation that no upload or publishing occurred unless it was explicitly requested and separately authorized.

public/generated/narrated/<run-id>/ and out/narrated/<run-id>/ are local, ignored, run-owned artifact roots. Do not repoint a run to a different capture or overwrite outputs outside those roots.

04 同仓库更多 Skills

StartupBros-com
StartupBros-com/hov-stack

review-adversarial-first

Operator-invoked adversarial review: 3 refute-by-default lenses plus a deterministic judge against a PR, branch, or path, then a decision on whether /ce-code-review's topic fan-out runs. Off the default merge path since the 2026-10-17 merge-and-end ruling; run only when asked for by name.

★ 1
StartupBros-com
StartupBros-com/hov-stack

memory-mine

Run one deterministic, budgeted directed-capture batch for an explicit Claude Code project store. Use manual workspace mode only when cross-agent cass sessions are intentionally in scope.

★ 1
StartupBros-com
StartupBros-com/hov-stack

papers

Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.

★ 1
StartupBros-com
StartupBros-com/hov-stack

papers

Search preprints from arXiv, bioRxiv, and medRxiv without API keys. Supports search, lookup, extraction, citations, enrichment, and paper downloads that bundle PDFs plus linked GitHub code.

★ 1
StartupBros-com
StartupBros-com/hov-stack

wsl-cdp

Drive the user's real, logged-in Windows browser (Chrome, Edge, or Brave) from a Claude Code session running inside WSL2, over the Chrome DevTools Protocol - read authenticated dashboards and consoles, take real screenshots, open and read pages with the user's sessions. Use when a task needs the user's existing web logins and the session runs in WSL2, where the Claude in Chrome extension cannot work. Not for native-Windows or macOS Claude Code (use the official Claude in Chrome extension), not for anonymous scraping or CI (use headless Playwright inside WSL), not for getting past login walls (those are handed to the human).

★ 1
StartupBros-com
StartupBros-com/hov-stack

token-eater

[BETA] Spend a subscription's idle, about-to-expire credits on safe, machine-verified cleanup. You pick the service (Grok / Codex / Claude) and a token-heavy maintenance skill (de-monolithize, de-slop, simplify, dead-code); that service runs the whole skill -> gate -> self-review -> draft-PR loop on its own credits, in an isolated worktree. token-eater verifies the gate and never merges. Manual invocation only during beta.

★ 1