Skip to content

majiayu000/spellbook / skills/build-product-demo

build-product-demo

Verified

Plan, produce, or diagnose evidence-backed product demo videos and screen-recorded promotional walkthroughs. Use when the user asks to make a product demo, launch video, feature showcase, app walkthrough, demo reel, or polished recording; wants every important capability shown without a slow feature tour; asks to remove dead time or fix narration-to-action pacing; or needs a repeatable script, capture plan, and verified final media. Do not use for fictional commercials with no real product proof, general-purpose video editing, or documentation-only tutorials that do not need a promotional narrative.

FileKindSize
SKILL.mdskill md13.0 KB
agents/openai.yamlconfig239 B
evals/evals.jsonconfig5.0 KB
references/artifact-contract.mdmarkdown6.8 KB
references/directing.mdmarkdown6.6 KB
references/production.mdmarkdown6.0 KB
scripts/analyze_demo_pacing.pyexecutablescript15.5 KB
scripts/probe_demo_media.pyexecutablescript6.8 KB
scripts/validate_demo_plan.pyexecutablescript16.3 KB

references/production.md

View on GitHub
# Product Demo Production

Use this reference before seeding data, recording product surfaces, generating
narration, or editing the final media.

## 1. Establish production truth

Record the exact repository, revision, executable or service version, runtime
configuration, and test data boundary. Classify every captured capability:

- **live**: the current product performs the action and produces the result;
- **deterministic**: a fixture or local provider stabilizes an input or external
  dependency while the real product path still runs;
- **composite**: multiple real captures are editorially combined;
- **title**: explanatory graphics that make no product-execution claim.

Do not label a bypass, static mockup, or manually edited result as live.

## 2. Use disposable state

- Prefer a temporary home, data directory, database, project, browser profile,
  or workspace.
- Seed only synthetic information. Do not record private memories, messages,
  credentials, filesystem history, notifications, or unrelated windows.
- Make setup rerunnable and explicit. Validate the seed before opening the
  recorder.
- Keep user worktrees and databases unchanged unless modification is expressly
  part of the request.
- Record cleanup instructions. Do not delete existing artifacts to make room.

## 3. Select the capture route

| Product surface | Preferred route |
|---|---|
| CLI or terminal | Repository script, VHS, asciinema, PTY capture, or direct command evidence |
| Browser-rendered app | Repository server plus Playwright or equivalent automation |
| Desktop-native app | App-owned automation or API first; bounded UI control only when required |
| API or service | Real request/response capture plus a legible presentation layer |
| Generated media | Preserve prompts, inputs, provider/model identity, outputs, and boundary frames |

Use the product's real interaction surface for the final proof. Lower-level
commands may prepare state or verify results, but should not replace a claimed
user-facing path.

For software promotion, target at least 60% of total duration on the native
surface and less than 20% on titles or explanatory composites. Record the
actual ratios in verification. If the proposition names an integration, record
the real consumer—not only the command that prepares its data.

## 4. Rehearse cheaply

Run a fast rehearsal without paid narration or full-duration holds. Verify:

- every command, selector, and endpoint resolves;
- the seeded state matches the opening frame;
- every beat produces the planned exit state;
- native product action begins by the declared first-action deadline;
- multiple observable events develop longer shots internally;
- the exact claimed consumer surface receives the result; storage, search,
  indexing, preview, and final delivery are separate gates;
- external waits have a bounded timeout and a visible failure;
- no secret or unrelated user data enters the frame;
- the estimated duration fits the delivery target.

After two empty long-running polls, continue independent work and inspect later.
Do not build a production loop around blind repeated retries.

## 5. Record narration

- Write narration after visible action and timing are stable.
- Say value, consequence, contrast, or context; let the screen show mechanics.
- Generate short per-beat files so one change does not invalidate the full mix.
- Audition the exact language with representative technical terms and product
  names. Reject pronunciation, accent, cadence, and energy mismatch.
- Measure every clip. A beat must accommodate its narration or be rewritten;
  truncation and silent fallback are failures.
- Keep narration, action sound, ambience, and music separable until the mix.

Do not infer a voice's accent or naturalness from its display name, locale, or
marketing label.

Duration, file integrity, and waveform presence do not prove a successful voice
audition. Require a listener decision for accent, pronunciation, and tone.

## 6. Handle waits honestly

When a real operation is slow, choose one explicit treatment:

- show a meaningful progress state;
- compress elapsed time and label the compression;
- use a sound or visual bridge while state continues to change;
- cut to evidence produced during the wait;
- redesign the demonstration around a deterministic local dependency;
- keep a motivated hold because reading or consequence is the event.

Never leave narration finished over an unchanged screen without a recorded
reason. Never fake completion because the provider or product path failed.

## 7. Edit from causality

- Preserve cause before consequence.
- Enter a shot late enough to remove setup noise and early enough to orient.
- Leave after the result registers, not after an arbitrary duration.
- Use hard cuts for clear state changes; use a transition only when it expresses
  time, location, comparison, or chapter change.
- Keep product text legible at delivery resolution and normal playback speed.
- Use music to shape the whole progression, not to disguise static footage.

## 8. Verify the delivery

Check mechanically:

- container, video codec, audio codec, dimensions, frame rate, and duration;
- presence of the expected video and audio streams;
- exact output path and nonzero file size.

Check editorially:

- opening hook and landing are legible;
- every core proof appears and produces a visible result;
- no unplanned gap, freeze, blank frame, clipped UI, or abrupt narration exists;
- cursor, highlights, captions, and crops point to the intended evidence;
- disclosure matches live, deterministic, and composite footage;
- no secret, personal data, or unrelated application surface is visible.

Use a contact sheet or full playback review. A successful `ffprobe` is not a
substitute for watching the sequence.

Run `analyze_demo_pacing.py` before sign-off. Its low-motion heuristic is not a
creative verdict, but a failing ratio or long segment is a blocking diagnostic
unless the beat plan identifies the interval as a motivated hold. Inspect dense
consecutive frames; scene-midpoint contact sheets are insufficient.
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