Skip to content

PRODUCT.md

Latest commit

 

History

History
64 lines (47 loc) · 2.54 KB

PRODUCT.md

File metadata and controls

64 lines (47 loc) · 2.54 KB
Raw

PRODUCT.md — web-tutorial-video

Written from the maintainer's brief without a live interview: the request was "make a landing page for the skill, auto do it as recommended." Items marked [assumed] were inferred and should be corrected if wrong.

What it is

An agent skill. Dropped into a coding agent's skills directory, it turns the prompt "make a tutorial video for this web workflow" into a finished, verified MP4 — instead of a hallucinated screencast.

Who it is for

Developers and technical writers who already run a coding agent (Claude Code, Codex CLI, OpenCode, or any agents.md-compatible harness) and need repeatable product walkthroughs. [assumed] Korean-first teams: narration ships through Supertonic Korean TTS by default.

The problem it removes

A generic agent asked for a screencast will describe a UI it never opened, guess at pacing, click the wrong control, and end on a frame that proves nothing. Three artifacts — the script, the timing sheet, the browser plan — drift apart by the second revision.

How it answers

  • One production contract, tutorial-plan.json, validated at three gates: planning, narrated, captured.
  • Verification before teaching: reconnaissance and a rendered rehearsal happen before a single frame is recorded.
  • Scene timing derived from measured TTS duration, never from guessed reading speed.
  • Side effects classified before rehearsal; consequential actions default to stop_before_commit. A tutorial request is not authorization to send, publish, purchase, or delete.
  • Machine QC plus a per-scene review contact sheet before anything ships.

What must be believed

That this one actually verifies. The landing page's job is to make the gates and the safety boundary legible in seconds — those are the difference, not the fact that it outputs an MP4.

Proof available

  • Nine scripts, a JSON Schema, an overlay asset, and an adversarial eval rubric, all in the repo.
  • The three validation commands, runnable and quotable verbatim.
  • The side-effect table from references/security.md.

Constraints

  • Ships as a static page on GitHub Pages: one HTML file, no build step.
  • [assumed] English-primary — the audience installing agent skills reads English docs; the narration being Korean is a feature, not the page language.
  • No fabricated metrics, no invented testimonials, no download counts.

Success

A visitor who has never heard of the skill can, within one screen, name what it produces and why it is trustworthy, and copy the install line without scrolling back.