SKILL: Product Tour Capture

Last used: 2026-03-24 Memory references: 1 Status: Active Version: 1.0 · Author: Loki · Created: 2026-03-24 Proven on: Reddi Agent Protocol (14 steps, agent-protocol.reddi.tech/tour)

What this skill does

Captures a series of Playwright screenshots from a live web app, builds a full-screen clickable slideshow page (/tour), and deploys it. Gives any web app a polished "product tour" in one playbook run.

Inputs

interface TourPlaybookInput {
  appRepo: string          // local path to Next.js repo
  baseUrl: string          // deployed app URL (https://...)
  outputDir: string        // public/tour/ relative to repo
  tourPagePath: string     // app/tour/page.tsx relative to repo
  steps: TourStep[]
}

interface TourStep {
  id: string               // e.g. "01-landing" — used as filename
  url: string              // path on the app, e.g. "/"
  title: string            // sidebar label
  caption: string          // 1-sentence description shown in tour
  scroll?: number          // pixels to scroll after load
  clickTab?: string        // text of a tab to click (shadcn Tabs)
  click?: string           // text of element to click
  fill?: string            // text to fill into first textarea
  wait?: number            // ms to wait after fill/click
}

Outputs

interface TourPlaybookOutput {
  screenshotsCapured: number
  screenshotsFailed: string[]   // IDs of failed captures
  tourUrl: string               // live URL of the tour page
  deployedAt: string            // ISO timestamp
}

Multi-agent playbook

Agent roles

Handoff pattern

Loki constructs TourPlaybookInput
  → spawn Kit (capture)
      → node scripts/capture-tour.mjs
      → returns: { captured: 14, failed: [], manifest: [...] }
  → spawn Kit (build)
      → writes app/tour/page.tsx from TOUR_STEPS template
      → npm run build → vercel deploy
      → returns: { tourUrl, httpStatus: 200 }
  → spawn Oli (QA)
      → checks each screenshot isn't blank (file size > 5KB)
      → spot-checks 3 captions match visible page content
      → returns: pass / conditional-pass / fail

Capture script template

Location: scripts/capture-tour.mjs in the target repo.

Key patterns:

• page.goto(url, { waitUntil: 'networkidle', timeout: 20000 })
• Tab clicks: page.getByRole('tab', { name: tabName }).first()
• Scroll: page.evaluate(y => window.scrollTo(0, y), scrollY)
• Fill + trigger: textarea.fill(text) → button[generate].click() → waitForTimeout(ms)
• Always waitForTimeout(600) after navigation before screenshot
• Viewport: 1280 x 800 (16:10, clean for sidebar layout)

Tour page template

Full-screen layout (no scroll):

• Top bar: "Product Tour · X of N" + auto-play toggle + "Try it live →" CTA
• Main: screenshot panel (with prev/next chevrons) + right sidebar (step info + list)
• Progress bar: Solana purple → green gradient

Template component lives in app/tour/page.tsx. TOUR_STEPS is the only thing that changes per app.

When to use this skill

• Any web app that needs a "product tour" page for demos, hackathons, or onboarding
• After a major UI rebuild (re-capture to keep tour current)
• For hackathon submissions — judges can self-tour without a live demo session

Lessons learned (from Reddi Agent Protocol run)

1. Tab clicks need getByRole('tab') — generic getByText can hit non-tab elements
2. Demo pipeline screenshots: fill + click + wait(7000) for mid-pipeline, wait(13000) for complete
3. 23 files in public/tour/ after a re-run — old screenshots persist, that's fine (overwritten)
4. Build command: use node node_modules/next/dist/bin/next build not npx next build — the .bin/next shim has path issues on Node 24 in some environments
5. Auto-play at 4s is the right pace — slow enough to read captions, fast enough to feel dynamic
6. 14 steps is the sweet spot — fewer feels thin, more feels exhausting for a tour

Playbook architecture notes

This skill is a good test case for the playbook architecture because:

• Clear 3-agent handoff with defined I/O at each step
• Capture agent is stateless (just runs a script)
• Builder agent has a fixed template, only data changes
• QA agent has a simple pass/fail rubric (file size + caption spot-check)
• Total runtime: ~5-8 minutes end-to-end
• Fully parameterisable — works for any Next.js app with any step list

Future extensions

• v2: Accept a Figma URL or sitemap and auto-generate the step list (Archie researches the app first)
• v2: QA agent uses vision model to verify screenshot content matches caption
• v3: Video export — stitch screenshots into a 60s walkthrough video (Finn)
• v3: Embed tour as a modal on the landing page (not just /tour route)