pitch

name	pitch
description	Run a live investor/demo pitch where TerranSoul (the pet) narrates a web slide deck aloud. The pet starts hidden; when the presenter clicks 'Click to reveal!' on slide 4, it appears, thanks Darren, introduces itself, and presents slides 5–9 — including a 30s walk through its human-brain architecture and the by-the-numbers comparison.
argument-hint	[event slug, default: southwest2026]
compatibility	Windows/macOS/Linux. Auto-installs recommended setup (incl. TTS/voice) and runs TerranSoul with CDP on :9222 via `npm run dev:recommended-setup`. Needs Playwright + chromium and Ollama.
user-invocable	true
disable-model-invocation	false

What this does

A two-presenter pitch:

• Darren presents slides 1–4 by hand (arrow keys / click).
• On slide 4 he clicks "Click to reveal!" → the hidden TerranSoul pet appears, says "Thanks, Darren!", introduces itself, and presents slides 5–9 on its own — speaking each line aloud (synced mouth + emotion) while the deck advances in lock-step.
• Slide 7 ("Why Me?") embeds the live brain-architecture diagram; the pet highlights each region (Focus, Graph, Recall, Memory, Affect, Skills, the LLM core and the MCP nerve) for a ~30s walkthrough, then pitches the intention: better memory + knowledge representation, retrieval, and updation — a different problem from Claude Code / OpenClaw / SIA, aimed at memory-heavy read+update work, and able to serve as external memory shared with other LLMs.
• Slide 8 is the combined comparison (COMPARISON.md + SELF-IMPROVE-COMPARISON.md) rendered as charts; the pet highlights the numbers.

Default event slug is southwest2026. Pass a different slug to present another deck under docs/pitch/<slug>/.

Moving parts already in the repo:

• Deck: docs/pitch/<event>/index.html + deck.css + deck.js (exposes window.TSDeck).
• Narration: docs/pitch/<event>/pitch-script.json (ordered beats).
• Pet bridges: window.__tsPitchControl (enter pet mode / hide / show) and window.__tsPitch.say() (speak a line + caption, resolves when audio ends).
• Driver: scripts/pitch/present.mjs (CDP orchestration).

---

Step 0 — Auto-install all recommended setup

Per the project rule, fix/install rather than work around. This makes the demo runnable from a clean machine, including the brain models and TTS/voice.

1. Run the prerequisites installer and let it install anything missing:

   • Invoke the /setup-prerequisites command (Node ≥ 20, Rust, Tauri CLI, WebView2 on Windows). Re-verify after install.

2. Ensure the Playwright browser is present (the driver launches the deck full-screen):

   npx playwright install chromium
   

3. The recommended models are installed automatically — in two aligned places, both running the same work as the first-launch "Recommended + Auto-accept all" popup (no manual clicking):

   • npm run dev:recommended-setup (Step 1) → App.vue runAutoRecommendedSetup runs it headlessly on startup and defers app-ready until it completes: brain.autoConfigureLocalFirst({ targetModel: 'gemma4:12b-it-qat' }) (ensures Ollama, pulls/activates gemma4:12b-it-qat chat and embeddinggemma embedder — the backend PREFERRED_EMBED_MODEL) + voice.downloadSupertonic() (on-device Supertonic TTS voice).
   • scripts/pitch/present.mjs (Step 2) re-runs the same via scripts/pitch/ensure-setup.mjs as a safety net — idempotent and fast when the app already configured it.

   A first-run cold model pull can take several minutes; app-ready waits for it. Skip the presenter's safety-net pass with TS_PITCH_SKIP_SETUP=1; override the chat model with TS_PITCH_CHAT_MODEL. Both paths fail-open: if a step fails, the pet still speaks via the browser-speech fallback with captions, so the demo never blocks.

---

Step 1 — Launch TerranSoul with recommended setup + CDP

In a separate terminal (keep it running):

npm run dev:recommended-setup    # auto setup + Tauri/Vite with WebView2 CDP on :9222

Wait until both are reachable:

curl -s http://127.0.0.1:9222/json/version | head -3   # CDP up
curl -s http://127.0.0.1:1420 -o /dev/null -w "%{http_code}\n"   # Vite up (200)

If they are not up yet, wait and retry — do not proceed until CDP responds.

---

Step 2 — Run the presenter

node scripts/pitch/present.mjs southwest2026

The driver will:

1. serve the deck + brain page over one local origin,
2. attach to the app over CDP, put it in pet mode, and hide the pet,
3. open the deck full-screen,
4. print: "deck open. Darren presents slides 1–4; click 'Click to reveal!' on slide 4."

---

Step 3 — Present

1. Darren walks slides 1 → 4 (→ / Space to advance, ← to go back).
2. On slide 4, Darren clicks "Click to reveal!".
3. TerranSoul appears and presents 5 → 9 on its own. Nothing else to do.
4. After the close, the pet and final slide stay up. Press Ctrl-C in the driver terminal to end (it exits pet mode and closes the deck window).

---

Editing the pitch

• Words: edit docs/pitch/southwest2026/pitch-script.json — each beat is { "slide", "text", "emotion", "action?" }. action is highlight:<regionId> (brain region on slide 7), appear, or clear. Keep language factual.
• Slides: edit docs/pitch/southwest2026/index.html. Adding/removing a slide changes the slide numbers referenced in the script.
• Charts (slide 8): the numbers live in docs/pitch/southwest2026/deck.js (buildCompareSlide), sourced from benchmark/COMPARISON.md and benchmark/SELF-IMPROVE-COMPARISON.md.

Notes

• One bench/MCP discipline still applies; this skill does not start a bench.
• The pet overlays the full-screen deck as a transparent always-on-top window; put the deck window on the projector/output display before revealing.