pitch-videolisted

# Automated pitch video (narration-first pipeline) ## Prerequisites — do not start without these 1. **The product is BUILT and reachable** at a stable URL (deployed URL strongly preferred over localhost — the address bar is part of the evidence). The demo flow must already work end-to-end; this skill records reality, it cannot fake it. 2. **A written pitch script already exists** (problem → live demo → tech → business → close). If it doesn't, write and approve it first; this skill starts from script beats, not from scratch. 3. **A visual design language exists** (fonts, color tokens, one signature element) — title cards must look like the product, not like PowerPoint. 4. Toolchain: `python + pip`, `ffmpeg` + `ffprobe` on PATH, `pip install edge-tts playwright` + `python -m playwright install chromium`. All free; edge-tts needs internet. 5. If the app has cold starts (serverless GPU, sleeping free tier): **warm every layer immediately before recording** and know the expected latencies. 6. Ask the human to choose the narration voice (play 2–3 edge-tts options: en-US-AndrewNeural is a safe default; offer a local-accent voice too). ## The core principle **Narration is the timeline; footage serves it.** Never record first and narrate over. Audio durations are measured to the centisecond and become the contract every video segment is cut to. ## Steps ### 1 — Narration (`narration.py`) - Split the script into 6–10 named beats (`s1_problem`, `s2_intro`,