Video-Editing
Wiki/Status.md
Status — Demo Factory
As of: 2026-05-04 (Session 2 closed) Current phase: Implementation complete — next: real e2e pipeline run
What's done
| Area | Status | Notes |
|---|---|---|
| Phase 1 — Foundation (Tasks 1–6) | ✅ Done | Types (Zod), validate, workspace state machine, preflight, readiness check |
| Phase 2 — Capture (Tasks 7–8) | ✅ Done | brand-detect (CSS/favicon scraper), capture (Playwright replay + record) |
| Phase 3 — LLM skills (Tasks 9–12) | ✅ Done | synopsis, script-writer, capture, brand-extractor — all via skill-creator |
| Phase 4 — Remotion engine (Tasks 13–15) | ✅ Done | BrandContext, ScreenLayer, TextOverlayLayer, VOPlayer, AvatarLayer, DemoComposition |
| Phase 5 — Adapters + scripts (Tasks 16–19) | ✅ Done | elevenlabs, heygen, gdrive adapters; render.ts, upload.ts |
| Phase 6 — Distribution (Tasks 20–22) | ✅ Done | plugin.json, /demo command, CLI binary, e2e smoke test |
| SOP remediation (Tasks 23–32) | ✅ Done | Prettier/ESLint, Zod at I/O boundaries, structured logging (lib/logger.ts), engine test infra |
| Remotion render fixes | ✅ Done | Base64 assets (webpack cache bug fixed), ScreenLayer objectFit, TextOverlay frosted pill |
| Real smoke render | ✅ Confirmed | 6s 9:16 MP4 at /tmp/smoke-render/renders/9-16/demo.mp4 — confirmed by Arijit |
| Test suite | ✅ 88 tests | Passing across 13 test files |
| TypeScript | ✅ Clean | tsc --noEmit passes |
| GitHub repo | ✅ Live | github.com/arijitchowdhury80/Video_Editing (private) |
| Real pipeline (live app) | ⏳ Blocked | Needs capture.yaml + running dev server |
CLI build → render wiring |
⏳ Missing | CLI stops after capture; render step not connected |
| 11Labs voice | ⏳ Next phase | Needs ELEVENLABS_API_KEY |
| Heygen avatar | ⏳ Next phase | Needs account + export bundle |
| Google Drive upload | ⏳ Next phase | Needs GDrive mount at vault path |
| Sub-skill eval loop | ⏳ Pending | synopsis + script-writer quality untested against real content |
What's next
- Wire render into CLI
build— currently stops after capture; connect render for--tracks text - Write
capture.yamlfor a product — pick one of: Crawler Factory, Algolia Central, PRISM, NoCo II - Run real e2e pipeline —
npx tsx cli/demo-factory.ts build --product <slug> --url <url> --tracks text - Sub-skill eval loop — run synopsis + script-writer against real captured content; iterate
- 11Labs voice (next phase)
Known issues
| ID | Issue | Severity | Status |
|---|---|---|---|
| KI-001 | CLI build stops after capture — render step not wired |
P1 | Open |
| KI-002 | Sub-skills not eval'd against real content — quality unknown | P2 | Open |
| ~~KI-003~~ | ~~White screen bug (webpack cache serving stale bundle → HTTP 404 on assets)~~ | ~~P0~~ | ✅ Fixed — switched to base64 data URIs |
| ~~KI-004~~ | ~~Screenshot filled only 42% of frame~~ | ~~P0~~ | ✅ Fixed — objectFit: cover |
| ~~KI-005~~ | ~~Text invisible against white background~~ | ~~P0~~ | ✅ Fixed — frosted-glass pill |
Open questions
| ID | Question | Status |
|---|---|---|
| OQ-1 | Cloud archive target: Google Drive (decided, Phase 1) vs S3/R2 (future) | Phase 1 = GDrive resolved |
| OQ-2 | How does the CLI trigger synopsis + script-writer LLM skills? Manual step or wired? | Open — CLI is deterministic-only today |
| OQ-3 | Which product gets the first real e2e run? | Pending Arijit pick |
Session history
| # | Date | Output |
|---|---|---|
| 1 | 2026-05-03 | Spec v1.1 approved (17 sections), all 22 plan tasks implemented, 75 tests passing |
| 2 | 2026-05-04 | SOP remediation (10 tasks), Remotion render fixes, 88 tests, real MP4 confirmed, GitHub repo created and pushed |