Session Handoff — YouTube Content Engine: roast + realization it's already scoped

Where it started

Sondra asked me to look at her old Base44 workshop app (GitHub AmplifAI-Your-Business/ai-youtube-content-engine-system-dashboard) and, more importantly, her universal-youtube-content-engine skill — proven by producing 250 video packages for her "Why Dachshunds" faceless channel. She wanted a discussion about turning it into an app where a user inputs info and gets a channel-launch package out. Framing evolved across the session: first "maybe a product," then "no — build it FOR ME, a personal tool I can also drop into my offers, and let me run it more than once."

Decisions locked + what shipped

• Ran /roast on the "self-serve app for a market" version → verdict RESHAPE (Contrarian 3 · Expansionist 9 · Logician 4 · Researcher 7 · Buyer 6). Council converged: a one-shot form at flat one-time $29-49 strips the human steering that makes output good, inverts unit economics ($1-5 LLM/run vs one-time price), and the "script→video" moat is commoditized. Real moat = her proof + warm audience.
• Sondra then rejected the market framing entirely: this is a PERSONAL tool (build-for-her), bonus-in-offers, not a product to sell. That voids most roast objections (they were about cold-market economics).
• Established the vidIQ constraint: vidIQ reaches Claude only via the "vidIQ for Claude" MCP connector; vidIQ has NO public API a standalone app can use. So either the tool runs through Claude (keeps vidIQ) OR a standalone app uses YouTube Data API instead.
• BIG realization at the end: the thing she's describing ("agents in a dashboard doing daily YouTube work on a current channel, push a button, connect my skill") ALREADY EXISTS as a PRD scoped earlier the SAME day (2026-07-06) — the "YouTube Content Engine" (her own version of Sandy Lee's Slee Studio). Her on-camera + faceless channels are just two "profiles." The universal-youtube-content-engine skill is the Script Writer / faceless branch inside it. The PRD already resolved vidIQ (app self-scores outliers off YouTube Data API; vidIQ stays an optional Claude bonus). NOTHING was built this session — this was discussion + discovery only.

Key files for next session

• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\YouTube Content Engine\PRD.md — READ FIRST. Full PRD: multi-channel (face|faceless) local content command center, 6 tools, phased build. This is the real project; the whole "build a content-engine app" discussion collapses into this.
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\YouTube Content Engine\FABLE-BUILD-PROMPT.md — ready-to-build Fable prompt (intended path: paste into a fresh Claude Code session opened in that folder, set the /goal finish line).
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\YouTube Content Engine\clone-run\ — cloneify artifacts: scope.md, research.md, PLAN.md (NOT yet Codex-grilled), STATE.md.
• C:\Users\srver\.claude\skills\universal-youtube-content-engine\SKILL.md — the proven engine (5-6 phases, approval-gated). References 7 knowledge files in its references/ (13 hook types, Hybrid Viral Formula, viewer psychology, viral hook library, title click triggers, thumbnail file). This plugs in as the app's Script Writer + faceless branch.
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\Sandy Lee AI Content Team\ — referenced by the PRD: has brand-voice.md (becomes the sondraverva profile voice) + a working Trend Scout prototype proven live 2026-07-06.
• c:\Users\srver\AIOS-Workspace\aios-starter-kit\PROJECTS\YT - Why Dachshunds Channel\Scripts\ — the 250 proven video packages (e.g. Batch 2 - 80 Video_Packages (DONE)\Video_01_*.md) showing the target output shape.
• Memory files touched: none.

Running state

• Background processes: none
• Dev servers / ports: none (PRD proposes local FastAPI + web UI on localhost:47xx via a Start Content Engine.bat, not yet built)
• Open worktrees / branches: none

Verification — how to confirm things still work

• ls "PROJECTS/YouTube Content Engine/" — should show PRD.md, FABLE-BUILD-PROMPT.md, clone-run/
• vidIQ is reachable only in Claude via ToolSearch mcp__claude_ai_vidIQ_for_Claude__*; a standalone app cannot use it — YouTube Data API (YOUTUBE_API_KEY in master .env) is the app's research source per PRD §6.

Deferred + open questions

• Open (Sondra's call — the live decision): how to move on the YouTube Content Engine — (1) build Phase 1 now (Profiles + Trend Scout, the daily-idea engine) [my recommendation], (2) run grill-me-codex to harden PLAN.md before building, or (3) just review the Fable prompt first.
• Deferred: standalone sellable/hosted version of the engine for her offers — PRD §3/§10 keeps v1 private-for-her; graduating to a product is explicitly out of scope for v1.
• Deferred: launching Why Dachshunds publicly to get real channel numbers (roast flagged the "proven" claim is currently "system produces 250 packages," NOT "channel gets views" — matters IF she ever sells this).

Pick up here

Read PROJECTS\YouTube Content Engine\PRD.md, then ask Sondra which of the three moves she wants; if she says build, start Phase 1 (Channel Profiles + Trend Scout via YouTube Data API) per PRD §8 as a real usable local tool before adding the rest.