project creatortrack scripts voice profile build 20260715T122156

CreatorTrack is getting a new capability: generate scripts in a creator's own voice by mining their past YouTube video transcripts. Build is a numbered task pipeline (Task 11 through 18+) on branch feat/social-data-layer-and-scripts, worktree /home/justinwieb/forge-suite-wt/warm-slot-1 (CT warm dev-slot pool, port 3067). Task 11 = bulk transcript ingest (target 548 videos, rate-limited, see [[youtube_bulk_ingest_ratelimit_backoff]]). Task 12 = schema for app.voice_profiles + app.generated_scripts (RLS-scoped, deployed to ct_warm_slot_1 dev DB via forge_workspace_migrate.py; commit f36c1c0 + fix aaa50bd). Task 13 = per-video structure extraction (hooks/beats/CTA/tone) via haiku, one call per transcript. Task 14 = voice profile synthesis, hard-blocked on the full transcript corpus converging. app.generated_scripts links back to CreatorTrack work items via task_node_id/page_node_id FKs to app.nodes (CT has no dedicated tasks table; all work items are app.nodes rows).

Why: this is an active multi-day build spanning many worker dispatches; a future session picking this up needs the task numbering, branch, and worktree to resume correctly. How to apply: when asked about "the scripts app", "voice profile feature", or CreatorTrack content generation from transcripts, check the result files at comms/results/creatortrack-social-data-layer-and-scripts-app-task-*.md and the task ledger before assuming state; this fact decays once the feature ships to prod.

[auto-memory session 60d132de-4659-485c-8044-427af0ccce7f, confidence 0.65, mode staged]