Repository navigation
v0.2.0 — production plans, scene by scene
Pre-release
Framewright helps AI agents plan and coordinate video production, scene by scene and layer by layer. Books and software promos are the initial test cases, not the product boundary.
New: readable production plans
Run python3 skills/video-production-router/scripts/plan.py BRIEF.json --format markdown to review scene timing, tool/capability evidence, explicit choices, required source kinds, suggested files, blocked requirements and separate video/still/voice QA. Default JSON and selection ranking are unchanged by this formatter. All generated checks remain planned or blocked; no asset is automatically verified.
Stage closeout and maintenance
- Complete the six-book, landing-tour and series-compilation stage, retaining owner feedback, technical QA, supplied publication links and incomplete attempts separately.
- Consolidate README/skill entrypoints and correct stale qualification wording.
- Retain scoped tool-chain evidence and manual OpenScreen/MoviePy recipes.
- Pin compilation music and thumbnail inputs, fix archive commands and reject unsafe summed PCM peaks before conversion.
Evidence and limits
167 offline tests passed, alongside retained evidence/link checks, generated matrix and standalone JSON/Markdown package smoke. No renders, provider calls or new model evaluation were performed for this release. Current source remains not_evaluated_current_source; historical scores are not promoted. Strict brief/budget/download-limit validation is deferred.
The ZIP contains the standalone router skill, not media, models, dependencies or a renderer. Review upstream licences and use your authorised runtime. See SHA256SUMS for the ZIP's checksum.