Video Forge v0.6.0

Video Forge v0.6.0 rebuilds the skill around creative direction, asset casting and cinematography before selecting effects. It adds a genuine 3D scene foundation and practical editorial tools while preserving the original 100-entry coverage register.

Creative library

• Original taste/critique, reference-breakdown, asset-casting, generated-shot continuity, editing-grammar, Premiere Pro and platform playbooks.
• Twelve individually loadable shot recipes with visual argument, assets, camera, rhythm, transitions, sound, do/avoid guidance and pinned source IDs.
• Fresh mechanism-level inspection of 11 repositories / 45 pinned files, plus 14 primary documentation/licensing pages. Full upstream applications and restricted media are not copied into the plugin.
• Instagram, YouTube Shorts and X direction separates storytelling/composition from dated account/upload facts. More effects are not an engagement score.

Native tools

• Real Three.js/R3F worlds through ThreeCanvas: camera position/target/FOV/roll, time-weighted path interpolation, independently keyed object transforms, lit geometry, local image planes and static self-contained GLB assets.
• Original rounded GLB prop, vectors, procedural material/stage kit and camera-study music/SFX.
• Camera key/midpoint previews and sampled target-clearance review; source scouting/contact sheets and one-recipe CLI loading.
• The original 12-second 1080×1920 camera study shows traversal, assembly and detail-to-hero travel. It validates the foundation; it is not a photoreal generation or audience benchmark.

Verification and scope

39 tests, runtime/renderer type checks, all nine skill packages and all six host archives validate. Native 3D/GLB/image staging renders and decodes; a GLB/camera/image frame repeated after an out-of-order seek is byte-identical on the tested machine. Linux/macOS CI, smoke renders and verified release hashes gate publication.

Premiere Pro support is a concrete workflow/handoff guide. No live Adobe application control, .prproj serialization, skeletal animation, compressed glTF decoder, physical motion blur or depth-of-field renderer is bundled. Paid generated-video quality and independent Claude/Codex creative performance remain separately unverified. No model weights were trained.

Use the Codex/Claude Code archives, Claude custom-skill ZIP, standalone skill ZIP, ChatGPT planning companion or complete source ZIP. The new reference files and indexes are generated into each applicable host package; ChatGPT's non-runtime companion links to binary source assets instead of claiming they are bundled.