[superche]

Summary

• support native partial video regeneration across the governed Seedance model catalog, preserving the source model when supported and defaulting to Seedance 2.0
• enforce provider-aligned edit windows: 4–15 seconds for Seedance 2.0-family models and 4–30.2 seconds for Seedance 2.5
• preserve both pippit_asset_id and asset_id from upload through native partial_edit_videos submission
• keep raw asset identities internal while public Facade, MCP, and ChatGPT callers continue to reference the source through source_job_id
• align Facade/OpenAPI, MCP, ChatGPT App, OpenCode, SDK, Widget, docs, generated contracts, and plugin version 0.3.0

Root cause

The previous regeneration path collapsed upload identity into the Pippit asset id and forced Seedance 2.5. Native partial editing requires the upstream asset_id, while the UI and public contracts also needed to preserve supported source models and validate model-specific edit durations.

The new flow fails early with an explicit source asset identity error when asset_id is unavailable instead of submitting an invalid upstream task.

User impact

• regeneration works without a selected frame region; the time range is authoritative and annotations remain optional soft guidance
• callers can use all governed public video models
• unsupported ranges are rejected before a charged upstream submission
• public MCP and ChatGPT contracts stay consistent with the Facade contract

Validation

• npm run check — 53 test files, 390 tests passed; lint, typecheck, build, and architecture checks passed
• npm run check:plugin-contract
• npm run check:plugin-version
• npm run check:public-lockfile
• npm run check:dev-gateway — 17 tools, 2 resources, 2 templates
• Codex Dev cold-contract acceptance completed; native partial regeneration verified successfully by the user