Introduce a new detailed Agent handbook (docs/【协作规范】Agent工作入口与执行准则-2026-06-22.md) and refactor AGENTS.md to be an entry-only navigation page. Update documentation cross-references (docs/README.md, project-memory README, document-map, team-conventions, development-workflow, decision-log and project-overview) to point to the new file and clarify reading/order rules, submission/validation constraints, and SpacetimeDB/back-end wording. Also adjust image canvas generation placement rules in the frontend doc and apply related changes to image-editor hooks/tests (src/components/image-editor/*). Purpose: separate high-level entry rules from complex execution details to reduce startup cost for complex tasks and centralize execution practices in a single guideline document.
Unify and lower default mud-point prices across editor generation features and update related API/schema, docs and tests. Changes set gpt-image-2 to 1K=3 / 2K=5, audio1.0 to 5, and chirp-v5 to 12, and update server default JSON, frontend models/workflows, OpenAPI schema (require aspectRatio/imageSize/priceMudPoints for UI extraction), multiple UX docs (quick-edit viewport, UI extraction panel, pricing guidance) and accompanying tests to match the new defaults and validation rules.
Require an explicit red-box selection before extracting UI-design assets: add rectangular/ellipse/brush selection UI, render red-outline marks into a marked Data URL on the client, and update the extraction prompt to "仅提取被红色框框选的素材并整理成spritesheet". Add frontend overlay/raster models and tests to support the selection flow and adjust numerous image-editor tests and views to wire up the selection lifecycle. On the dev ops side, enhance scripts/dev to detect a local %LOCALAPPDATA%/Genarrative/ffmpeg/bin installation on Windows, prepend it to PATH and inject CHARACTER_ANIMATION_FFMPEG_PATH/FFPROBE_PATH (with tests), and export the helper. Server config now prefers GENARRATIVE_*-prefixed FFmpeg env vars and includes a Rust test to validate reading the prefixed variables. Update docs to describe the new UI extraction behavior and local FFmpeg handling.
Introduce centralized model pricing for editor generations: add EditorGenerationPricing types and admin API client methods (GET/POST), backend default config and runtime override support, and server route/handler changes to expose admin editor-generation-pricing. Add an AdminEditorGenerationPricingPage component, styles, route entry, icon, and tests, plus a test for admin routes. Update docs and decision log to describe the pricing contract and management workflow. This enables managing per-model pricing (perGeneration / perSecond, flat or tiered) from the admin panel and exposes runtime pricing to the main site.
Handle character-action generation results as real video assets: use backend previewVideoPath as layer src, set mediaType="video" and assetKind="character-animation", and store the first frame only as thumbnail/poster. Update docs to document the new behavior and preserve backend frame extraction/green-screen workflows. Update UI/UX: compact parameter labels (e.g. "16:9·1K"), quick-edit treats original image as an implicit last reference (hidden from the reference bar) and allows editable bottom capsules, and ensure exports preserve true media file types. Change upload toolbar icon to Upload and add tests to cover the icon and video-layer behavior; also add serialization/hydration test for video thumbnails and adjust many integration/unit tests to reflect new labels, sizes and generation semantics. Files touched include multiple docs under docs/, edited components and tests under src/components/image-editor/, and minor stylesheet updates.
Unify UI aliases for the Nanobanana image model by mapping display and legacy names to the real model ID (gemini-3.1-flash-image-preview). Added normalizeEditorImageModel and used it across generation, dialog, submission and workflow code paths to ensure consistent model selection and pricing. Server-side editor_generation_config and editor_project were updated to accept the display alias (nanobanana2) and legacy alias (nano-banana) for validation/pricing. Tests and docs were added/updated to cover alias normalization and related UI behavior; minor code formatting tweaks included.
Introduce a new UI design asset extraction flow: add POST /api/editor/ui-designs/assets/extractions handler that always uses gpt-image-2 with the fixed prompt "提取画面中的所有独立并整理成spritesheet"; parse Data URL references, call OpenAI image edits, return a spritesheet and sliced icon assets. Increase request body limit for image Data URLs and wire the route with bearer auth. Add platform-image slicing: slice_generated_icon_spritesheet_all_by_connected_components (with auto-naming "素材 1..N", foreground estimation and heuristics) and export it. Add Rust tests for the handler, prompt constant, and slicing behavior. Update frontend client and integration tests to call the new extractEditorUiDesignAssets API and verify spritesheet + icon layers are added to the canvas and persisted. Update multiple docs to describe the extraction UI, toolbar placement, and related audio/playback pitfalls. Minor related updates: toolbar button ordering and audio duration / signed read-url notes.
Adds Seedance 2.0 reference-media constraints and forbids video data URLs; updates docs/decision-log and pitfalls to record the new rules. Introduces typed audio params (type: one-shot|loop, tempo: number|null), generator-specific placeholder styles/avoidance and UI changes for generation panels. Implements related front-end changes (new crop/ raster edit components, many image-editor models, tests and workflow hooks) and server-side contract/handler updates across api-server, shared-contracts, platform-audio, platform-image and vector_engine_audio_generation to support the new submission/asset flow and validations.
JenkenB
kdletters