Initial TrueGrowth source import

openspec/changes/add-video-prompt-pdf-context/design.md

@@ -0,0 +1,37 @@
+## Context
+The multi-image tool already supports PDF context by caching the uploaded PDF and passing `pdfCacheUrl`, `pdfMimeType`, and `pdfName` into a `TaskType.CHAT` task. `TaskQueueService.buildChatInlineDataParts()` converts that cached PDF to Gemini `inline_data`, so the new popular-video flow should reuse this path instead of introducing a PDF extraction service.
+
+The popular video tool stores workflow records as `AnalysisRecord` and downstream pages expect `analysis: VideoAnalysisData`. A prompt-based start can therefore create the same analysis shape without a source video.
+
+## Goals
+- Add a first tab/input mode named `提示词生成` before `上传视频`.
+- Let users enter a creative prompt and optionally upload a PDF as Gemini context.
+- Submit prompt generation through the unified task queue.
+- Store the result as a normal popular-video record so Script and Generate pages keep working.
+- Avoid persisting PDF base64 or full provider responses in history.
+
+## Non-Goals
+- No new PDF parser or backend endpoint.
+- No new storage table/entity.
+- No changes to video generation semantics after the script is created.
+
+## Design
+Extend the video analyzer task action from `analyze | rewrite` to include a prompt-generation action, for example `prompt-generate`. The task params carry:
+- `prompt`: short task label for queue display
+- `videoAnalyzerPrompt`: full Gemini instruction
+- `videoAnalyzerAction: 'prompt-generate'`
+- `videoAnalyzerSource: 'prompt'`
+- optional `pdfCacheUrl`, `pdfMimeType`, `pdfName`
+
+The task queue routes this action through the normal text generation path with `buildChatInlineDataParts()`, then finalizes the task with both markdown preview and structured `VideoAnalysisData`. Task sync creates an `AnalysisRecord` with `source: 'prompt'`, a prompt source label, and a lightweight source snapshot containing prompt text plus optional PDF metadata/cache reference.
+
+Prompt generation should instruct Gemini to return the same JSON contract as video analysis: total duration, aspect ratio, style, BGM mood, characters, and ordered shots with first/last frame prompts, narration/dialogue, camera movement, and transitions. This keeps existing script editing, versioning, and video generation surfaces compatible.
+
+## Error Handling
+- Reject non-PDF uploads and PDFs above the existing multi-image limit.
+- If a PDF is attached but no Gemini text model is available, show the same style of warning as the multi-image tool.
+- If a cached PDF cannot be read while the task runs, fail the queue task with a clear message.
+- If Gemini returns invalid JSON, mark the task failed and keep the current form state.
+
+## Performance
+The PDF remains a cached Blob reference until the task executes. Conversion to base64 happens only during Gemini submission, matching current multi-image behavior. No new long-lived large strings should be stored in workflow history.

openspec/changes/add-video-prompt-pdf-context/proposal.md

@@ -0,0 +1,14 @@
+# Change: Add PDF-backed prompt start for popular video
+
+## Why
+The popular video tool currently starts from an uploaded video or YouTube URL. Users also need to start from a creative prompt and optional PDF context, then continue through the same script editing and video generation workflow.
+
+## What Changes
+- Add a prompt-generation input mode before the existing upload video mode in the popular video analysis step.
+- Reuse the existing Gemini `TaskType.CHAT` PDF inline-data path from the multi-image tool, avoiding a separate PDF parser or storage entity.
+- Generate a `VideoAnalysisData`-compatible script plan from user prompt plus optional PDF context so the existing Script and Generate pages can continue unchanged.
+- Persist only lightweight prompt/PDF source metadata and cache references needed for pending or repeatable tasks.
+
+## Impact
+- Affected specs: `video-analyzer`
+- Affected code: `video-analyzer` Analyze page/types/utils/task sync, `task-queue-service` chat action routing, focused tests

openspec/changes/add-video-prompt-pdf-context/specs/video-analyzer/spec.md

@@ -0,0 +1,35 @@
+## ADDED Requirements
+
+### Requirement: Popular Video Prompt Start
+The popular video tool SHALL allow users to start the workflow from a creative prompt before uploading a source video.
+
+#### Scenario: Generate script plan from prompt
+- **WHEN** the user selects the prompt-generation input mode and submits a creative prompt
+- **THEN** the system SHALL create a queued text task
+- **AND** the completed task SHALL create a popular-video record with `VideoAnalysisData`-compatible shots
+- **AND** the user SHALL be able to continue to script editing and video generation without uploading a source video
+
+#### Scenario: Preserve existing source-video starts
+- **WHEN** the user selects upload video or YouTube URL
+- **THEN** the existing video analysis behavior SHALL remain available
+- **AND** the prompt-generation mode SHALL NOT change uploaded-video or YouTube task semantics
+
+### Requirement: Popular Video Prompt Start SHALL Support PDF Context
+The prompt-generation input mode SHALL accept an optional PDF context and send it to Gemini with the prompt.
+
+#### Scenario: Submit prompt with PDF context
+- **GIVEN** the user uploads a valid PDF in prompt-generation mode
+- **WHEN** the user submits prompt generation
+- **THEN** the task SHALL include the cached PDF reference and MIME metadata
+- **AND** Gemini SHALL receive the PDF as inline context together with the prompt-generation instruction
+
+#### Scenario: Reject invalid PDF context
+- **WHEN** the user uploads a non-PDF file or a PDF above the allowed size
+- **THEN** the system SHALL reject the file before task submission
+- **AND** it SHALL keep the current prompt text intact
+
+#### Scenario: Missing cached PDF during execution
+- **GIVEN** a prompt-generation task references a cached PDF
+- **WHEN** the cached PDF cannot be read during execution
+- **THEN** the task SHALL fail with a clear PDF-read error
+- **AND** no partial popular-video record SHALL be created

openspec/changes/add-video-prompt-pdf-context/tasks.md

@@ -0,0 +1,11 @@
+## 1. Implementation
+- [x] 1.1 Extend popular-video source/task types to support prompt-based source records and prompt-generation tasks.
+- [x] 1.2 Add a video prompt-generation prompt builder that returns `VideoAnalysisData` JSON compatible with the existing workflow.
+- [x] 1.3 Add `提示词生成` input mode before `上传视频`, with prompt text, optional PDF upload/removal, model selection, and queue submission.
+- [x] 1.4 Route prompt-generation tasks through the existing Gemini chat + PDF inline-data path and finalize structured results.
+- [x] 1.5 Sync completed prompt-generation tasks into `AnalysisRecord` history without storing PDF base64.
+- [x] 1.6 Add focused tests for prompt builder/parsing, task sync, and PDF task params.
+
+## 2. Verification
+- [x] 2.1 Run targeted unit tests for video analyzer utilities/task sync.
+- [ ] 2.2 Manually verify prompt-only generation, prompt + PDF generation, invalid PDF, oversized PDF, and history reopen behavior.